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;
21 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
27 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
28 InstrItinClass arg_ri> {
29 InstrItinClass rr = arg_rr;
30 InstrItinClass rm = arg_rm;
31 InstrItinClass ri = arg_ri;
36 def SSE_ALU_F32S : OpndItins<
37 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
40 def SSE_ALU_F64S : OpndItins<
41 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
44 def SSE_ALU_ITINS_S : SizeItins<
45 SSE_ALU_F32S, SSE_ALU_F64S
48 def SSE_MUL_F32S : OpndItins<
49 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
52 def SSE_MUL_F64S : OpndItins<
53 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
56 def SSE_MUL_ITINS_S : SizeItins<
57 SSE_MUL_F32S, SSE_MUL_F64S
60 def SSE_DIV_F32S : OpndItins<
61 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
64 def SSE_DIV_F64S : OpndItins<
65 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
68 def SSE_DIV_ITINS_S : SizeItins<
69 SSE_DIV_F32S, SSE_DIV_F64S
73 def SSE_ALU_F32P : OpndItins<
74 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
77 def SSE_ALU_F64P : OpndItins<
78 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
81 def SSE_ALU_ITINS_P : SizeItins<
82 SSE_ALU_F32P, SSE_ALU_F64P
85 def SSE_MUL_F32P : OpndItins<
86 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
89 def SSE_MUL_F64P : OpndItins<
90 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
93 def SSE_MUL_ITINS_P : SizeItins<
94 SSE_MUL_F32P, SSE_MUL_F64P
97 def SSE_DIV_F32P : OpndItins<
98 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
101 def SSE_DIV_F64P : OpndItins<
102 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
105 def SSE_DIV_ITINS_P : SizeItins<
106 SSE_DIV_F32P, SSE_DIV_F64P
109 def SSE_BIT_ITINS_P : OpndItins<
110 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
113 def SSE_INTALU_ITINS_P : OpndItins<
114 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
117 def SSE_INTALUQ_ITINS_P : OpndItins<
118 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
121 def SSE_INTMUL_ITINS_P : OpndItins<
122 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
125 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
126 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
129 def SSE_MOVA_ITINS : OpndItins<
130 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
133 def SSE_MOVU_ITINS : OpndItins<
134 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
137 //===----------------------------------------------------------------------===//
138 // SSE 1 & 2 Instructions Classes
139 //===----------------------------------------------------------------------===//
141 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
142 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
143 RegisterClass RC, X86MemOperand x86memop,
146 let isCommutable = 1 in {
147 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
149 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
150 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
151 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
153 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
157 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
160 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
161 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
162 string asm, string SSEVer, string FPSizeStr,
163 Operand memopr, ComplexPattern mem_cpat,
166 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
168 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
169 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
170 [(set RC:$dst, (!cast<Intrinsic>(
171 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
172 RC:$src1, RC:$src2))], itins.rr>;
173 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
175 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
176 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
177 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
178 SSEVer, "_", OpcodeStr, FPSizeStr))
179 RC:$src1, mem_cpat:$src2))], itins.rm>;
182 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
183 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
184 RegisterClass RC, ValueType vt,
185 X86MemOperand x86memop, PatFrag mem_frag,
186 Domain d, OpndItins itins, bit Is2Addr = 1> {
187 let isCommutable = 1 in
188 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
190 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
191 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
192 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
194 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
196 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
197 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
198 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
202 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
203 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
204 string OpcodeStr, X86MemOperand x86memop,
205 list<dag> pat_rr, list<dag> pat_rm,
207 bit rr_hasSideEffects = 0> {
208 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects 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, $src2}")),
213 pat_rr, IIC_DEFAULT, d>;
214 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
218 pat_rm, IIC_DEFAULT, d>;
221 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
222 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
223 string asm, string SSEVer, string FPSizeStr,
224 X86MemOperand x86memop, PatFrag mem_frag,
225 Domain d, OpndItins itins, bit Is2Addr = 1> {
226 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
228 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
229 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
230 [(set RC:$dst, (!cast<Intrinsic>(
231 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
232 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
233 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
235 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
236 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
237 [(set RC:$dst, (!cast<Intrinsic>(
238 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
239 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
242 //===----------------------------------------------------------------------===//
243 // Non-instruction patterns
244 //===----------------------------------------------------------------------===//
246 // A vector extract of the first f32/f64 position is a subregister copy
247 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
248 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
249 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
250 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
252 // A 128-bit subvector extract from the first 256-bit vector position
253 // is a subregister copy that needs no instruction.
254 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
255 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
256 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
257 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
259 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
260 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
261 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
262 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
264 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
265 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
266 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
267 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
269 // A 128-bit subvector insert to the first 256-bit vector position
270 // is a subregister copy that needs no instruction.
271 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
272 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
273 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
274 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
275 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
276 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
277 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
278 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
279 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
280 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
281 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
282 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
284 // Implicitly promote a 32-bit scalar to a vector.
285 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
286 (COPY_TO_REGCLASS FR32:$src, VR128)>;
287 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
288 (COPY_TO_REGCLASS FR32:$src, VR128)>;
289 // Implicitly promote a 64-bit scalar to a vector.
290 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
291 (COPY_TO_REGCLASS FR64:$src, VR128)>;
292 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
293 (COPY_TO_REGCLASS FR64:$src, VR128)>;
295 // Bitcasts between 128-bit vector types. Return the original type since
296 // no instruction is needed for the conversion
297 let Predicates = [HasSSE2] in {
298 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
299 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
300 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
301 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
302 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
303 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
304 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
305 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
306 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
307 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
308 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
309 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
310 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
311 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
312 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
313 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
314 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
315 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
316 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
317 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
318 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
319 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
320 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
321 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
322 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
323 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
324 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
325 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
326 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
327 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
330 // Bitcasts between 256-bit vector types. Return the original type since
331 // no instruction is needed for the conversion
332 let Predicates = [HasAVX] in {
333 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
334 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
335 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
336 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
337 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
338 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
339 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
340 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
341 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
342 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
343 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
344 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
345 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
346 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
347 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
348 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
349 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
350 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
351 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
352 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
353 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
354 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
355 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
356 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
357 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
358 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
359 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
360 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
361 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
362 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
365 // Alias instructions that map fld0 to pxor for sse.
366 // This is expanded by ExpandPostRAPseudos.
367 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
369 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
370 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
371 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
372 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
375 //===----------------------------------------------------------------------===//
376 // AVX & SSE - Zero/One Vectors
377 //===----------------------------------------------------------------------===//
379 // Alias instruction that maps zero vector to pxor / xorp* for sse.
380 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
381 // swizzled by ExecutionDepsFix to pxor.
382 // We set canFoldAsLoad because this can be converted to a constant-pool
383 // load of an all-zeros value if folding it would be beneficial.
384 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
385 isPseudo = 1, neverHasSideEffects = 1 in {
386 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
389 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
390 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
391 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
392 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
393 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
394 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
397 // The same as done above but for AVX. The 256-bit ISA does not support PI,
398 // and doesn't need it because on sandy bridge the register is set to zero
399 // at the rename stage without using any execution unit, so SET0PSY
400 // and SET0PDY can be used for vector int instructions without penalty
401 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
402 // JIT implementatioan, it does not expand the instructions below like
403 // X86MCInstLower does.
404 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
405 isCodeGenOnly = 1 in {
406 let Predicates = [HasAVX] in {
407 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
408 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
409 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
410 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
412 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
413 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
417 let Predicates = [HasAVX2], AddedComplexity = 5 in {
418 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
419 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
420 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
421 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
424 // AVX has no support for 256-bit integer instructions, but since the 128-bit
425 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
426 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
427 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
428 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
430 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
431 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
432 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
434 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
435 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
436 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
438 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
439 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
440 (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 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
445 // JIT implementation, it does not expand the instructions below like
446 // X86MCInstLower does.
447 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
448 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
449 let Predicates = [HasAVX] in
450 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
451 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
452 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
453 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
454 let Predicates = [HasAVX2] in
455 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
456 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
460 //===----------------------------------------------------------------------===//
461 // SSE 1 & 2 - Move FP Scalar Instructions
463 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
464 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
465 // is used instead. Register-to-register movss/movsd is not modeled as an
466 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
467 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
468 //===----------------------------------------------------------------------===//
470 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
471 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
472 [(set VR128:$dst, (vt (OpNode VR128:$src1,
473 (scalar_to_vector RC:$src2))))],
476 // Loading from memory automatically zeroing upper bits.
477 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
478 PatFrag mem_pat, string OpcodeStr> :
479 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
481 [(set RC:$dst, (mem_pat addr:$src))],
485 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
486 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
488 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
489 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
492 // For the disassembler
493 let isCodeGenOnly = 1 in {
494 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
495 (ins VR128:$src1, FR32:$src2),
496 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
499 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
500 (ins VR128:$src1, FR64:$src2),
501 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
506 let canFoldAsLoad = 1, isReMaterializable = 1 in {
507 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
509 let AddedComplexity = 20 in
510 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
514 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
515 "movss\t{$src, $dst|$dst, $src}",
516 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
518 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
519 "movsd\t{$src, $dst|$dst, $src}",
520 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
524 let Constraints = "$src1 = $dst" in {
525 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
526 "movss\t{$src2, $dst|$dst, $src2}">, XS;
527 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
528 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
530 // For the disassembler
531 let isCodeGenOnly = 1 in {
532 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
533 (ins VR128:$src1, FR32:$src2),
534 "movss\t{$src2, $dst|$dst, $src2}", [],
535 IIC_SSE_MOV_S_RR>, XS;
536 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
537 (ins VR128:$src1, FR64:$src2),
538 "movsd\t{$src2, $dst|$dst, $src2}", [],
539 IIC_SSE_MOV_S_RR>, XD;
543 let canFoldAsLoad = 1, isReMaterializable = 1 in {
544 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
546 let AddedComplexity = 20 in
547 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
550 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
551 "movss\t{$src, $dst|$dst, $src}",
552 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
553 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
554 "movsd\t{$src, $dst|$dst, $src}",
555 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
558 let Predicates = [HasAVX] in {
559 let AddedComplexity = 15 in {
560 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
561 // MOVS{S,D} to the lower bits.
562 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
563 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
564 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
565 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
566 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
567 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
568 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
569 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
571 // Move low f32 and clear high bits.
572 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
573 (SUBREG_TO_REG (i32 0),
574 (VMOVSSrr (v4f32 (V_SET0)),
575 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
576 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
577 (SUBREG_TO_REG (i32 0),
578 (VMOVSSrr (v4i32 (V_SET0)),
579 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
582 let AddedComplexity = 20 in {
583 // MOVSSrm zeros the high parts of the register; represent this
584 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
585 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
586 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
587 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
588 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
589 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
590 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
592 // MOVSDrm zeros the high parts of the register; represent this
593 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
594 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
595 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
596 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
597 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
598 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
599 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
600 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
601 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
602 def : Pat<(v2f64 (X86vzload addr:$src)),
603 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
605 // Represent the same patterns above but in the form they appear for
607 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
608 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
609 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
610 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
611 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
612 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
613 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
614 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
615 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
617 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
618 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
619 (SUBREG_TO_REG (i32 0),
620 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
622 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
623 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
624 (SUBREG_TO_REG (i64 0),
625 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
627 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
628 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
629 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
631 // Move low f64 and clear high bits.
632 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
633 (SUBREG_TO_REG (i32 0),
634 (VMOVSDrr (v2f64 (V_SET0)),
635 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
637 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
638 (SUBREG_TO_REG (i32 0),
639 (VMOVSDrr (v2i64 (V_SET0)),
640 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
642 // Extract and store.
643 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
645 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
646 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
648 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
650 // Shuffle with VMOVSS
651 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
652 (VMOVSSrr (v4i32 VR128:$src1),
653 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
654 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
655 (VMOVSSrr (v4f32 VR128:$src1),
656 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
659 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
660 (SUBREG_TO_REG (i32 0),
661 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
662 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
664 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
665 (SUBREG_TO_REG (i32 0),
666 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
667 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
670 // Shuffle with VMOVSD
671 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
672 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
673 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
674 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
675 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
676 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
677 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
678 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
681 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
682 (SUBREG_TO_REG (i32 0),
683 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
684 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
686 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
687 (SUBREG_TO_REG (i32 0),
688 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
689 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
693 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
694 // is during lowering, where it's not possible to recognize the fold cause
695 // it has two uses through a bitcast. One use disappears at isel time and the
696 // fold opportunity reappears.
697 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
698 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
699 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
700 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
701 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
702 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
703 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
704 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
707 let Predicates = [HasSSE1] in {
708 let AddedComplexity = 15 in {
709 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
710 // MOVSS to the lower bits.
711 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
712 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
713 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
714 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
715 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
716 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
719 let AddedComplexity = 20 in {
720 // MOVSSrm already zeros the high parts of the register.
721 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
722 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
723 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
724 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
725 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
726 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
729 // Extract and store.
730 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
732 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
734 // Shuffle with MOVSS
735 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
736 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
737 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
738 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
741 let Predicates = [HasSSE2] in {
742 let AddedComplexity = 15 in {
743 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
744 // MOVSD to the lower bits.
745 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
746 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
749 let AddedComplexity = 20 in {
750 // MOVSDrm already zeros the high parts of the register.
751 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
752 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
753 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
754 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
755 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
756 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
757 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
758 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
759 def : Pat<(v2f64 (X86vzload addr:$src)),
760 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
763 // Extract and store.
764 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
766 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
768 // Shuffle with MOVSD
769 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
770 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
771 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
772 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
773 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
774 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
775 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
776 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
778 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
779 // is during lowering, where it's not possible to recognize the fold cause
780 // it has two uses through a bitcast. One use disappears at isel time and the
781 // fold opportunity reappears.
782 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
783 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
784 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
785 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
786 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
787 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
788 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
789 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
792 //===----------------------------------------------------------------------===//
793 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
794 //===----------------------------------------------------------------------===//
796 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
797 X86MemOperand x86memop, PatFrag ld_frag,
798 string asm, Domain d,
800 bit IsReMaterializable = 1> {
801 let neverHasSideEffects = 1 in
802 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
803 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
804 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
805 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
806 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
807 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
810 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
811 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
813 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
814 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
816 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
817 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
819 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
820 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
823 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
824 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
826 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
827 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
829 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
830 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
832 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
833 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
835 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
836 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
838 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
839 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
841 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
842 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
844 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
845 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
848 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
849 "movaps\t{$src, $dst|$dst, $src}",
850 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
851 IIC_SSE_MOVA_P_MR>, VEX;
852 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
853 "movapd\t{$src, $dst|$dst, $src}",
854 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
855 IIC_SSE_MOVA_P_MR>, VEX;
856 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
857 "movups\t{$src, $dst|$dst, $src}",
858 [(store (v4f32 VR128:$src), addr:$dst)],
859 IIC_SSE_MOVU_P_MR>, VEX;
860 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
861 "movupd\t{$src, $dst|$dst, $src}",
862 [(store (v2f64 VR128:$src), addr:$dst)],
863 IIC_SSE_MOVU_P_MR>, VEX;
864 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
865 "movaps\t{$src, $dst|$dst, $src}",
866 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
867 IIC_SSE_MOVA_P_MR>, VEX;
868 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
869 "movapd\t{$src, $dst|$dst, $src}",
870 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
871 IIC_SSE_MOVA_P_MR>, VEX;
872 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
873 "movups\t{$src, $dst|$dst, $src}",
874 [(store (v8f32 VR256:$src), addr:$dst)],
875 IIC_SSE_MOVU_P_MR>, VEX;
876 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
877 "movupd\t{$src, $dst|$dst, $src}",
878 [(store (v4f64 VR256:$src), addr:$dst)],
879 IIC_SSE_MOVU_P_MR>, VEX;
882 let isCodeGenOnly = 1 in {
883 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
885 "movaps\t{$src, $dst|$dst, $src}", [],
886 IIC_SSE_MOVA_P_RR>, VEX;
887 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
889 "movapd\t{$src, $dst|$dst, $src}", [],
890 IIC_SSE_MOVA_P_RR>, VEX;
891 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
893 "movups\t{$src, $dst|$dst, $src}", [],
894 IIC_SSE_MOVU_P_RR>, VEX;
895 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
897 "movupd\t{$src, $dst|$dst, $src}", [],
898 IIC_SSE_MOVU_P_RR>, VEX;
899 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
901 "movaps\t{$src, $dst|$dst, $src}", [],
902 IIC_SSE_MOVA_P_RR>, VEX;
903 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
905 "movapd\t{$src, $dst|$dst, $src}", [],
906 IIC_SSE_MOVA_P_RR>, VEX;
907 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
909 "movups\t{$src, $dst|$dst, $src}", [],
910 IIC_SSE_MOVU_P_RR>, VEX;
911 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
913 "movupd\t{$src, $dst|$dst, $src}", [],
914 IIC_SSE_MOVU_P_RR>, VEX;
917 let Predicates = [HasAVX] in {
918 def : Pat<(v8i32 (X86vzmovl
919 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
920 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
921 def : Pat<(v4i64 (X86vzmovl
922 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
923 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
924 def : Pat<(v8f32 (X86vzmovl
925 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
926 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
927 def : Pat<(v4f64 (X86vzmovl
928 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
929 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
933 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
934 (VMOVUPSYmr addr:$dst, VR256:$src)>;
935 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
936 (VMOVUPDYmr addr:$dst, VR256:$src)>;
938 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
939 "movaps\t{$src, $dst|$dst, $src}",
940 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
942 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
943 "movapd\t{$src, $dst|$dst, $src}",
944 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
946 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
947 "movups\t{$src, $dst|$dst, $src}",
948 [(store (v4f32 VR128:$src), addr:$dst)],
950 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
951 "movupd\t{$src, $dst|$dst, $src}",
952 [(store (v2f64 VR128:$src), addr:$dst)],
956 let isCodeGenOnly = 1 in {
957 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
958 "movaps\t{$src, $dst|$dst, $src}", [],
960 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
961 "movapd\t{$src, $dst|$dst, $src}", [],
963 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
964 "movups\t{$src, $dst|$dst, $src}", [],
966 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
967 "movupd\t{$src, $dst|$dst, $src}", [],
971 let Predicates = [HasAVX] in {
972 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
973 (VMOVUPSmr addr:$dst, VR128:$src)>;
974 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
975 (VMOVUPDmr addr:$dst, VR128:$src)>;
978 let Predicates = [HasSSE1] in
979 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
980 (MOVUPSmr addr:$dst, VR128:$src)>;
981 let Predicates = [HasSSE2] in
982 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
983 (MOVUPDmr addr:$dst, VR128:$src)>;
985 // Use vmovaps/vmovups for AVX integer load/store.
986 let Predicates = [HasAVX] in {
987 // 128-bit load/store
988 def : Pat<(alignedloadv2i64 addr:$src),
989 (VMOVAPSrm addr:$src)>;
990 def : Pat<(loadv2i64 addr:$src),
991 (VMOVUPSrm addr:$src)>;
993 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
994 (VMOVAPSmr addr:$dst, VR128:$src)>;
995 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
996 (VMOVAPSmr addr:$dst, VR128:$src)>;
997 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
998 (VMOVAPSmr addr:$dst, VR128:$src)>;
999 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1000 (VMOVAPSmr addr:$dst, VR128:$src)>;
1001 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1002 (VMOVUPSmr addr:$dst, VR128:$src)>;
1003 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1004 (VMOVUPSmr addr:$dst, VR128:$src)>;
1005 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1006 (VMOVUPSmr addr:$dst, VR128:$src)>;
1007 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1008 (VMOVUPSmr addr:$dst, VR128:$src)>;
1010 // 256-bit load/store
1011 def : Pat<(alignedloadv4i64 addr:$src),
1012 (VMOVAPSYrm addr:$src)>;
1013 def : Pat<(loadv4i64 addr:$src),
1014 (VMOVUPSYrm addr:$src)>;
1015 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1016 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1017 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1018 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1019 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1020 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1021 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1022 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1023 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1024 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1025 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1026 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1027 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1028 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1029 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1030 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1033 // Use movaps / movups for SSE integer load / store (one byte shorter).
1034 // The instructions selected below are then converted to MOVDQA/MOVDQU
1035 // during the SSE domain pass.
1036 let Predicates = [HasSSE1] in {
1037 def : Pat<(alignedloadv2i64 addr:$src),
1038 (MOVAPSrm addr:$src)>;
1039 def : Pat<(loadv2i64 addr:$src),
1040 (MOVUPSrm addr:$src)>;
1042 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1043 (MOVAPSmr addr:$dst, VR128:$src)>;
1044 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1045 (MOVAPSmr addr:$dst, VR128:$src)>;
1046 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1047 (MOVAPSmr addr:$dst, VR128:$src)>;
1048 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1049 (MOVAPSmr addr:$dst, VR128:$src)>;
1050 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1051 (MOVUPSmr addr:$dst, VR128:$src)>;
1052 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1053 (MOVUPSmr addr:$dst, VR128:$src)>;
1054 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1055 (MOVUPSmr addr:$dst, VR128:$src)>;
1056 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1057 (MOVUPSmr addr:$dst, VR128:$src)>;
1060 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1061 // bits are disregarded. FIXME: Set encoding to pseudo!
1062 let neverHasSideEffects = 1 in {
1063 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1064 "movaps\t{$src, $dst|$dst, $src}", [],
1065 IIC_SSE_MOVA_P_RR>, VEX;
1066 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1067 "movapd\t{$src, $dst|$dst, $src}", [],
1068 IIC_SSE_MOVA_P_RR>, VEX;
1069 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1070 "movaps\t{$src, $dst|$dst, $src}", [],
1072 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1073 "movapd\t{$src, $dst|$dst, $src}", [],
1077 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1078 // bits are disregarded. FIXME: Set encoding to pseudo!
1079 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1080 let isCodeGenOnly = 1 in {
1081 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1082 "movaps\t{$src, $dst|$dst, $src}",
1083 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1084 IIC_SSE_MOVA_P_RM>, VEX;
1085 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1086 "movapd\t{$src, $dst|$dst, $src}",
1087 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1088 IIC_SSE_MOVA_P_RM>, VEX;
1090 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1091 "movaps\t{$src, $dst|$dst, $src}",
1092 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1094 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1095 "movapd\t{$src, $dst|$dst, $src}",
1096 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1100 //===----------------------------------------------------------------------===//
1101 // SSE 1 & 2 - Move Low packed FP Instructions
1102 //===----------------------------------------------------------------------===//
1104 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1105 SDNode psnode, SDNode pdnode, string base_opc,
1106 string asm_opr, InstrItinClass itin> {
1107 def PSrm : PI<opc, MRMSrcMem,
1108 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1109 !strconcat(base_opc, "s", asm_opr),
1112 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1113 itin, SSEPackedSingle>, TB;
1115 def PDrm : PI<opc, MRMSrcMem,
1116 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1117 !strconcat(base_opc, "d", asm_opr),
1118 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1119 (scalar_to_vector (loadf64 addr:$src2)))))],
1120 itin, SSEPackedDouble>, TB, OpSize;
1123 let AddedComplexity = 20 in {
1124 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1125 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1126 IIC_SSE_MOV_LH>, VEX_4V;
1128 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1129 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1130 "\t{$src2, $dst|$dst, $src2}",
1134 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1135 "movlps\t{$src, $dst|$dst, $src}",
1136 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1137 (iPTR 0))), addr:$dst)],
1138 IIC_SSE_MOV_LH>, VEX;
1139 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1140 "movlpd\t{$src, $dst|$dst, $src}",
1141 [(store (f64 (vector_extract (v2f64 VR128:$src),
1142 (iPTR 0))), addr:$dst)],
1143 IIC_SSE_MOV_LH>, VEX;
1144 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1145 "movlps\t{$src, $dst|$dst, $src}",
1146 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1147 (iPTR 0))), addr:$dst)],
1149 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1150 "movlpd\t{$src, $dst|$dst, $src}",
1151 [(store (f64 (vector_extract (v2f64 VR128:$src),
1152 (iPTR 0))), addr:$dst)],
1155 let Predicates = [HasAVX] in {
1156 // Shuffle with VMOVLPS
1157 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1158 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1159 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1160 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1162 // Shuffle with VMOVLPD
1163 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1164 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1165 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1166 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1169 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1171 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1172 def : Pat<(store (v4i32 (X86Movlps
1173 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1174 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1175 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1177 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1178 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1180 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1183 let Predicates = [HasSSE1] in {
1184 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1185 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1186 (iPTR 0))), addr:$src1),
1187 (MOVLPSmr addr:$src1, VR128:$src2)>;
1189 // Shuffle with MOVLPS
1190 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1191 (MOVLPSrm VR128:$src1, addr:$src2)>;
1192 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1193 (MOVLPSrm VR128:$src1, addr:$src2)>;
1194 def : Pat<(X86Movlps VR128:$src1,
1195 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1196 (MOVLPSrm VR128:$src1, addr:$src2)>;
1199 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1201 (MOVLPSmr addr:$src1, VR128:$src2)>;
1202 def : Pat<(store (v4i32 (X86Movlps
1203 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1205 (MOVLPSmr addr:$src1, VR128:$src2)>;
1208 let Predicates = [HasSSE2] in {
1209 // Shuffle with MOVLPD
1210 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1211 (MOVLPDrm VR128:$src1, addr:$src2)>;
1212 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1213 (MOVLPDrm VR128:$src1, addr:$src2)>;
1216 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1218 (MOVLPDmr addr:$src1, VR128:$src2)>;
1219 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1221 (MOVLPDmr addr:$src1, VR128:$src2)>;
1224 //===----------------------------------------------------------------------===//
1225 // SSE 1 & 2 - Move Hi packed FP Instructions
1226 //===----------------------------------------------------------------------===//
1228 let AddedComplexity = 20 in {
1229 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1230 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 IIC_SSE_MOV_LH>, VEX_4V;
1233 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1234 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1235 "\t{$src2, $dst|$dst, $src2}",
1239 // v2f64 extract element 1 is always custom lowered to unpack high to low
1240 // and extract element 0 so the non-store version isn't too horrible.
1241 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1242 "movhps\t{$src, $dst|$dst, $src}",
1243 [(store (f64 (vector_extract
1244 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1245 (bc_v2f64 (v4f32 VR128:$src))),
1246 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1247 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1248 "movhpd\t{$src, $dst|$dst, $src}",
1249 [(store (f64 (vector_extract
1250 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1251 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1252 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1253 "movhps\t{$src, $dst|$dst, $src}",
1254 [(store (f64 (vector_extract
1255 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1256 (bc_v2f64 (v4f32 VR128:$src))),
1257 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1258 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1259 "movhpd\t{$src, $dst|$dst, $src}",
1260 [(store (f64 (vector_extract
1261 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1262 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1264 let Predicates = [HasAVX] in {
1266 def : Pat<(X86Movlhps VR128:$src1,
1267 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1268 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1269 def : Pat<(X86Movlhps VR128:$src1,
1270 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1271 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1273 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1274 // is during lowering, where it's not possible to recognize the load fold
1275 // cause it has two uses through a bitcast. One use disappears at isel time
1276 // and the fold opportunity reappears.
1277 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1278 (scalar_to_vector (loadf64 addr:$src2)))),
1279 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1282 let Predicates = [HasSSE1] in {
1284 def : Pat<(X86Movlhps VR128:$src1,
1285 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1286 (MOVHPSrm VR128:$src1, addr:$src2)>;
1287 def : Pat<(X86Movlhps VR128:$src1,
1288 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1289 (MOVHPSrm VR128:$src1, addr:$src2)>;
1292 let Predicates = [HasSSE2] in {
1293 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1294 // is during lowering, where it's not possible to recognize the load fold
1295 // cause it has two uses through a bitcast. One use disappears at isel time
1296 // and the fold opportunity reappears.
1297 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1298 (scalar_to_vector (loadf64 addr:$src2)))),
1299 (MOVHPDrm VR128:$src1, addr:$src2)>;
1302 //===----------------------------------------------------------------------===//
1303 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1304 //===----------------------------------------------------------------------===//
1306 let AddedComplexity = 20 in {
1307 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1308 (ins VR128:$src1, VR128:$src2),
1309 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1311 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1314 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1315 (ins VR128:$src1, VR128:$src2),
1316 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1318 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1322 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1323 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1324 (ins VR128:$src1, VR128:$src2),
1325 "movlhps\t{$src2, $dst|$dst, $src2}",
1327 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1329 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1330 (ins VR128:$src1, VR128:$src2),
1331 "movhlps\t{$src2, $dst|$dst, $src2}",
1333 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1337 let Predicates = [HasAVX] in {
1339 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1340 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1341 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1342 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1345 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1349 let Predicates = [HasSSE1] in {
1351 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1352 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1353 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1354 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1357 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1358 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1361 //===----------------------------------------------------------------------===//
1362 // SSE 1 & 2 - Conversion Instructions
1363 //===----------------------------------------------------------------------===//
1365 def SSE_CVT_PD : OpndItins<
1366 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1369 def SSE_CVT_PS : OpndItins<
1370 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1373 def SSE_CVT_Scalar : OpndItins<
1374 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1377 def SSE_CVT_SS2SI_32 : OpndItins<
1378 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1381 def SSE_CVT_SS2SI_64 : OpndItins<
1382 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1385 def SSE_CVT_SD2SI : OpndItins<
1386 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1389 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1390 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1391 string asm, OpndItins itins> {
1392 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1393 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1395 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1396 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1400 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1401 X86MemOperand x86memop, string asm, Domain d,
1403 let neverHasSideEffects = 1 in {
1404 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1407 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1412 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1413 X86MemOperand x86memop, string asm> {
1414 let neverHasSideEffects = 1 in {
1415 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1416 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1418 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1419 (ins DstRC:$src1, x86memop:$src),
1420 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1421 } // neverHasSideEffects = 1
1424 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1425 "cvttss2si\t{$src, $dst|$dst, $src}",
1428 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1429 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1431 XS, VEX, VEX_W, VEX_LIG;
1432 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1433 "cvttsd2si\t{$src, $dst|$dst, $src}",
1436 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1437 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1439 XD, VEX, VEX_W, VEX_LIG;
1441 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1442 // register, but the same isn't true when only using memory operands,
1443 // provide other assembly "l" and "q" forms to address this explicitly
1444 // where appropriate to do so.
1445 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1446 XS, VEX_4V, VEX_LIG;
1447 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1448 XS, VEX_4V, VEX_W, VEX_LIG;
1449 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1450 XD, VEX_4V, VEX_LIG;
1451 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1452 XD, VEX_4V, VEX_W, VEX_LIG;
1454 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1455 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1456 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1457 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1459 let Predicates = [HasAVX], AddedComplexity = 1 in {
1460 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1461 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1462 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1463 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1464 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1465 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1466 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1467 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1469 def : Pat<(f32 (sint_to_fp GR32:$src)),
1470 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1471 def : Pat<(f32 (sint_to_fp GR64:$src)),
1472 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1473 def : Pat<(f64 (sint_to_fp GR32:$src)),
1474 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1475 def : Pat<(f64 (sint_to_fp GR64:$src)),
1476 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1479 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1480 "cvttss2si\t{$src, $dst|$dst, $src}",
1481 SSE_CVT_SS2SI_32>, XS;
1482 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1483 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1484 SSE_CVT_SS2SI_64>, XS, REX_W;
1485 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1486 "cvttsd2si\t{$src, $dst|$dst, $src}",
1488 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1489 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1490 SSE_CVT_SD2SI>, XD, REX_W;
1491 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1492 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1493 SSE_CVT_Scalar>, XS;
1494 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1495 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1496 SSE_CVT_Scalar>, XS, REX_W;
1497 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1498 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1499 SSE_CVT_Scalar>, XD;
1500 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1501 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1502 SSE_CVT_Scalar>, XD, REX_W;
1504 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1505 // and/or XMM operand(s).
1507 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1508 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1509 string asm, OpndItins itins> {
1510 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1511 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1512 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1513 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1514 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1515 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1518 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1519 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1520 PatFrag ld_frag, string asm, OpndItins itins,
1522 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1524 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1525 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1526 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1528 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1529 (ins DstRC:$src1, x86memop:$src2),
1531 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1532 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1533 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1537 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1538 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
1539 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1540 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1541 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
1542 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1544 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1545 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1546 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1547 sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1550 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1551 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1552 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1553 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1554 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1555 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1557 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1558 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1559 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1560 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1561 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1562 SSE_CVT_Scalar, 0>, XD,
1565 let Constraints = "$src1 = $dst" in {
1566 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1567 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1568 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1569 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1570 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1571 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1572 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1573 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1574 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1575 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1576 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1577 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1582 // Aliases for intrinsics
1583 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1584 ssmem, sse_load_f32, "cvttss2si",
1585 SSE_CVT_SS2SI_32>, XS, VEX;
1586 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1587 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1588 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1590 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1591 sdmem, sse_load_f64, "cvttsd2si",
1592 SSE_CVT_SD2SI>, XD, VEX;
1593 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1594 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1595 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1597 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1598 ssmem, sse_load_f32, "cvttss2si",
1599 SSE_CVT_SS2SI_32>, XS;
1600 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1601 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1602 "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
1603 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1604 sdmem, sse_load_f64, "cvttsd2si",
1606 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1607 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1608 "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1610 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1611 ssmem, sse_load_f32, "cvtss2si{l}",
1612 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1613 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1614 ssmem, sse_load_f32, "cvtss2si{q}",
1615 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1617 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1618 ssmem, sse_load_f32, "cvtss2si{l}",
1619 SSE_CVT_SS2SI_32>, XS;
1620 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1621 ssmem, sse_load_f32, "cvtss2si{q}",
1622 SSE_CVT_SS2SI_64>, XS, REX_W;
1624 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1625 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1626 SSEPackedSingle, SSE_CVT_PS>,
1627 TB, VEX, Requires<[HasAVX]>;
1628 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1629 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1630 SSEPackedSingle, SSE_CVT_PS>,
1631 TB, VEX, Requires<[HasAVX]>;
1633 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1634 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1635 SSEPackedSingle, SSE_CVT_PS>,
1636 TB, Requires<[HasSSE2]>;
1640 // Convert scalar double to scalar single
1641 let neverHasSideEffects = 1 in {
1642 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1643 (ins FR64:$src1, FR64:$src2),
1644 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1645 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1647 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1648 (ins FR64:$src1, f64mem:$src2),
1649 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1650 [], IIC_SSE_CVT_Scalar_RM>,
1651 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1654 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1657 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1658 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1659 [(set FR32:$dst, (fround FR64:$src))],
1660 IIC_SSE_CVT_Scalar_RR>;
1661 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1662 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1663 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1664 IIC_SSE_CVT_Scalar_RM>,
1666 Requires<[HasSSE2, OptForSize]>;
1668 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1669 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1670 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1672 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1673 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
1674 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1675 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1676 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1677 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1678 VR128:$src1, sse_load_f64:$src2))],
1679 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
1681 let Constraints = "$src1 = $dst" in {
1682 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1683 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1684 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1686 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1687 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[HasSSE2]>;
1688 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1689 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1690 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1691 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1692 VR128:$src1, sse_load_f64:$src2))],
1693 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[HasSSE2]>;
1696 // Convert scalar single to scalar double
1697 // SSE2 instructions with XS prefix
1698 let neverHasSideEffects = 1 in {
1699 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1700 (ins FR32:$src1, FR32:$src2),
1701 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1702 [], IIC_SSE_CVT_Scalar_RR>,
1703 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1705 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1706 (ins FR32:$src1, f32mem:$src2),
1707 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1708 [], IIC_SSE_CVT_Scalar_RM>,
1709 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1712 let AddedComplexity = 1 in { // give AVX priority
1713 def : Pat<(f64 (fextend FR32:$src)),
1714 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1715 def : Pat<(fextend (loadf32 addr:$src)),
1716 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1718 def : Pat<(extloadf32 addr:$src),
1719 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1720 Requires<[HasAVX, OptForSize]>;
1721 def : Pat<(extloadf32 addr:$src),
1722 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1723 Requires<[HasAVX, OptForSpeed]>;
1724 } // AddedComplexity = 1
1726 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1727 "cvtss2sd\t{$src, $dst|$dst, $src}",
1728 [(set FR64:$dst, (fextend FR32:$src))],
1729 IIC_SSE_CVT_Scalar_RR>, XS,
1730 Requires<[HasSSE2]>;
1731 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1732 "cvtss2sd\t{$src, $dst|$dst, $src}",
1733 [(set FR64:$dst, (extloadf32 addr:$src))],
1734 IIC_SSE_CVT_Scalar_RM>, XS,
1735 Requires<[HasSSE2, OptForSize]>;
1737 // extload f32 -> f64. This matches load+fextend because we have a hack in
1738 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1740 // Since these loads aren't folded into the fextend, we have to match it
1742 def : Pat<(fextend (loadf32 addr:$src)),
1743 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1744 def : Pat<(extloadf32 addr:$src),
1745 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1747 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1748 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1749 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1751 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1752 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
1753 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1754 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1755 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1757 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1758 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
1759 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1760 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1761 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1762 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1764 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1765 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[HasSSE2]>;
1766 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1767 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1768 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1770 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1771 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[HasSSE2]>;
1774 // Convert packed single/double fp to doubleword
1775 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 "cvtps2dq\t{$src, $dst|$dst, $src}",
1777 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1778 IIC_SSE_CVT_PS_RR>, VEX;
1779 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1780 "cvtps2dq\t{$src, $dst|$dst, $src}",
1782 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1783 IIC_SSE_CVT_PS_RM>, VEX;
1784 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1785 "cvtps2dq\t{$src, $dst|$dst, $src}",
1787 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1788 IIC_SSE_CVT_PS_RR>, VEX;
1789 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1790 "cvtps2dq\t{$src, $dst|$dst, $src}",
1792 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1793 IIC_SSE_CVT_PS_RM>, VEX;
1794 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}",
1796 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1798 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1799 "cvtps2dq\t{$src, $dst|$dst, $src}",
1801 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1805 // Convert Packed Double FP to Packed DW Integers
1806 let Predicates = [HasAVX] in {
1807 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1808 // register, but the same isn't true when using memory operands instead.
1809 // Provide other assembly rr and rm forms to address this explicitly.
1810 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1811 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1812 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1816 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1817 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1818 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1819 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1821 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1824 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1825 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1827 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX;
1828 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1829 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1831 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1833 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1834 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1837 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1838 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1840 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1842 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1843 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1844 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1847 // Convert with truncation packed single/double fp to doubleword
1848 // SSE2 packed instructions with XS prefix
1849 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1850 "cvttps2dq\t{$src, $dst|$dst, $src}",
1852 (int_x86_sse2_cvttps2dq VR128:$src))],
1853 IIC_SSE_CVT_PS_RR>, VEX;
1854 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1855 "cvttps2dq\t{$src, $dst|$dst, $src}",
1856 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1857 (memopv4f32 addr:$src)))],
1858 IIC_SSE_CVT_PS_RM>, VEX;
1859 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1860 "cvttps2dq\t{$src, $dst|$dst, $src}",
1862 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1863 IIC_SSE_CVT_PS_RR>, VEX;
1864 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1865 "cvttps2dq\t{$src, $dst|$dst, $src}",
1866 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1867 (memopv8f32 addr:$src)))],
1868 IIC_SSE_CVT_PS_RM>, VEX;
1870 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1871 "cvttps2dq\t{$src, $dst|$dst, $src}",
1872 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1874 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1875 "cvttps2dq\t{$src, $dst|$dst, $src}",
1877 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1880 let Predicates = [HasAVX] in {
1881 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1882 (VCVTDQ2PSrr VR128:$src)>;
1883 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1884 (VCVTDQ2PSrm addr:$src)>;
1886 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1887 (VCVTDQ2PSrr VR128:$src)>;
1888 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1889 (VCVTDQ2PSrm addr:$src)>;
1891 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1892 (VCVTTPS2DQrr VR128:$src)>;
1893 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1894 (VCVTTPS2DQrm addr:$src)>;
1896 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1897 (VCVTDQ2PSYrr VR256:$src)>;
1898 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1899 (VCVTDQ2PSYrm addr:$src)>;
1901 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1902 (VCVTTPS2DQYrr VR256:$src)>;
1903 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1904 (VCVTTPS2DQYrm addr:$src)>;
1907 let Predicates = [HasSSE2] in {
1908 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1909 (CVTDQ2PSrr VR128:$src)>;
1910 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1911 (CVTDQ2PSrm addr:$src)>;
1913 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1914 (CVTDQ2PSrr VR128:$src)>;
1915 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1916 (CVTDQ2PSrm addr:$src)>;
1918 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1919 (CVTTPS2DQrr VR128:$src)>;
1920 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1921 (CVTTPS2DQrm addr:$src)>;
1924 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1925 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1927 (int_x86_sse2_cvttpd2dq VR128:$src))],
1928 IIC_SSE_CVT_PD_RR>, VEX;
1930 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1931 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1932 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1934 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1935 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1936 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1937 (memopv2f64 addr:$src)))],
1940 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1941 // register, but the same isn't true when using memory operands instead.
1942 // Provide other assembly rr and rm forms to address this explicitly.
1945 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1946 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1947 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1948 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1949 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1950 (memopv2f64 addr:$src)))],
1951 IIC_SSE_CVT_PD_RM>, VEX;
1954 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1955 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1957 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
1958 IIC_SSE_CVT_PD_RR>, VEX;
1959 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1960 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1962 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
1963 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1964 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1965 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1967 let Predicates = [HasAVX] in {
1968 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1969 (VCVTTPD2DQYrr VR256:$src)>;
1970 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1971 (VCVTTPD2DQYrm addr:$src)>;
1972 } // Predicates = [HasAVX]
1974 // Convert packed single to packed double
1975 let Predicates = [HasAVX] in {
1976 // SSE2 instructions without OpSize prefix
1977 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1978 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1979 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1980 IIC_SSE_CVT_PD_RR>, TB, VEX;
1981 let neverHasSideEffects = 1, mayLoad = 1 in
1982 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1983 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1984 IIC_SSE_CVT_PD_RM>, TB, VEX;
1985 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1986 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1988 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
1989 IIC_SSE_CVT_PD_RR>, TB, VEX;
1990 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1991 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1993 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
1994 IIC_SSE_CVT_PD_RM>, TB, VEX;
1997 let Predicates = [HasSSE2] in {
1998 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1999 "cvtps2pd\t{$src, $dst|$dst, $src}",
2000 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2001 IIC_SSE_CVT_PD_RR>, TB;
2002 let neverHasSideEffects = 1, mayLoad = 1 in
2003 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2004 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2005 IIC_SSE_CVT_PD_RM>, TB;
2008 // Convert Packed DW Integers to Packed Double FP
2009 let Predicates = [HasAVX] in {
2010 let neverHasSideEffects = 1, mayLoad = 1 in
2011 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2012 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2014 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2015 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2017 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2018 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2019 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2021 (int_x86_avx_cvtdq2_pd_256
2022 (bitconvert (memopv2i64 addr:$src))))]>, VEX;
2023 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2024 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2026 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX;
2029 let neverHasSideEffects = 1, mayLoad = 1 in
2030 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2031 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2033 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2034 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2035 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2038 // AVX 256-bit register conversion intrinsics
2039 let Predicates = [HasAVX] in {
2040 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2041 (VCVTDQ2PDYrr VR128:$src)>;
2042 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2043 (VCVTDQ2PDYrm addr:$src)>;
2044 } // Predicates = [HasAVX]
2046 // Convert packed double to packed single
2047 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2048 // register, but the same isn't true when using memory operands instead.
2049 // Provide other assembly rr and rm forms to address this explicitly.
2050 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2051 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2052 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2053 IIC_SSE_CVT_PD_RR>, VEX;
2056 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2057 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2058 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2059 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2061 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2062 IIC_SSE_CVT_PD_RM>, VEX;
2065 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2066 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2068 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2069 IIC_SSE_CVT_PD_RR>, VEX;
2070 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2071 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2073 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2074 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2075 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2076 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2078 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2079 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2080 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2082 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2083 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2085 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2089 // AVX 256-bit register conversion intrinsics
2090 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2091 // whenever possible to avoid declaring two versions of each one.
2092 let Predicates = [HasAVX] in {
2093 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2094 (VCVTDQ2PSYrr VR256:$src)>;
2095 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2096 (VCVTDQ2PSYrm addr:$src)>;
2098 // Match fround and fextend for 128/256-bit conversions
2099 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2100 (VCVTPD2PSYrr VR256:$src)>;
2101 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2102 (VCVTPD2PSYrm addr:$src)>;
2104 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2105 (VCVTPS2PDYrr VR128:$src)>;
2106 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2107 (VCVTPS2PDYrm addr:$src)>;
2110 //===----------------------------------------------------------------------===//
2111 // SSE 1 & 2 - Compare Instructions
2112 //===----------------------------------------------------------------------===//
2114 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2115 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2116 Operand CC, SDNode OpNode, ValueType VT,
2117 PatFrag ld_frag, string asm, string asm_alt,
2119 def rr : SIi8<0xC2, MRMSrcReg,
2120 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2121 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2123 def rm : SIi8<0xC2, MRMSrcMem,
2124 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2125 [(set RC:$dst, (OpNode (VT RC:$src1),
2126 (ld_frag addr:$src2), imm:$cc))],
2129 // Accept explicit immediate argument form instead of comparison code.
2130 let neverHasSideEffects = 1 in {
2131 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2132 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2133 IIC_SSE_ALU_F32S_RR>;
2135 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2136 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2137 IIC_SSE_ALU_F32S_RM>;
2141 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2142 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2143 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2145 XS, VEX_4V, VEX_LIG;
2146 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2147 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2148 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2149 SSE_ALU_F32S>, // same latency as 32 bit compare
2150 XD, VEX_4V, VEX_LIG;
2152 let Constraints = "$src1 = $dst" in {
2153 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2154 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2155 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2157 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2158 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2159 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2160 SSE_ALU_F32S>, // same latency as 32 bit compare
2164 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2165 Intrinsic Int, string asm, OpndItins itins> {
2166 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2167 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2168 [(set VR128:$dst, (Int VR128:$src1,
2169 VR128:$src, imm:$cc))],
2171 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2172 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2173 [(set VR128:$dst, (Int VR128:$src1,
2174 (load addr:$src), imm:$cc))],
2178 // Aliases to match intrinsics which expect XMM operand(s).
2179 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2180 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2183 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2184 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2185 SSE_ALU_F32S>, // same latency as f32
2187 let Constraints = "$src1 = $dst" in {
2188 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2189 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2191 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2192 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2193 SSE_ALU_F32S>, // same latency as f32
2198 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2199 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2200 ValueType vt, X86MemOperand x86memop,
2201 PatFrag ld_frag, string OpcodeStr, Domain d> {
2202 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2203 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2204 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2205 IIC_SSE_COMIS_RR, d>;
2206 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2207 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2208 [(set EFLAGS, (OpNode (vt RC:$src1),
2209 (ld_frag addr:$src2)))],
2210 IIC_SSE_COMIS_RM, d>;
2213 let Defs = [EFLAGS] in {
2214 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2215 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2216 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2217 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2219 let Pattern = []<dag> in {
2220 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2221 "comiss", SSEPackedSingle>, TB, VEX,
2223 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2224 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2228 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2229 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2230 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2231 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2233 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2234 load, "comiss", SSEPackedSingle>, TB, VEX;
2235 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2236 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2237 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2238 "ucomiss", SSEPackedSingle>, TB;
2239 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2240 "ucomisd", SSEPackedDouble>, TB, OpSize;
2242 let Pattern = []<dag> in {
2243 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2244 "comiss", SSEPackedSingle>, TB;
2245 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2246 "comisd", SSEPackedDouble>, TB, OpSize;
2249 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2250 load, "ucomiss", SSEPackedSingle>, TB;
2251 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2252 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2254 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2255 "comiss", SSEPackedSingle>, TB;
2256 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2257 "comisd", SSEPackedDouble>, TB, OpSize;
2258 } // Defs = [EFLAGS]
2260 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2261 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2262 Operand CC, Intrinsic Int, string asm,
2263 string asm_alt, Domain d> {
2264 def rri : PIi8<0xC2, MRMSrcReg,
2265 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2266 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2267 IIC_SSE_CMPP_RR, d>;
2268 def rmi : PIi8<0xC2, MRMSrcMem,
2269 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2270 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2271 IIC_SSE_CMPP_RM, d>;
2273 // Accept explicit immediate argument form instead of comparison code.
2274 let neverHasSideEffects = 1 in {
2275 def rri_alt : PIi8<0xC2, MRMSrcReg,
2276 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2277 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2278 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2279 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2280 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2284 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2285 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2286 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2287 SSEPackedSingle>, TB, VEX_4V;
2288 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2289 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2290 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2291 SSEPackedDouble>, TB, OpSize, VEX_4V;
2292 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2293 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2294 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2295 SSEPackedSingle>, TB, VEX_4V;
2296 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2297 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2298 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2299 SSEPackedDouble>, TB, OpSize, VEX_4V;
2300 let Constraints = "$src1 = $dst" in {
2301 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2302 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2303 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2304 SSEPackedSingle>, TB;
2305 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2306 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2307 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2308 SSEPackedDouble>, TB, OpSize;
2311 let Predicates = [HasAVX] in {
2312 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2313 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2314 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2315 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2316 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2317 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2318 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2319 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2321 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2322 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2323 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2324 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2325 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2326 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2327 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2328 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2331 let Predicates = [HasSSE1] in {
2332 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2333 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2334 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2335 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2338 let Predicates = [HasSSE2] in {
2339 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2340 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2341 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2342 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2345 //===----------------------------------------------------------------------===//
2346 // SSE 1 & 2 - Shuffle Instructions
2347 //===----------------------------------------------------------------------===//
2349 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2350 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2351 ValueType vt, string asm, PatFrag mem_frag,
2352 Domain d, bit IsConvertibleToThreeAddress = 0> {
2353 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2354 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2355 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2356 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2357 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2358 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2359 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2360 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2361 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2364 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2365 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2366 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2367 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2368 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2369 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2370 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2371 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2372 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2373 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2374 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2375 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2377 let Constraints = "$src1 = $dst" in {
2378 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2379 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2380 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2382 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2383 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2384 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2388 let Predicates = [HasAVX] in {
2389 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2390 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2391 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2392 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2393 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2395 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2396 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2397 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2398 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2399 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2402 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2403 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2404 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2405 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2406 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2408 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2409 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2410 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2411 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2412 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2415 let Predicates = [HasSSE1] in {
2416 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2417 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2418 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2419 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2420 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2423 let Predicates = [HasSSE2] in {
2424 // Generic SHUFPD patterns
2425 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2426 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2427 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2428 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2429 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2432 //===----------------------------------------------------------------------===//
2433 // SSE 1 & 2 - Unpack Instructions
2434 //===----------------------------------------------------------------------===//
2436 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2437 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2438 PatFrag mem_frag, RegisterClass RC,
2439 X86MemOperand x86memop, string asm,
2441 def rr : PI<opc, MRMSrcReg,
2442 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2444 (vt (OpNode RC:$src1, RC:$src2)))],
2446 def rm : PI<opc, MRMSrcMem,
2447 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2449 (vt (OpNode RC:$src1,
2450 (mem_frag addr:$src2))))],
2454 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2455 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2456 SSEPackedSingle>, TB, VEX_4V;
2457 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2458 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2459 SSEPackedDouble>, TB, OpSize, VEX_4V;
2460 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2461 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2462 SSEPackedSingle>, TB, VEX_4V;
2463 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2464 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2465 SSEPackedDouble>, TB, OpSize, VEX_4V;
2467 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2468 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 SSEPackedSingle>, TB, VEX_4V;
2470 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2471 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 SSEPackedDouble>, TB, OpSize, VEX_4V;
2473 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2474 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2475 SSEPackedSingle>, TB, VEX_4V;
2476 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2477 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2478 SSEPackedDouble>, TB, OpSize, VEX_4V;
2480 let Constraints = "$src1 = $dst" in {
2481 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2482 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2483 SSEPackedSingle>, TB;
2484 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2485 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2486 SSEPackedDouble>, TB, OpSize;
2487 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2488 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2489 SSEPackedSingle>, TB;
2490 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2491 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2492 SSEPackedDouble>, TB, OpSize;
2493 } // Constraints = "$src1 = $dst"
2495 let Predicates = [HasAVX], AddedComplexity = 1 in {
2496 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2497 // problem is during lowering, where it's not possible to recognize the load
2498 // fold cause it has two uses through a bitcast. One use disappears at isel
2499 // time and the fold opportunity reappears.
2500 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2501 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2504 let Predicates = [HasSSE2] in {
2505 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2506 // problem is during lowering, where it's not possible to recognize the load
2507 // fold cause it has two uses through a bitcast. One use disappears at isel
2508 // time and the fold opportunity reappears.
2509 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2510 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2513 //===----------------------------------------------------------------------===//
2514 // SSE 1 & 2 - Extract Floating-Point Sign mask
2515 //===----------------------------------------------------------------------===//
2517 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2518 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2520 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2521 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2522 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2523 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2524 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2525 IIC_SSE_MOVMSK, d>, REX_W;
2528 let Predicates = [HasAVX] in {
2529 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2530 "movmskps", SSEPackedSingle>, TB, VEX;
2531 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2532 "movmskpd", SSEPackedDouble>, TB,
2534 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2535 "movmskps", SSEPackedSingle>, TB, VEX;
2536 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2537 "movmskpd", SSEPackedDouble>, TB,
2540 def : Pat<(i32 (X86fgetsign FR32:$src)),
2541 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2542 def : Pat<(i64 (X86fgetsign FR32:$src)),
2543 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2544 def : Pat<(i32 (X86fgetsign FR64:$src)),
2545 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2546 def : Pat<(i64 (X86fgetsign FR64:$src)),
2547 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2550 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2551 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2552 SSEPackedSingle>, TB, VEX;
2553 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2554 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2555 SSEPackedDouble>, TB,
2557 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2558 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2559 SSEPackedSingle>, TB, VEX;
2560 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2561 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2562 SSEPackedDouble>, TB,
2566 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2567 SSEPackedSingle>, TB;
2568 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2569 SSEPackedDouble>, TB, OpSize;
2571 def : Pat<(i32 (X86fgetsign FR32:$src)),
2572 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2573 Requires<[HasSSE1]>;
2574 def : Pat<(i64 (X86fgetsign FR32:$src)),
2575 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2576 Requires<[HasSSE1]>;
2577 def : Pat<(i32 (X86fgetsign FR64:$src)),
2578 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2579 Requires<[HasSSE2]>;
2580 def : Pat<(i64 (X86fgetsign FR64:$src)),
2581 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2582 Requires<[HasSSE2]>;
2584 //===---------------------------------------------------------------------===//
2585 // SSE2 - Packed Integer Logical Instructions
2586 //===---------------------------------------------------------------------===//
2588 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2590 /// PDI_binop_rm - Simple SSE2 binary operator.
2591 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2592 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2593 X86MemOperand x86memop,
2595 bit IsCommutable = 0,
2597 let isCommutable = IsCommutable in
2598 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2599 (ins RC:$src1, RC:$src2),
2601 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2602 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2603 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2604 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2605 (ins RC:$src1, x86memop:$src2),
2607 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2608 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2609 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2610 (bitconvert (memop_frag addr:$src2)))))],
2613 } // ExeDomain = SSEPackedInt
2615 // These are ordered here for pattern ordering requirements with the fp versions
2617 let Predicates = [HasAVX] in {
2618 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2619 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2620 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2621 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2622 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2623 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2624 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2625 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2628 let Constraints = "$src1 = $dst" in {
2629 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2630 i128mem, SSE_BIT_ITINS_P, 1>;
2631 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2632 i128mem, SSE_BIT_ITINS_P, 1>;
2633 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2634 i128mem, SSE_BIT_ITINS_P, 1>;
2635 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2636 i128mem, SSE_BIT_ITINS_P, 0>;
2637 } // Constraints = "$src1 = $dst"
2639 let Predicates = [HasAVX2] in {
2640 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2641 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2642 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2643 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2644 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2645 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2646 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2647 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2650 //===----------------------------------------------------------------------===//
2651 // SSE 1 & 2 - Logical Instructions
2652 //===----------------------------------------------------------------------===//
2654 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2656 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2657 SDNode OpNode, OpndItins itins> {
2658 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2659 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2662 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2663 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2666 let Constraints = "$src1 = $dst" in {
2667 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2668 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2671 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2672 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2677 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2678 let mayLoad = 0 in {
2679 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2681 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2683 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2687 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2688 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2691 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2693 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2695 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2696 // are all promoted to v2i64, and the patterns are covered by the int
2697 // version. This is needed in SSE only, because v2i64 isn't supported on
2698 // SSE1, but only on SSE2.
2699 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2700 !strconcat(OpcodeStr, "ps"), f128mem, [],
2701 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2702 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2704 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2705 !strconcat(OpcodeStr, "pd"), f128mem,
2706 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2707 (bc_v2i64 (v2f64 VR128:$src2))))],
2708 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2709 (memopv2i64 addr:$src2)))], 0>,
2711 let Constraints = "$src1 = $dst" in {
2712 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2713 !strconcat(OpcodeStr, "ps"), f128mem,
2714 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2715 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2716 (memopv2i64 addr:$src2)))]>, TB;
2718 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2719 !strconcat(OpcodeStr, "pd"), f128mem,
2720 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2721 (bc_v2i64 (v2f64 VR128:$src2))))],
2722 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2723 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2727 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2729 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2731 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2732 !strconcat(OpcodeStr, "ps"), f256mem,
2733 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2734 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2735 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2737 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2738 !strconcat(OpcodeStr, "pd"), f256mem,
2739 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2740 (bc_v4i64 (v4f64 VR256:$src2))))],
2741 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2742 (memopv4i64 addr:$src2)))], 0>,
2746 // AVX 256-bit packed logical ops forms
2747 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2748 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2749 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2750 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2752 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2753 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2754 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2755 let isCommutable = 0 in
2756 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2758 //===----------------------------------------------------------------------===//
2759 // SSE 1 & 2 - Arithmetic Instructions
2760 //===----------------------------------------------------------------------===//
2762 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2765 /// In addition, we also have a special variant of the scalar form here to
2766 /// represent the associated intrinsic operation. This form is unlike the
2767 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2768 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2770 /// These three forms can each be reg+reg or reg+mem.
2773 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2775 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2778 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2779 OpNode, FR32, f32mem,
2780 itins.s, Is2Addr>, XS;
2781 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2782 OpNode, FR64, f64mem,
2783 itins.d, Is2Addr>, XD;
2786 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2789 let mayLoad = 0 in {
2790 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2791 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2793 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2794 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2799 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2802 let mayLoad = 0 in {
2803 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2804 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2806 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2807 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2812 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2815 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2816 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2817 itins.s, Is2Addr>, XS;
2818 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2819 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2820 itins.d, Is2Addr>, XD;
2823 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2826 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2827 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2828 SSEPackedSingle, itins.s, Is2Addr>,
2831 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2832 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2833 SSEPackedDouble, itins.d, Is2Addr>,
2837 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2839 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2840 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2841 SSEPackedSingle, itins.s, 0>, TB;
2843 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2844 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2845 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2848 // Binary Arithmetic instructions
2849 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2850 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2852 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2853 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2855 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2856 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2858 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2859 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2862 let isCommutable = 0 in {
2863 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2864 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2866 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2867 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2868 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2869 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2871 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2872 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2874 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2875 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2877 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2878 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2879 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2880 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2882 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2883 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2885 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2886 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2887 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2888 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2892 let Constraints = "$src1 = $dst" in {
2893 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2894 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2895 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2896 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2897 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2898 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2900 let isCommutable = 0 in {
2901 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2902 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2903 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2904 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2905 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2906 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2907 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2908 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2909 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2910 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2911 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2912 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2913 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2914 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2919 /// In addition, we also have a special variant of the scalar form here to
2920 /// represent the associated intrinsic operation. This form is unlike the
2921 /// plain scalar form, in that it takes an entire vector (instead of a
2922 /// scalar) and leaves the top elements undefined.
2924 /// And, we have a special variant form for a full-vector intrinsic form.
2926 def SSE_SQRTP : OpndItins<
2927 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2930 def SSE_SQRTS : OpndItins<
2931 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2934 def SSE_RCPP : OpndItins<
2935 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2938 def SSE_RCPS : OpndItins<
2939 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2942 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2943 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2944 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2945 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2946 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2947 [(set FR32:$dst, (OpNode FR32:$src))]>;
2948 // For scalar unary operations, fold a load into the operation
2949 // only in OptForSize mode. It eliminates an instruction, but it also
2950 // eliminates a whole-register clobber (the load), so it introduces a
2951 // partial register update condition.
2952 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2953 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2954 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2955 Requires<[HasSSE1, OptForSize]>;
2956 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2957 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2958 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2959 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2960 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2961 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2964 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2965 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2966 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2967 !strconcat(OpcodeStr,
2968 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2970 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2971 !strconcat(OpcodeStr,
2972 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2973 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2974 (ins VR128:$src1, ssmem:$src2),
2975 !strconcat(OpcodeStr,
2976 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2979 /// sse1_fp_unop_p - SSE1 unops in packed form.
2980 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2982 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2983 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2984 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
2985 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2986 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2987 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
2990 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2991 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
2993 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2994 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2995 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
2997 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2998 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2999 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3003 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3004 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3005 Intrinsic V4F32Int, OpndItins itins> {
3006 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3007 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3008 [(set VR128:$dst, (V4F32Int VR128:$src))],
3010 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3011 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3012 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3016 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3017 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3018 Intrinsic V4F32Int, OpndItins itins> {
3019 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3020 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3021 [(set VR256:$dst, (V4F32Int VR256:$src))],
3023 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3024 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3025 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3029 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3030 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3031 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3032 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3033 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3034 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3035 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3036 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3037 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3038 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3039 Requires<[HasSSE2, OptForSize]>;
3040 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3041 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3042 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3043 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3044 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3045 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3048 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3049 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3050 let neverHasSideEffects = 1 in {
3051 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3052 !strconcat(OpcodeStr,
3053 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3055 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3056 !strconcat(OpcodeStr,
3057 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3059 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3060 (ins VR128:$src1, sdmem:$src2),
3061 !strconcat(OpcodeStr,
3062 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3065 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3066 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3067 SDNode OpNode, OpndItins itins> {
3068 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3069 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3070 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3071 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3072 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3073 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3076 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3077 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3079 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3080 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3081 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3083 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3084 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3085 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3089 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3090 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3091 Intrinsic V2F64Int, OpndItins itins> {
3092 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3093 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3094 [(set VR128:$dst, (V2F64Int VR128:$src))],
3096 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3097 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3098 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3102 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3103 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3104 Intrinsic V2F64Int, OpndItins itins> {
3105 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3106 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3107 [(set VR256:$dst, (V2F64Int VR256:$src))],
3109 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3110 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3111 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3115 let Predicates = [HasAVX] in {
3117 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3118 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3120 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3121 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3122 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3123 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3124 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3126 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3128 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3130 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3134 // Reciprocal approximations. Note that these typically require refinement
3135 // in order to obtain suitable precision.
3136 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3137 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3138 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3139 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3141 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3144 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3145 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3146 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3147 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3149 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3153 let AddedComplexity = 1 in {
3154 def : Pat<(f32 (fsqrt FR32:$src)),
3155 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3156 def : Pat<(f32 (fsqrt (load addr:$src))),
3157 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3158 Requires<[HasAVX, OptForSize]>;
3159 def : Pat<(f64 (fsqrt FR64:$src)),
3160 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3161 def : Pat<(f64 (fsqrt (load addr:$src))),
3162 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3163 Requires<[HasAVX, OptForSize]>;
3165 def : Pat<(f32 (X86frsqrt FR32:$src)),
3166 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3167 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3168 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3169 Requires<[HasAVX, OptForSize]>;
3171 def : Pat<(f32 (X86frcp FR32:$src)),
3172 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3173 def : Pat<(f32 (X86frcp (load addr:$src))),
3174 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3175 Requires<[HasAVX, OptForSize]>;
3178 let Predicates = [HasAVX], AddedComplexity = 1 in {
3179 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3180 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3181 (COPY_TO_REGCLASS VR128:$src, FR32)),
3183 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3184 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3186 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3187 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3188 (COPY_TO_REGCLASS VR128:$src, FR64)),
3190 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3191 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3193 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3194 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3195 (COPY_TO_REGCLASS VR128:$src, FR32)),
3197 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3198 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3200 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3201 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3202 (COPY_TO_REGCLASS VR128:$src, FR32)),
3204 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3205 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3209 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3211 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3212 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3213 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3215 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3216 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3218 // Reciprocal approximations. Note that these typically require refinement
3219 // in order to obtain suitable precision.
3220 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3222 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3223 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3225 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3227 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3228 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3230 // There is no f64 version of the reciprocal approximation instructions.
3232 //===----------------------------------------------------------------------===//
3233 // SSE 1 & 2 - Non-temporal stores
3234 //===----------------------------------------------------------------------===//
3236 let AddedComplexity = 400 in { // Prefer non-temporal versions
3237 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3238 (ins f128mem:$dst, VR128:$src),
3239 "movntps\t{$src, $dst|$dst, $src}",
3240 [(alignednontemporalstore (v4f32 VR128:$src),
3242 IIC_SSE_MOVNT>, VEX;
3243 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3244 (ins f128mem:$dst, VR128:$src),
3245 "movntpd\t{$src, $dst|$dst, $src}",
3246 [(alignednontemporalstore (v2f64 VR128:$src),
3248 IIC_SSE_MOVNT>, VEX;
3250 let ExeDomain = SSEPackedInt in
3251 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3252 (ins f128mem:$dst, VR128:$src),
3253 "movntdq\t{$src, $dst|$dst, $src}",
3254 [(alignednontemporalstore (v2i64 VR128:$src),
3256 IIC_SSE_MOVNT>, VEX;
3258 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3259 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3261 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3262 (ins f256mem:$dst, VR256:$src),
3263 "movntps\t{$src, $dst|$dst, $src}",
3264 [(alignednontemporalstore (v8f32 VR256:$src),
3266 IIC_SSE_MOVNT>, VEX;
3267 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3268 (ins f256mem:$dst, VR256:$src),
3269 "movntpd\t{$src, $dst|$dst, $src}",
3270 [(alignednontemporalstore (v4f64 VR256:$src),
3272 IIC_SSE_MOVNT>, VEX;
3273 let ExeDomain = SSEPackedInt in
3274 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3275 (ins f256mem:$dst, VR256:$src),
3276 "movntdq\t{$src, $dst|$dst, $src}",
3277 [(alignednontemporalstore (v4i64 VR256:$src),
3279 IIC_SSE_MOVNT>, VEX;
3282 let AddedComplexity = 400 in { // Prefer non-temporal versions
3283 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3284 "movntps\t{$src, $dst|$dst, $src}",
3285 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3287 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3288 "movntpd\t{$src, $dst|$dst, $src}",
3289 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3292 let ExeDomain = SSEPackedInt in
3293 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3294 "movntdq\t{$src, $dst|$dst, $src}",
3295 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3298 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3299 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3301 // There is no AVX form for instructions below this point
3302 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3303 "movnti{l}\t{$src, $dst|$dst, $src}",
3304 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3306 TB, Requires<[HasSSE2]>;
3307 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3308 "movnti{q}\t{$src, $dst|$dst, $src}",
3309 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3311 TB, Requires<[HasSSE2]>;
3314 //===----------------------------------------------------------------------===//
3315 // SSE 1 & 2 - Prefetch and memory fence
3316 //===----------------------------------------------------------------------===//
3318 // Prefetch intrinsic.
3319 let Predicates = [HasSSE1] in {
3320 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3321 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3322 IIC_SSE_PREFETCH>, TB;
3323 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3324 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3325 IIC_SSE_PREFETCH>, TB;
3326 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3327 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3328 IIC_SSE_PREFETCH>, TB;
3329 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3330 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3331 IIC_SSE_PREFETCH>, TB;
3335 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3336 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3337 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3339 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3340 // was introduced with SSE2, it's backward compatible.
3341 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3343 // Load, store, and memory fence
3344 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3345 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3346 TB, Requires<[HasSSE1]>;
3347 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3348 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3349 TB, Requires<[HasSSE2]>;
3350 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3351 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3352 TB, Requires<[HasSSE2]>;
3354 def : Pat<(X86SFence), (SFENCE)>;
3355 def : Pat<(X86LFence), (LFENCE)>;
3356 def : Pat<(X86MFence), (MFENCE)>;
3358 //===----------------------------------------------------------------------===//
3359 // SSE 1 & 2 - Load/Store XCSR register
3360 //===----------------------------------------------------------------------===//
3362 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3363 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3364 IIC_SSE_LDMXCSR>, VEX;
3365 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3366 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3367 IIC_SSE_STMXCSR>, VEX;
3369 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3370 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3372 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3373 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3376 //===---------------------------------------------------------------------===//
3377 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3378 //===---------------------------------------------------------------------===//
3380 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3382 let neverHasSideEffects = 1 in {
3383 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3384 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3386 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3387 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3390 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3391 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3393 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3394 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3398 let isCodeGenOnly = 1 in {
3399 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3400 "movdqa\t{$src, $dst|$dst, $src}", [],
3403 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3404 "movdqa\t{$src, $dst|$dst, $src}", [],
3407 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3408 "movdqu\t{$src, $dst|$dst, $src}", [],
3411 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3412 "movdqu\t{$src, $dst|$dst, $src}", [],
3417 let canFoldAsLoad = 1, mayLoad = 1 in {
3418 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3419 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3421 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3422 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3424 let Predicates = [HasAVX] in {
3425 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3426 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3428 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3429 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3434 let mayStore = 1 in {
3435 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3436 (ins i128mem:$dst, VR128:$src),
3437 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3439 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3440 (ins i256mem:$dst, VR256:$src),
3441 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3443 let Predicates = [HasAVX] in {
3444 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3445 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3447 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3448 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3453 let neverHasSideEffects = 1 in
3454 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3455 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3457 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3458 "movdqu\t{$src, $dst|$dst, $src}",
3459 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3462 let isCodeGenOnly = 1 in {
3463 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3464 "movdqa\t{$src, $dst|$dst, $src}", [],
3467 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3468 "movdqu\t{$src, $dst|$dst, $src}",
3469 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3472 let canFoldAsLoad = 1, mayLoad = 1 in {
3473 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3474 "movdqa\t{$src, $dst|$dst, $src}",
3475 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3477 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3478 "movdqu\t{$src, $dst|$dst, $src}",
3479 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3481 XS, Requires<[HasSSE2]>;
3484 let mayStore = 1 in {
3485 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3486 "movdqa\t{$src, $dst|$dst, $src}",
3487 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3489 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3490 "movdqu\t{$src, $dst|$dst, $src}",
3491 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3493 XS, Requires<[HasSSE2]>;
3496 // Intrinsic forms of MOVDQU load and store
3497 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3498 "vmovdqu\t{$src, $dst|$dst, $src}",
3499 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3501 XS, VEX, Requires<[HasAVX]>;
3503 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3504 "movdqu\t{$src, $dst|$dst, $src}",
3505 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3507 XS, Requires<[HasSSE2]>;
3509 } // ExeDomain = SSEPackedInt
3511 let Predicates = [HasAVX] in {
3512 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3513 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3516 //===---------------------------------------------------------------------===//
3517 // SSE2 - Packed Integer Arithmetic Instructions
3518 //===---------------------------------------------------------------------===//
3520 def SSE_PMADD : OpndItins<
3521 IIC_SSE_PMADD, IIC_SSE_PMADD
3524 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3526 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3527 RegisterClass RC, PatFrag memop_frag,
3528 X86MemOperand x86memop,
3530 bit IsCommutable = 0,
3532 let isCommutable = IsCommutable in
3533 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3534 (ins RC:$src1, RC:$src2),
3536 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3537 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3538 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3539 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3540 (ins RC:$src1, x86memop:$src2),
3542 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3543 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3544 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3548 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3549 string OpcodeStr, SDNode OpNode,
3550 SDNode OpNode2, RegisterClass RC,
3551 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3552 ShiftOpndItins itins,
3554 // src2 is always 128-bit
3555 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3556 (ins RC:$src1, VR128:$src2),
3558 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3559 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3560 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3562 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3563 (ins RC:$src1, i128mem:$src2),
3565 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3566 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3567 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3568 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3569 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3570 (ins RC:$src1, i32i8imm:$src2),
3572 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3573 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3574 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3577 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3578 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3579 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3580 PatFrag memop_frag, X86MemOperand x86memop,
3582 bit IsCommutable = 0, bit Is2Addr = 1> {
3583 let isCommutable = IsCommutable in
3584 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3585 (ins RC:$src1, RC:$src2),
3587 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3588 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3589 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3590 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3591 (ins RC:$src1, x86memop:$src2),
3593 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3594 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3595 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3596 (bitconvert (memop_frag addr:$src2)))))]>;
3598 } // ExeDomain = SSEPackedInt
3600 // 128-bit Integer Arithmetic
3602 let Predicates = [HasAVX] in {
3603 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3604 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3606 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3607 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3608 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3609 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3610 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3611 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3612 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3613 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3614 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3615 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3616 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3617 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3618 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3619 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3620 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3621 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3622 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3623 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3627 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3628 VR128, memopv2i64, i128mem,
3629 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3630 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3631 VR128, memopv2i64, i128mem,
3632 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3633 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3634 VR128, memopv2i64, i128mem,
3635 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3636 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3637 VR128, memopv2i64, i128mem,
3638 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3639 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3640 VR128, memopv2i64, i128mem,
3641 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3642 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3643 VR128, memopv2i64, i128mem,
3644 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3645 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3646 VR128, memopv2i64, i128mem,
3647 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3648 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3649 VR128, memopv2i64, i128mem,
3650 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3651 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3652 VR128, memopv2i64, i128mem,
3653 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3654 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3655 VR128, memopv2i64, i128mem,
3656 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3657 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3658 VR128, memopv2i64, i128mem,
3659 SSE_PMADD, 1, 0>, VEX_4V;
3660 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3661 VR128, memopv2i64, i128mem,
3662 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3663 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3664 VR128, memopv2i64, i128mem,
3665 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3666 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3667 VR128, memopv2i64, i128mem,
3668 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3669 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3670 VR128, memopv2i64, i128mem,
3671 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3672 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3673 VR128, memopv2i64, i128mem,
3674 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3675 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3676 VR128, memopv2i64, i128mem,
3677 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3678 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3679 VR128, memopv2i64, i128mem,
3680 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3683 let Predicates = [HasAVX2] in {
3684 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3685 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3686 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3687 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3688 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3689 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3690 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3691 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3692 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3693 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3694 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3695 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3696 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3697 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3698 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3699 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3700 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3701 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3702 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3703 VR256, memopv4i64, i256mem,
3704 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3707 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3708 VR256, memopv4i64, i256mem,
3709 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3710 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3711 VR256, memopv4i64, i256mem,
3712 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3713 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3714 VR256, memopv4i64, i256mem,
3715 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3716 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3717 VR256, memopv4i64, i256mem,
3718 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3719 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3720 VR256, memopv4i64, i256mem,
3721 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3722 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3723 VR256, memopv4i64, i256mem,
3724 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3725 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3726 VR256, memopv4i64, i256mem,
3727 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3728 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3729 VR256, memopv4i64, i256mem,
3730 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3731 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3732 VR256, memopv4i64, i256mem,
3733 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3734 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3735 VR256, memopv4i64, i256mem,
3736 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3737 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3738 VR256, memopv4i64, i256mem,
3739 SSE_PMADD, 1, 0>, VEX_4V;
3740 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3741 VR256, memopv4i64, i256mem,
3742 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3743 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3744 VR256, memopv4i64, i256mem,
3745 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3746 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3747 VR256, memopv4i64, i256mem,
3748 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3749 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3750 VR256, memopv4i64, i256mem,
3751 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3752 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3753 VR256, memopv4i64, i256mem,
3754 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3755 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3756 VR256, memopv4i64, i256mem,
3757 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3758 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3759 VR256, memopv4i64, i256mem,
3760 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3763 let Constraints = "$src1 = $dst" in {
3764 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3765 i128mem, SSE_INTALU_ITINS_P, 1>;
3766 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3767 i128mem, SSE_INTALU_ITINS_P, 1>;
3768 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3769 i128mem, SSE_INTALU_ITINS_P, 1>;
3770 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3771 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3772 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3773 i128mem, SSE_INTMUL_ITINS_P, 1>;
3774 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3775 i128mem, SSE_INTALU_ITINS_P>;
3776 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3777 i128mem, SSE_INTALU_ITINS_P>;
3778 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3779 i128mem, SSE_INTALU_ITINS_P>;
3780 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3781 i128mem, SSE_INTALUQ_ITINS_P>;
3782 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3783 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3786 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3787 VR128, memopv2i64, i128mem,
3788 SSE_INTALU_ITINS_P>;
3789 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3790 VR128, memopv2i64, i128mem,
3791 SSE_INTALU_ITINS_P>;
3792 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3793 VR128, memopv2i64, i128mem,
3794 SSE_INTALU_ITINS_P>;
3795 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3796 VR128, memopv2i64, i128mem,
3797 SSE_INTALU_ITINS_P>;
3798 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3799 VR128, memopv2i64, i128mem,
3800 SSE_INTALU_ITINS_P, 1>;
3801 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3802 VR128, memopv2i64, i128mem,
3803 SSE_INTALU_ITINS_P, 1>;
3804 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3805 VR128, memopv2i64, i128mem,
3806 SSE_INTALU_ITINS_P, 1>;
3807 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3808 VR128, memopv2i64, i128mem,
3809 SSE_INTALU_ITINS_P, 1>;
3810 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3811 VR128, memopv2i64, i128mem,
3812 SSE_INTMUL_ITINS_P, 1>;
3813 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3814 VR128, memopv2i64, i128mem,
3815 SSE_INTMUL_ITINS_P, 1>;
3816 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3817 VR128, memopv2i64, i128mem,
3819 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3820 VR128, memopv2i64, i128mem,
3821 SSE_INTALU_ITINS_P, 1>;
3822 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3823 VR128, memopv2i64, i128mem,
3824 SSE_INTALU_ITINS_P, 1>;
3825 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3826 VR128, memopv2i64, i128mem,
3827 SSE_INTALU_ITINS_P, 1>;
3828 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3829 VR128, memopv2i64, i128mem,
3830 SSE_INTALU_ITINS_P, 1>;
3831 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3832 VR128, memopv2i64, i128mem,
3833 SSE_INTALU_ITINS_P, 1>;
3834 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3835 VR128, memopv2i64, i128mem,
3836 SSE_INTALU_ITINS_P, 1>;
3837 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3838 VR128, memopv2i64, i128mem,
3839 SSE_INTALU_ITINS_P, 1>;
3841 } // Constraints = "$src1 = $dst"
3843 //===---------------------------------------------------------------------===//
3844 // SSE2 - Packed Integer Logical Instructions
3845 //===---------------------------------------------------------------------===//
3847 let Predicates = [HasAVX] in {
3848 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3849 VR128, v8i16, v8i16, bc_v8i16,
3850 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3851 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3852 VR128, v4i32, v4i32, bc_v4i32,
3853 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3854 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3855 VR128, v2i64, v2i64, bc_v2i64,
3856 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3858 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3859 VR128, v8i16, v8i16, bc_v8i16,
3860 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3861 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3862 VR128, v4i32, v4i32, bc_v4i32,
3863 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3864 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3865 VR128, v2i64, v2i64, bc_v2i64,
3866 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3868 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3869 VR128, v8i16, v8i16, bc_v8i16,
3870 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3871 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3872 VR128, v4i32, v4i32, bc_v4i32,
3873 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3875 let ExeDomain = SSEPackedInt in {
3876 // 128-bit logical shifts.
3877 def VPSLLDQri : PDIi8<0x73, MRM7r,
3878 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3879 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3881 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3883 def VPSRLDQri : PDIi8<0x73, MRM3r,
3884 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3885 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3887 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3889 // PSRADQri doesn't exist in SSE[1-3].
3891 } // Predicates = [HasAVX]
3893 let Predicates = [HasAVX2] in {
3894 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3895 VR256, v16i16, v8i16, bc_v8i16,
3896 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3897 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3898 VR256, v8i32, v4i32, bc_v4i32,
3899 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3900 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3901 VR256, v4i64, v2i64, bc_v2i64,
3902 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3904 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3905 VR256, v16i16, v8i16, bc_v8i16,
3906 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3907 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3908 VR256, v8i32, v4i32, bc_v4i32,
3909 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3910 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3911 VR256, v4i64, v2i64, bc_v2i64,
3912 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3914 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3915 VR256, v16i16, v8i16, bc_v8i16,
3916 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3917 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3918 VR256, v8i32, v4i32, bc_v4i32,
3919 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3921 let ExeDomain = SSEPackedInt in {
3922 // 256-bit logical shifts.
3923 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3924 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3925 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3927 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3929 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3930 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3931 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3933 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3935 // PSRADQYri doesn't exist in SSE[1-3].
3937 } // Predicates = [HasAVX2]
3939 let Constraints = "$src1 = $dst" in {
3940 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3941 VR128, v8i16, v8i16, bc_v8i16,
3942 SSE_INTSHIFT_ITINS_P>;
3943 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3944 VR128, v4i32, v4i32, bc_v4i32,
3945 SSE_INTSHIFT_ITINS_P>;
3946 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3947 VR128, v2i64, v2i64, bc_v2i64,
3948 SSE_INTSHIFT_ITINS_P>;
3950 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3951 VR128, v8i16, v8i16, bc_v8i16,
3952 SSE_INTSHIFT_ITINS_P>;
3953 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3954 VR128, v4i32, v4i32, bc_v4i32,
3955 SSE_INTSHIFT_ITINS_P>;
3956 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3957 VR128, v2i64, v2i64, bc_v2i64,
3958 SSE_INTSHIFT_ITINS_P>;
3960 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3961 VR128, v8i16, v8i16, bc_v8i16,
3962 SSE_INTSHIFT_ITINS_P>;
3963 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3964 VR128, v4i32, v4i32, bc_v4i32,
3965 SSE_INTSHIFT_ITINS_P>;
3967 let ExeDomain = SSEPackedInt in {
3968 // 128-bit logical shifts.
3969 def PSLLDQri : PDIi8<0x73, MRM7r,
3970 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3971 "pslldq\t{$src2, $dst|$dst, $src2}",
3973 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3974 def PSRLDQri : PDIi8<0x73, MRM3r,
3975 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3976 "psrldq\t{$src2, $dst|$dst, $src2}",
3978 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3979 // PSRADQri doesn't exist in SSE[1-3].
3981 } // Constraints = "$src1 = $dst"
3983 let Predicates = [HasAVX] in {
3984 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3985 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3986 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3987 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3988 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3989 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3991 // Shift up / down and insert zero's.
3992 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3993 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3994 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3995 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3998 let Predicates = [HasAVX2] in {
3999 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4000 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4001 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4002 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4005 let Predicates = [HasSSE2] in {
4006 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4007 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4008 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4009 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4010 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4011 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4013 // Shift up / down and insert zero's.
4014 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4015 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4016 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4017 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4020 //===---------------------------------------------------------------------===//
4021 // SSE2 - Packed Integer Comparison Instructions
4022 //===---------------------------------------------------------------------===//
4024 let Predicates = [HasAVX] in {
4025 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4026 VR128, memopv2i64, i128mem,
4027 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4028 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4029 VR128, memopv2i64, i128mem,
4030 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4031 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4032 VR128, memopv2i64, i128mem,
4033 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4034 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4035 VR128, memopv2i64, i128mem,
4036 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4037 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4038 VR128, memopv2i64, i128mem,
4039 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4040 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4041 VR128, memopv2i64, i128mem,
4042 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4045 let Predicates = [HasAVX2] in {
4046 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4047 VR256, memopv4i64, i256mem,
4048 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4049 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4050 VR256, memopv4i64, i256mem,
4051 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4052 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4053 VR256, memopv4i64, i256mem,
4054 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4055 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4056 VR256, memopv4i64, i256mem,
4057 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4058 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4059 VR256, memopv4i64, i256mem,
4060 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4061 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4062 VR256, memopv4i64, i256mem,
4063 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4066 let Constraints = "$src1 = $dst" in {
4067 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4068 VR128, memopv2i64, i128mem,
4069 SSE_INTALU_ITINS_P, 1>;
4070 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4071 VR128, memopv2i64, i128mem,
4072 SSE_INTALU_ITINS_P, 1>;
4073 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4074 VR128, memopv2i64, i128mem,
4075 SSE_INTALU_ITINS_P, 1>;
4076 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4077 VR128, memopv2i64, i128mem,
4078 SSE_INTALU_ITINS_P>;
4079 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4080 VR128, memopv2i64, i128mem,
4081 SSE_INTALU_ITINS_P>;
4082 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4083 VR128, memopv2i64, i128mem,
4084 SSE_INTALU_ITINS_P>;
4085 } // Constraints = "$src1 = $dst"
4087 //===---------------------------------------------------------------------===//
4088 // SSE2 - Packed Integer Pack Instructions
4089 //===---------------------------------------------------------------------===//
4091 let Predicates = [HasAVX] in {
4092 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4093 VR128, memopv2i64, i128mem,
4094 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4095 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4096 VR128, memopv2i64, i128mem,
4097 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4098 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4099 VR128, memopv2i64, i128mem,
4100 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4103 let Predicates = [HasAVX2] in {
4104 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4105 VR256, memopv4i64, i256mem,
4106 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4107 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4108 VR256, memopv4i64, i256mem,
4109 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4110 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4111 VR256, memopv4i64, i256mem,
4112 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4115 let Constraints = "$src1 = $dst" in {
4116 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4117 VR128, memopv2i64, i128mem,
4118 SSE_INTALU_ITINS_P>;
4119 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4120 VR128, memopv2i64, i128mem,
4121 SSE_INTALU_ITINS_P>;
4122 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4123 VR128, memopv2i64, i128mem,
4124 SSE_INTALU_ITINS_P>;
4125 } // Constraints = "$src1 = $dst"
4127 //===---------------------------------------------------------------------===//
4128 // SSE2 - Packed Integer Shuffle Instructions
4129 //===---------------------------------------------------------------------===//
4131 let ExeDomain = SSEPackedInt in {
4132 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4133 def ri : Ii8<0x70, MRMSrcReg,
4134 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4135 !strconcat(OpcodeStr,
4136 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4137 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4139 def mi : Ii8<0x70, MRMSrcMem,
4140 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4141 !strconcat(OpcodeStr,
4142 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4144 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4149 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4150 def Yri : Ii8<0x70, MRMSrcReg,
4151 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4152 !strconcat(OpcodeStr,
4153 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4154 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4155 def Ymi : Ii8<0x70, MRMSrcMem,
4156 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4157 !strconcat(OpcodeStr,
4158 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4160 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4161 (i8 imm:$src2))))]>;
4163 } // ExeDomain = SSEPackedInt
4165 let Predicates = [HasAVX] in {
4166 let AddedComplexity = 5 in
4167 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4169 // SSE2 with ImmT == Imm8 and XS prefix.
4170 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4172 // SSE2 with ImmT == Imm8 and XD prefix.
4173 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4175 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4176 (VPSHUFDmi addr:$src1, imm:$imm)>;
4177 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4178 (VPSHUFDri VR128:$src1, imm:$imm)>;
4181 let Predicates = [HasAVX2] in {
4182 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4183 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4184 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4187 let Predicates = [HasSSE2] in {
4188 let AddedComplexity = 5 in
4189 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4191 // SSE2 with ImmT == Imm8 and XS prefix.
4192 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4194 // SSE2 with ImmT == Imm8 and XD prefix.
4195 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4197 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4198 (PSHUFDmi addr:$src1, imm:$imm)>;
4199 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4200 (PSHUFDri VR128:$src1, imm:$imm)>;
4203 //===---------------------------------------------------------------------===//
4204 // SSE2 - Packed Integer Unpack Instructions
4205 //===---------------------------------------------------------------------===//
4207 let ExeDomain = SSEPackedInt in {
4208 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4209 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4210 def rr : PDI<opc, MRMSrcReg,
4211 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4213 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4214 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4215 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4217 def rm : PDI<opc, MRMSrcMem,
4218 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4220 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4221 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4222 [(set VR128:$dst, (OpNode VR128:$src1,
4223 (bc_frag (memopv2i64
4228 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4229 SDNode OpNode, PatFrag bc_frag> {
4230 def Yrr : PDI<opc, MRMSrcReg,
4231 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4232 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4233 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4234 def Yrm : PDI<opc, MRMSrcMem,
4235 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4236 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4237 [(set VR256:$dst, (OpNode VR256:$src1,
4238 (bc_frag (memopv4i64 addr:$src2))))]>;
4241 let Predicates = [HasAVX] in {
4242 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4243 bc_v16i8, 0>, VEX_4V;
4244 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4245 bc_v8i16, 0>, VEX_4V;
4246 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4247 bc_v4i32, 0>, VEX_4V;
4248 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4249 bc_v2i64, 0>, VEX_4V;
4251 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4252 bc_v16i8, 0>, VEX_4V;
4253 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4254 bc_v8i16, 0>, VEX_4V;
4255 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4256 bc_v4i32, 0>, VEX_4V;
4257 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4258 bc_v2i64, 0>, VEX_4V;
4261 let Predicates = [HasAVX2] in {
4262 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4264 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4266 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4268 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4271 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4273 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4275 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4277 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4281 let Constraints = "$src1 = $dst" in {
4282 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4284 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4286 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4288 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4291 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4293 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4295 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4297 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4300 } // ExeDomain = SSEPackedInt
4302 // Patterns for using AVX1 instructions with integer vectors
4303 // Here to give AVX2 priority
4304 let Predicates = [HasAVX] in {
4305 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4306 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4307 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4308 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4309 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4310 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4311 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4312 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4314 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4315 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4316 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4317 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4318 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4319 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4320 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4321 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4324 //===---------------------------------------------------------------------===//
4325 // SSE2 - Packed Integer Extract and Insert
4326 //===---------------------------------------------------------------------===//
4328 let ExeDomain = SSEPackedInt in {
4329 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4330 def rri : Ii8<0xC4, MRMSrcReg,
4331 (outs VR128:$dst), (ins VR128:$src1,
4332 GR32:$src2, i32i8imm:$src3),
4334 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4335 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4337 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4338 def rmi : Ii8<0xC4, MRMSrcMem,
4339 (outs VR128:$dst), (ins VR128:$src1,
4340 i16mem:$src2, i32i8imm:$src3),
4342 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4343 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4345 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4346 imm:$src3))], IIC_SSE_PINSRW>;
4350 let Predicates = [HasAVX] in
4351 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4352 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4353 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4354 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4355 imm:$src2))]>, TB, OpSize, VEX;
4356 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4357 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4358 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4359 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4360 imm:$src2))], IIC_SSE_PEXTRW>;
4363 let Predicates = [HasAVX] in {
4364 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4365 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4366 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4367 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4368 []>, TB, OpSize, VEX_4V;
4371 let Constraints = "$src1 = $dst" in
4372 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4374 } // ExeDomain = SSEPackedInt
4376 //===---------------------------------------------------------------------===//
4377 // SSE2 - Packed Mask Creation
4378 //===---------------------------------------------------------------------===//
4380 let ExeDomain = SSEPackedInt in {
4382 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4383 "pmovmskb\t{$src, $dst|$dst, $src}",
4384 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4385 IIC_SSE_MOVMSK>, VEX;
4386 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4387 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4389 let Predicates = [HasAVX2] in {
4390 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4391 "pmovmskb\t{$src, $dst|$dst, $src}",
4392 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4393 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4394 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4397 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4398 "pmovmskb\t{$src, $dst|$dst, $src}",
4399 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4402 } // ExeDomain = SSEPackedInt
4404 //===---------------------------------------------------------------------===//
4405 // SSE2 - Conditional Store
4406 //===---------------------------------------------------------------------===//
4408 let ExeDomain = SSEPackedInt in {
4411 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4412 (ins VR128:$src, VR128:$mask),
4413 "maskmovdqu\t{$mask, $src|$src, $mask}",
4414 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4415 IIC_SSE_MASKMOV>, VEX;
4417 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4418 (ins VR128:$src, VR128:$mask),
4419 "maskmovdqu\t{$mask, $src|$src, $mask}",
4420 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4421 IIC_SSE_MASKMOV>, VEX;
4424 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4425 "maskmovdqu\t{$mask, $src|$src, $mask}",
4426 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4429 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4430 "maskmovdqu\t{$mask, $src|$src, $mask}",
4431 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4434 } // ExeDomain = SSEPackedInt
4436 //===---------------------------------------------------------------------===//
4437 // SSE2 - Move Doubleword
4438 //===---------------------------------------------------------------------===//
4440 //===---------------------------------------------------------------------===//
4441 // Move Int Doubleword to Packed Double Int
4443 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4444 "movd\t{$src, $dst|$dst, $src}",
4446 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4448 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4449 "movd\t{$src, $dst|$dst, $src}",
4451 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4454 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4455 "mov{d|q}\t{$src, $dst|$dst, $src}",
4457 (v2i64 (scalar_to_vector GR64:$src)))],
4458 IIC_SSE_MOVDQ>, VEX;
4459 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4460 "mov{d|q}\t{$src, $dst|$dst, $src}",
4461 [(set FR64:$dst, (bitconvert GR64:$src))],
4462 IIC_SSE_MOVDQ>, VEX;
4464 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4465 "movd\t{$src, $dst|$dst, $src}",
4467 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4468 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4469 "movd\t{$src, $dst|$dst, $src}",
4471 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4473 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4474 "mov{d|q}\t{$src, $dst|$dst, $src}",
4476 (v2i64 (scalar_to_vector GR64:$src)))],
4478 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4479 "mov{d|q}\t{$src, $dst|$dst, $src}",
4480 [(set FR64:$dst, (bitconvert GR64:$src))],
4483 //===---------------------------------------------------------------------===//
4484 // Move Int Doubleword to Single Scalar
4486 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4487 "movd\t{$src, $dst|$dst, $src}",
4488 [(set FR32:$dst, (bitconvert GR32:$src))],
4489 IIC_SSE_MOVDQ>, VEX;
4491 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4492 "movd\t{$src, $dst|$dst, $src}",
4493 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4496 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4497 "movd\t{$src, $dst|$dst, $src}",
4498 [(set FR32:$dst, (bitconvert GR32:$src))],
4501 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4502 "movd\t{$src, $dst|$dst, $src}",
4503 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4506 //===---------------------------------------------------------------------===//
4507 // Move Packed Doubleword Int to Packed Double Int
4509 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4510 "movd\t{$src, $dst|$dst, $src}",
4511 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4512 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4513 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4514 (ins i32mem:$dst, VR128:$src),
4515 "movd\t{$src, $dst|$dst, $src}",
4516 [(store (i32 (vector_extract (v4i32 VR128:$src),
4517 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4519 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4520 "movd\t{$src, $dst|$dst, $src}",
4521 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4522 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4523 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4524 "movd\t{$src, $dst|$dst, $src}",
4525 [(store (i32 (vector_extract (v4i32 VR128:$src),
4526 (iPTR 0))), addr:$dst)],
4529 //===---------------------------------------------------------------------===//
4530 // Move Packed Doubleword Int first element to Doubleword Int
4532 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4533 "mov{d|q}\t{$src, $dst|$dst, $src}",
4534 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4537 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4539 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4540 "mov{d|q}\t{$src, $dst|$dst, $src}",
4541 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4545 //===---------------------------------------------------------------------===//
4546 // Bitcast FR64 <-> GR64
4548 let Predicates = [HasAVX] in
4549 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4550 "vmovq\t{$src, $dst|$dst, $src}",
4551 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4553 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4554 "mov{d|q}\t{$src, $dst|$dst, $src}",
4555 [(set GR64:$dst, (bitconvert FR64:$src))],
4556 IIC_SSE_MOVDQ>, VEX;
4557 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4558 "movq\t{$src, $dst|$dst, $src}",
4559 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4560 IIC_SSE_MOVDQ>, VEX;
4562 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4563 "movq\t{$src, $dst|$dst, $src}",
4564 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4566 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4567 "mov{d|q}\t{$src, $dst|$dst, $src}",
4568 [(set GR64:$dst, (bitconvert FR64:$src))],
4570 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4571 "movq\t{$src, $dst|$dst, $src}",
4572 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4575 //===---------------------------------------------------------------------===//
4576 // Move Scalar Single to Double Int
4578 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4579 "movd\t{$src, $dst|$dst, $src}",
4580 [(set GR32:$dst, (bitconvert FR32:$src))],
4581 IIC_SSE_MOVD_ToGP>, VEX;
4582 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4583 "movd\t{$src, $dst|$dst, $src}",
4584 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4585 IIC_SSE_MOVDQ>, VEX;
4586 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4587 "movd\t{$src, $dst|$dst, $src}",
4588 [(set GR32:$dst, (bitconvert FR32:$src))],
4590 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4592 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4595 //===---------------------------------------------------------------------===//
4596 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4598 let AddedComplexity = 15 in {
4599 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4600 "movd\t{$src, $dst|$dst, $src}",
4601 [(set VR128:$dst, (v4i32 (X86vzmovl
4602 (v4i32 (scalar_to_vector GR32:$src)))))],
4603 IIC_SSE_MOVDQ>, VEX;
4604 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4605 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4606 [(set VR128:$dst, (v2i64 (X86vzmovl
4607 (v2i64 (scalar_to_vector GR64:$src)))))],
4611 let AddedComplexity = 15 in {
4612 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4613 "movd\t{$src, $dst|$dst, $src}",
4614 [(set VR128:$dst, (v4i32 (X86vzmovl
4615 (v4i32 (scalar_to_vector GR32:$src)))))],
4617 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4618 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4619 [(set VR128:$dst, (v2i64 (X86vzmovl
4620 (v2i64 (scalar_to_vector GR64:$src)))))],
4624 let AddedComplexity = 20 in {
4625 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4626 "movd\t{$src, $dst|$dst, $src}",
4628 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4629 (loadi32 addr:$src))))))],
4630 IIC_SSE_MOVDQ>, VEX;
4631 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4632 "movd\t{$src, $dst|$dst, $src}",
4634 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4635 (loadi32 addr:$src))))))],
4639 let Predicates = [HasAVX] in {
4640 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4641 let AddedComplexity = 20 in {
4642 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4643 (VMOVZDI2PDIrm addr:$src)>;
4644 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4645 (VMOVZDI2PDIrm addr:$src)>;
4647 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4648 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4649 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4650 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4651 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4652 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4653 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4656 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4657 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4658 (MOVZDI2PDIrm addr:$src)>;
4659 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4660 (MOVZDI2PDIrm addr:$src)>;
4663 // These are the correct encodings of the instructions so that we know how to
4664 // read correct assembly, even though we continue to emit the wrong ones for
4665 // compatibility with Darwin's buggy assembler.
4666 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4667 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4668 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4669 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4670 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4671 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4672 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4673 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4674 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4675 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4676 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4677 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4679 //===---------------------------------------------------------------------===//
4680 // SSE2 - Move Quadword
4681 //===---------------------------------------------------------------------===//
4683 //===---------------------------------------------------------------------===//
4684 // Move Quadword Int to Packed Quadword Int
4686 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4687 "vmovq\t{$src, $dst|$dst, $src}",
4689 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4690 VEX, Requires<[HasAVX]>;
4691 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4692 "movq\t{$src, $dst|$dst, $src}",
4694 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4696 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4698 //===---------------------------------------------------------------------===//
4699 // Move Packed Quadword Int to Quadword Int
4701 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4702 "movq\t{$src, $dst|$dst, $src}",
4703 [(store (i64 (vector_extract (v2i64 VR128:$src),
4704 (iPTR 0))), addr:$dst)],
4705 IIC_SSE_MOVDQ>, VEX;
4706 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4707 "movq\t{$src, $dst|$dst, $src}",
4708 [(store (i64 (vector_extract (v2i64 VR128:$src),
4709 (iPTR 0))), addr:$dst)],
4712 //===---------------------------------------------------------------------===//
4713 // Store / copy lower 64-bits of a XMM register.
4715 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4716 "movq\t{$src, $dst|$dst, $src}",
4717 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4718 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4719 "movq\t{$src, $dst|$dst, $src}",
4720 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4723 let AddedComplexity = 20 in
4724 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4725 "vmovq\t{$src, $dst|$dst, $src}",
4727 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4728 (loadi64 addr:$src))))))],
4730 XS, VEX, Requires<[HasAVX]>;
4732 let AddedComplexity = 20 in
4733 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4734 "movq\t{$src, $dst|$dst, $src}",
4736 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4737 (loadi64 addr:$src))))))],
4739 XS, Requires<[HasSSE2]>;
4741 let Predicates = [HasAVX], AddedComplexity = 20 in {
4742 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4743 (VMOVZQI2PQIrm addr:$src)>;
4744 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4745 (VMOVZQI2PQIrm addr:$src)>;
4746 def : Pat<(v2i64 (X86vzload addr:$src)),
4747 (VMOVZQI2PQIrm addr:$src)>;
4750 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4751 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4752 (MOVZQI2PQIrm addr:$src)>;
4753 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4754 (MOVZQI2PQIrm addr:$src)>;
4755 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4758 let Predicates = [HasAVX] in {
4759 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4760 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4761 def : Pat<(v4i64 (X86vzload addr:$src)),
4762 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4765 //===---------------------------------------------------------------------===//
4766 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4767 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4769 let AddedComplexity = 15 in
4770 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4771 "vmovq\t{$src, $dst|$dst, $src}",
4772 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4774 XS, VEX, Requires<[HasAVX]>;
4775 let AddedComplexity = 15 in
4776 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4777 "movq\t{$src, $dst|$dst, $src}",
4778 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4780 XS, Requires<[HasSSE2]>;
4782 let AddedComplexity = 20 in
4783 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4784 "vmovq\t{$src, $dst|$dst, $src}",
4785 [(set VR128:$dst, (v2i64 (X86vzmovl
4786 (loadv2i64 addr:$src))))],
4788 XS, VEX, Requires<[HasAVX]>;
4789 let AddedComplexity = 20 in {
4790 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4791 "movq\t{$src, $dst|$dst, $src}",
4792 [(set VR128:$dst, (v2i64 (X86vzmovl
4793 (loadv2i64 addr:$src))))],
4795 XS, Requires<[HasSSE2]>;
4798 let AddedComplexity = 20 in {
4799 let Predicates = [HasAVX] in {
4800 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4801 (VMOVZPQILo2PQIrm addr:$src)>;
4802 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4803 (VMOVZPQILo2PQIrr VR128:$src)>;
4805 let Predicates = [HasSSE2] in {
4806 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4807 (MOVZPQILo2PQIrm addr:$src)>;
4808 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4809 (MOVZPQILo2PQIrr VR128:$src)>;
4813 // Instructions to match in the assembler
4814 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4815 "movq\t{$src, $dst|$dst, $src}", [],
4816 IIC_SSE_MOVDQ>, VEX, VEX_W;
4817 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4818 "movq\t{$src, $dst|$dst, $src}", [],
4819 IIC_SSE_MOVDQ>, VEX, VEX_W;
4820 // Recognize "movd" with GR64 destination, but encode as a "movq"
4821 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4822 "movd\t{$src, $dst|$dst, $src}", [],
4823 IIC_SSE_MOVDQ>, VEX, VEX_W;
4825 // Instructions for the disassembler
4826 // xr = XMM register
4829 let Predicates = [HasAVX] in
4830 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4831 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4832 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4833 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4835 //===---------------------------------------------------------------------===//
4836 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4837 //===---------------------------------------------------------------------===//
4838 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4839 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4840 X86MemOperand x86memop> {
4841 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4842 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4843 [(set RC:$dst, (vt (OpNode RC:$src)))],
4845 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4846 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4847 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4851 let Predicates = [HasAVX] in {
4852 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4853 v4f32, VR128, memopv4f32, f128mem>, VEX;
4854 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4855 v4f32, VR128, memopv4f32, f128mem>, VEX;
4856 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4857 v8f32, VR256, memopv8f32, f256mem>, VEX;
4858 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4859 v8f32, VR256, memopv8f32, f256mem>, VEX;
4861 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4862 memopv4f32, f128mem>;
4863 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4864 memopv4f32, f128mem>;
4866 let Predicates = [HasAVX] in {
4867 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4868 (VMOVSHDUPrr VR128:$src)>;
4869 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4870 (VMOVSHDUPrm addr:$src)>;
4871 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4872 (VMOVSLDUPrr VR128:$src)>;
4873 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4874 (VMOVSLDUPrm addr:$src)>;
4875 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4876 (VMOVSHDUPYrr VR256:$src)>;
4877 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4878 (VMOVSHDUPYrm addr:$src)>;
4879 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4880 (VMOVSLDUPYrr VR256:$src)>;
4881 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4882 (VMOVSLDUPYrm addr:$src)>;
4885 let Predicates = [HasSSE3] in {
4886 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4887 (MOVSHDUPrr VR128:$src)>;
4888 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4889 (MOVSHDUPrm addr:$src)>;
4890 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4891 (MOVSLDUPrr VR128:$src)>;
4892 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4893 (MOVSLDUPrm addr:$src)>;
4896 //===---------------------------------------------------------------------===//
4897 // SSE3 - Replicate Double FP - MOVDDUP
4898 //===---------------------------------------------------------------------===//
4900 multiclass sse3_replicate_dfp<string OpcodeStr> {
4901 let neverHasSideEffects = 1 in
4902 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4903 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4904 [], IIC_SSE_MOV_LH>;
4905 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4906 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4909 (scalar_to_vector (loadf64 addr:$src)))))],
4913 // FIXME: Merge with above classe when there're patterns for the ymm version
4914 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4915 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4916 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4917 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4918 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4919 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4922 (scalar_to_vector (loadf64 addr:$src)))))]>;
4925 let Predicates = [HasAVX] in {
4926 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4927 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4930 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4932 let Predicates = [HasAVX] in {
4933 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4934 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4935 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4936 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4937 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4938 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4939 def : Pat<(X86Movddup (bc_v2f64
4940 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4941 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4944 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4945 (VMOVDDUPYrm addr:$src)>;
4946 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4947 (VMOVDDUPYrm addr:$src)>;
4948 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4949 (VMOVDDUPYrm addr:$src)>;
4950 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4951 (VMOVDDUPYrr VR256:$src)>;
4954 let Predicates = [HasSSE3] in {
4955 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4956 (MOVDDUPrm addr:$src)>;
4957 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4958 (MOVDDUPrm addr:$src)>;
4959 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4960 (MOVDDUPrm addr:$src)>;
4961 def : Pat<(X86Movddup (bc_v2f64
4962 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4963 (MOVDDUPrm addr:$src)>;
4966 //===---------------------------------------------------------------------===//
4967 // SSE3 - Move Unaligned Integer
4968 //===---------------------------------------------------------------------===//
4970 let Predicates = [HasAVX] in {
4971 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4972 "vlddqu\t{$src, $dst|$dst, $src}",
4973 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4974 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4975 "vlddqu\t{$src, $dst|$dst, $src}",
4976 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4978 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4979 "lddqu\t{$src, $dst|$dst, $src}",
4980 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4983 //===---------------------------------------------------------------------===//
4984 // SSE3 - Arithmetic
4985 //===---------------------------------------------------------------------===//
4987 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4988 X86MemOperand x86memop, OpndItins itins,
4990 def rr : I<0xD0, MRMSrcReg,
4991 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4993 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4994 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4995 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4996 def rm : I<0xD0, MRMSrcMem,
4997 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4999 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5000 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5001 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5004 let Predicates = [HasAVX] in {
5005 let ExeDomain = SSEPackedSingle in {
5006 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5007 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5008 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5009 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5011 let ExeDomain = SSEPackedDouble in {
5012 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5013 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5014 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5015 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5018 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5019 let ExeDomain = SSEPackedSingle in
5020 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5021 f128mem, SSE_ALU_F32P>, TB, XD;
5022 let ExeDomain = SSEPackedDouble in
5023 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5024 f128mem, SSE_ALU_F64P>, TB, OpSize;
5027 //===---------------------------------------------------------------------===//
5028 // SSE3 Instructions
5029 //===---------------------------------------------------------------------===//
5032 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5033 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5034 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5036 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5037 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5038 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5040 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5042 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5043 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5044 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5045 IIC_SSE_HADDSUB_RM>;
5047 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5048 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5049 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5051 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5052 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5053 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5055 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5057 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5058 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5059 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5060 IIC_SSE_HADDSUB_RM>;
5063 let Predicates = [HasAVX] in {
5064 let ExeDomain = SSEPackedSingle in {
5065 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5066 X86fhadd, 0>, VEX_4V;
5067 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5068 X86fhsub, 0>, VEX_4V;
5069 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5070 X86fhadd, 0>, VEX_4V;
5071 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5072 X86fhsub, 0>, VEX_4V;
5074 let ExeDomain = SSEPackedDouble in {
5075 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5076 X86fhadd, 0>, VEX_4V;
5077 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5078 X86fhsub, 0>, VEX_4V;
5079 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5080 X86fhadd, 0>, VEX_4V;
5081 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5082 X86fhsub, 0>, VEX_4V;
5086 let Constraints = "$src1 = $dst" in {
5087 let ExeDomain = SSEPackedSingle in {
5088 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5089 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5091 let ExeDomain = SSEPackedDouble in {
5092 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5093 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5097 //===---------------------------------------------------------------------===//
5098 // SSSE3 - Packed Absolute Instructions
5099 //===---------------------------------------------------------------------===//
5102 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5103 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5104 Intrinsic IntId128> {
5105 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5107 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5108 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5111 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5113 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5116 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5120 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5121 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5122 Intrinsic IntId256> {
5123 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5125 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5126 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5129 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5131 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5134 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5137 let Predicates = [HasAVX] in {
5138 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5139 int_x86_ssse3_pabs_b_128>, VEX;
5140 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5141 int_x86_ssse3_pabs_w_128>, VEX;
5142 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5143 int_x86_ssse3_pabs_d_128>, VEX;
5146 let Predicates = [HasAVX2] in {
5147 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5148 int_x86_avx2_pabs_b>, VEX;
5149 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5150 int_x86_avx2_pabs_w>, VEX;
5151 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5152 int_x86_avx2_pabs_d>, VEX;
5155 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5156 int_x86_ssse3_pabs_b_128>;
5157 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5158 int_x86_ssse3_pabs_w_128>;
5159 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5160 int_x86_ssse3_pabs_d_128>;
5162 //===---------------------------------------------------------------------===//
5163 // SSSE3 - Packed Binary Operator Instructions
5164 //===---------------------------------------------------------------------===//
5166 def SSE_PHADDSUBD : OpndItins<
5167 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5169 def SSE_PHADDSUBSW : OpndItins<
5170 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5172 def SSE_PHADDSUBW : OpndItins<
5173 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5175 def SSE_PSHUFB : OpndItins<
5176 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5178 def SSE_PSIGN : OpndItins<
5179 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5181 def SSE_PMULHRSW : OpndItins<
5182 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5185 /// SS3I_binop_rm - Simple SSSE3 bin op
5186 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5187 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5188 X86MemOperand x86memop, OpndItins itins,
5190 let isCommutable = 1 in
5191 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5192 (ins RC:$src1, RC:$src2),
5194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5196 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5198 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5199 (ins RC:$src1, x86memop:$src2),
5201 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5202 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5204 (OpVT (OpNode RC:$src1,
5205 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5208 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5209 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5210 Intrinsic IntId128, OpndItins itins,
5212 let isCommutable = 1 in
5213 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5214 (ins VR128:$src1, VR128:$src2),
5216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5218 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5220 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5221 (ins VR128:$src1, i128mem:$src2),
5223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5226 (IntId128 VR128:$src1,
5227 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5230 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5231 Intrinsic IntId256> {
5232 let isCommutable = 1 in
5233 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5234 (ins VR256:$src1, VR256:$src2),
5235 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5236 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5238 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5239 (ins VR256:$src1, i256mem:$src2),
5240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5242 (IntId256 VR256:$src1,
5243 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5246 let ImmT = NoImm, Predicates = [HasAVX] in {
5247 let isCommutable = 0 in {
5248 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5249 memopv2i64, i128mem,
5250 SSE_PHADDSUBW, 0>, VEX_4V;
5251 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5252 memopv2i64, i128mem,
5253 SSE_PHADDSUBD, 0>, VEX_4V;
5254 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5255 memopv2i64, i128mem,
5256 SSE_PHADDSUBW, 0>, VEX_4V;
5257 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5258 memopv2i64, i128mem,
5259 SSE_PHADDSUBD, 0>, VEX_4V;
5260 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5261 memopv2i64, i128mem,
5262 SSE_PSIGN, 0>, VEX_4V;
5263 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5264 memopv2i64, i128mem,
5265 SSE_PSIGN, 0>, VEX_4V;
5266 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5267 memopv2i64, i128mem,
5268 SSE_PSIGN, 0>, VEX_4V;
5269 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5270 memopv2i64, i128mem,
5271 SSE_PSHUFB, 0>, VEX_4V;
5272 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5273 int_x86_ssse3_phadd_sw_128,
5274 SSE_PHADDSUBSW, 0>, VEX_4V;
5275 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5276 int_x86_ssse3_phsub_sw_128,
5277 SSE_PHADDSUBSW, 0>, VEX_4V;
5278 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5279 int_x86_ssse3_pmadd_ub_sw_128,
5280 SSE_PMADD, 0>, VEX_4V;
5282 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5283 int_x86_ssse3_pmul_hr_sw_128,
5284 SSE_PMULHRSW, 0>, VEX_4V;
5287 let ImmT = NoImm, Predicates = [HasAVX2] in {
5288 let isCommutable = 0 in {
5289 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5290 memopv4i64, i256mem,
5291 SSE_PHADDSUBW, 0>, VEX_4V;
5292 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5293 memopv4i64, i256mem,
5294 SSE_PHADDSUBW, 0>, VEX_4V;
5295 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5296 memopv4i64, i256mem,
5297 SSE_PHADDSUBW, 0>, VEX_4V;
5298 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5299 memopv4i64, i256mem,
5300 SSE_PHADDSUBW, 0>, VEX_4V;
5301 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5302 memopv4i64, i256mem,
5303 SSE_PHADDSUBW, 0>, VEX_4V;
5304 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5305 memopv4i64, i256mem,
5306 SSE_PHADDSUBW, 0>, VEX_4V;
5307 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5308 memopv4i64, i256mem,
5309 SSE_PHADDSUBW, 0>, VEX_4V;
5310 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5311 memopv4i64, i256mem,
5312 SSE_PHADDSUBW, 0>, VEX_4V;
5313 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5314 int_x86_avx2_phadd_sw>, VEX_4V;
5315 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5316 int_x86_avx2_phsub_sw>, VEX_4V;
5317 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5318 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5320 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5321 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5324 // None of these have i8 immediate fields.
5325 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5326 let isCommutable = 0 in {
5327 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5328 memopv2i64, i128mem, SSE_PHADDSUBW>;
5329 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5330 memopv2i64, i128mem, SSE_PHADDSUBD>;
5331 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5332 memopv2i64, i128mem, SSE_PHADDSUBW>;
5333 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5334 memopv2i64, i128mem, SSE_PHADDSUBD>;
5335 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5336 memopv2i64, i128mem, SSE_PSIGN>;
5337 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5338 memopv2i64, i128mem, SSE_PSIGN>;
5339 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5340 memopv2i64, i128mem, SSE_PSIGN>;
5341 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5342 memopv2i64, i128mem, SSE_PSHUFB>;
5343 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5344 int_x86_ssse3_phadd_sw_128,
5346 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5347 int_x86_ssse3_phsub_sw_128,
5349 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5350 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5352 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5353 int_x86_ssse3_pmul_hr_sw_128,
5357 //===---------------------------------------------------------------------===//
5358 // SSSE3 - Packed Align Instruction Patterns
5359 //===---------------------------------------------------------------------===//
5361 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5362 let neverHasSideEffects = 1 in {
5363 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5364 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5366 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5368 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5369 [], IIC_SSE_PALIGNR>, OpSize;
5371 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5372 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5374 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5376 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5377 [], IIC_SSE_PALIGNR>, OpSize;
5381 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5382 let neverHasSideEffects = 1 in {
5383 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5384 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5386 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5389 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5390 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5392 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5397 let Predicates = [HasAVX] in
5398 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5399 let Predicates = [HasAVX2] in
5400 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5401 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5402 defm PALIGN : ssse3_palign<"palignr">;
5404 let Predicates = [HasAVX2] in {
5405 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5406 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5407 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5408 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5409 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5410 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5411 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5412 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5415 let Predicates = [HasAVX] in {
5416 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5417 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5418 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5419 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5420 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5421 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5422 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5423 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5426 let Predicates = [HasSSSE3] in {
5427 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5428 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5429 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5430 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5431 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5432 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5433 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5434 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5437 //===---------------------------------------------------------------------===//
5438 // SSSE3 - Thread synchronization
5439 //===---------------------------------------------------------------------===//
5441 let usesCustomInserter = 1 in {
5442 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5443 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5444 Requires<[HasSSE3]>;
5445 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5446 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5447 Requires<[HasSSE3]>;
5450 let Uses = [EAX, ECX, EDX] in
5451 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5452 TB, Requires<[HasSSE3]>;
5453 let Uses = [ECX, EAX] in
5454 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5455 TB, Requires<[HasSSE3]>;
5457 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5458 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5460 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5461 Requires<[In32BitMode]>;
5462 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5463 Requires<[In64BitMode]>;
5465 //===----------------------------------------------------------------------===//
5466 // SSE4.1 - Packed Move with Sign/Zero Extend
5467 //===----------------------------------------------------------------------===//
5469 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5470 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5471 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5472 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5474 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5475 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5477 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5481 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5483 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5484 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5485 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5487 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5488 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5489 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5492 let Predicates = [HasAVX] in {
5493 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5495 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5497 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5499 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5501 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5503 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5507 let Predicates = [HasAVX2] in {
5508 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5509 int_x86_avx2_pmovsxbw>, VEX;
5510 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5511 int_x86_avx2_pmovsxwd>, VEX;
5512 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5513 int_x86_avx2_pmovsxdq>, VEX;
5514 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5515 int_x86_avx2_pmovzxbw>, VEX;
5516 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5517 int_x86_avx2_pmovzxwd>, VEX;
5518 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5519 int_x86_avx2_pmovzxdq>, VEX;
5522 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5523 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5524 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5525 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5526 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5527 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5529 let Predicates = [HasAVX] in {
5530 // Common patterns involving scalar load.
5531 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5532 (VPMOVSXBWrm addr:$src)>;
5533 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5534 (VPMOVSXBWrm addr:$src)>;
5536 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5537 (VPMOVSXWDrm addr:$src)>;
5538 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5539 (VPMOVSXWDrm addr:$src)>;
5541 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5542 (VPMOVSXDQrm addr:$src)>;
5543 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5544 (VPMOVSXDQrm addr:$src)>;
5546 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5547 (VPMOVZXBWrm addr:$src)>;
5548 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5549 (VPMOVZXBWrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5552 (VPMOVZXWDrm addr:$src)>;
5553 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5554 (VPMOVZXWDrm addr:$src)>;
5556 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5557 (VPMOVZXDQrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5559 (VPMOVZXDQrm addr:$src)>;
5562 let Predicates = [HasSSE41] in {
5563 // Common patterns involving scalar load.
5564 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5565 (PMOVSXBWrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5567 (PMOVSXBWrm addr:$src)>;
5569 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5570 (PMOVSXWDrm addr:$src)>;
5571 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5572 (PMOVSXWDrm addr:$src)>;
5574 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5575 (PMOVSXDQrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5577 (PMOVSXDQrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5580 (PMOVZXBWrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5582 (PMOVZXBWrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5585 (PMOVZXWDrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5587 (PMOVZXWDrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5590 (PMOVZXDQrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5592 (PMOVZXDQrm addr:$src)>;
5595 let Predicates = [HasAVX2] in {
5596 let AddedComplexity = 15 in {
5597 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5598 (VPMOVZXDQYrr VR128:$src)>;
5599 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5600 (VPMOVZXWDYrr VR128:$src)>;
5603 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5604 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5607 let Predicates = [HasAVX] in {
5608 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5609 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5612 let Predicates = [HasSSE41] in {
5613 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5614 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5618 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5619 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5620 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5621 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5623 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5624 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5626 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5630 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5632 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5633 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5634 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5636 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5637 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5639 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5643 let Predicates = [HasAVX] in {
5644 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5646 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5648 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5650 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5654 let Predicates = [HasAVX2] in {
5655 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5656 int_x86_avx2_pmovsxbd>, VEX;
5657 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5658 int_x86_avx2_pmovsxwq>, VEX;
5659 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5660 int_x86_avx2_pmovzxbd>, VEX;
5661 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5662 int_x86_avx2_pmovzxwq>, VEX;
5665 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5666 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5667 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5668 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5670 let Predicates = [HasAVX] in {
5671 // Common patterns involving scalar load
5672 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5673 (VPMOVSXBDrm addr:$src)>;
5674 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5675 (VPMOVSXWQrm addr:$src)>;
5677 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5678 (VPMOVZXBDrm addr:$src)>;
5679 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5680 (VPMOVZXWQrm addr:$src)>;
5683 let Predicates = [HasSSE41] in {
5684 // Common patterns involving scalar load
5685 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5686 (PMOVSXBDrm addr:$src)>;
5687 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5688 (PMOVSXWQrm addr:$src)>;
5690 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5691 (PMOVZXBDrm addr:$src)>;
5692 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5693 (PMOVZXWQrm addr:$src)>;
5696 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5697 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5698 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5699 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5701 // Expecting a i16 load any extended to i32 value.
5702 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5703 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5704 [(set VR128:$dst, (IntId (bitconvert
5705 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5709 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5711 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5712 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5713 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5715 // Expecting a i16 load any extended to i32 value.
5716 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5717 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5718 [(set VR256:$dst, (IntId (bitconvert
5719 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5723 let Predicates = [HasAVX] in {
5724 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5726 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5729 let Predicates = [HasAVX2] in {
5730 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5731 int_x86_avx2_pmovsxbq>, VEX;
5732 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5733 int_x86_avx2_pmovzxbq>, VEX;
5735 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5736 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5738 let Predicates = [HasAVX] in {
5739 // Common patterns involving scalar load
5740 def : Pat<(int_x86_sse41_pmovsxbq
5741 (bitconvert (v4i32 (X86vzmovl
5742 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5743 (VPMOVSXBQrm addr:$src)>;
5745 def : Pat<(int_x86_sse41_pmovzxbq
5746 (bitconvert (v4i32 (X86vzmovl
5747 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5748 (VPMOVZXBQrm addr:$src)>;
5751 let Predicates = [HasSSE41] in {
5752 // Common patterns involving scalar load
5753 def : Pat<(int_x86_sse41_pmovsxbq
5754 (bitconvert (v4i32 (X86vzmovl
5755 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5756 (PMOVSXBQrm addr:$src)>;
5758 def : Pat<(int_x86_sse41_pmovzxbq
5759 (bitconvert (v4i32 (X86vzmovl
5760 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5761 (PMOVZXBQrm addr:$src)>;
5764 //===----------------------------------------------------------------------===//
5765 // SSE4.1 - Extract Instructions
5766 //===----------------------------------------------------------------------===//
5768 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5769 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5770 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5771 (ins VR128:$src1, i32i8imm:$src2),
5772 !strconcat(OpcodeStr,
5773 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5774 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5776 let neverHasSideEffects = 1, mayStore = 1 in
5777 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5778 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5779 !strconcat(OpcodeStr,
5780 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5783 // There's an AssertZext in the way of writing the store pattern
5784 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5787 let Predicates = [HasAVX] in {
5788 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5789 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5790 (ins VR128:$src1, i32i8imm:$src2),
5791 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5794 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5797 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5798 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5799 let neverHasSideEffects = 1, mayStore = 1 in
5800 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5801 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5802 !strconcat(OpcodeStr,
5803 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5806 // There's an AssertZext in the way of writing the store pattern
5807 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5810 let Predicates = [HasAVX] in
5811 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5813 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5816 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5817 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5818 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5819 (ins VR128:$src1, i32i8imm:$src2),
5820 !strconcat(OpcodeStr,
5821 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5823 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5824 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5825 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5826 !strconcat(OpcodeStr,
5827 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5828 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5829 addr:$dst)]>, OpSize;
5832 let Predicates = [HasAVX] in
5833 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5835 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5837 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5838 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5839 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5840 (ins VR128:$src1, i32i8imm:$src2),
5841 !strconcat(OpcodeStr,
5842 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5844 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5845 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5846 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5847 !strconcat(OpcodeStr,
5848 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5849 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5850 addr:$dst)]>, OpSize, REX_W;
5853 let Predicates = [HasAVX] in
5854 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5856 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5858 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5860 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5861 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5862 (ins VR128:$src1, i32i8imm:$src2),
5863 !strconcat(OpcodeStr,
5864 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5866 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5868 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5869 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5870 !strconcat(OpcodeStr,
5871 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5872 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5873 addr:$dst)]>, OpSize;
5876 let ExeDomain = SSEPackedSingle in {
5877 let Predicates = [HasAVX] in {
5878 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5879 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5880 (ins VR128:$src1, i32i8imm:$src2),
5881 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5884 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5887 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5888 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5891 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5893 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5896 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5897 Requires<[HasSSE41]>;
5899 //===----------------------------------------------------------------------===//
5900 // SSE4.1 - Insert Instructions
5901 //===----------------------------------------------------------------------===//
5903 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5904 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5905 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5907 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5909 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5911 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5912 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5913 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5915 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5917 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5919 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5920 imm:$src3))]>, OpSize;
5923 let Predicates = [HasAVX] in
5924 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5925 let Constraints = "$src1 = $dst" in
5926 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5928 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5929 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5930 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5932 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5934 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5936 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5938 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5939 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5941 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5943 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5945 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5946 imm:$src3)))]>, OpSize;
5949 let Predicates = [HasAVX] in
5950 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5951 let Constraints = "$src1 = $dst" in
5952 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5954 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5955 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5956 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5958 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5960 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5962 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5964 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5965 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5967 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5969 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5971 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5972 imm:$src3)))]>, OpSize;
5975 let Predicates = [HasAVX] in
5976 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5977 let Constraints = "$src1 = $dst" in
5978 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5980 // insertps has a few different modes, there's the first two here below which
5981 // are optimized inserts that won't zero arbitrary elements in the destination
5982 // vector. The next one matches the intrinsic and could zero arbitrary elements
5983 // in the target vector.
5984 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5985 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5986 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5988 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5990 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5992 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5994 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5995 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5997 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5999 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6001 (X86insrtps VR128:$src1,
6002 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6003 imm:$src3))]>, OpSize;
6006 let ExeDomain = SSEPackedSingle in {
6007 let Predicates = [HasAVX] in
6008 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6009 let Constraints = "$src1 = $dst" in
6010 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6013 //===----------------------------------------------------------------------===//
6014 // SSE4.1 - Round Instructions
6015 //===----------------------------------------------------------------------===//
6017 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6018 X86MemOperand x86memop, RegisterClass RC,
6019 PatFrag mem_frag32, PatFrag mem_frag64,
6020 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6021 let ExeDomain = SSEPackedSingle in {
6022 // Intrinsic operation, reg.
6023 // Vector intrinsic operation, reg
6024 def PSr : SS4AIi8<opcps, MRMSrcReg,
6025 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6026 !strconcat(OpcodeStr,
6027 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6028 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6031 // Vector intrinsic operation, mem
6032 def PSm : SS4AIi8<opcps, MRMSrcMem,
6033 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6034 !strconcat(OpcodeStr,
6035 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6037 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6039 } // ExeDomain = SSEPackedSingle
6041 let ExeDomain = SSEPackedDouble in {
6042 // Vector intrinsic operation, reg
6043 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6044 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6045 !strconcat(OpcodeStr,
6046 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6047 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6050 // Vector intrinsic operation, mem
6051 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6052 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6053 !strconcat(OpcodeStr,
6054 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6056 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6058 } // ExeDomain = SSEPackedDouble
6061 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6064 Intrinsic F64Int, bit Is2Addr = 1> {
6065 let ExeDomain = GenericDomain in {
6067 def SSr : SS4AIi8<opcss, MRMSrcReg,
6068 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6070 !strconcat(OpcodeStr,
6071 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6072 !strconcat(OpcodeStr,
6073 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6076 // Intrinsic operation, reg.
6077 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6078 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6080 !strconcat(OpcodeStr,
6081 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6082 !strconcat(OpcodeStr,
6083 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6084 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6087 // Intrinsic operation, mem.
6088 def SSm : SS4AIi8<opcss, MRMSrcMem,
6089 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6091 !strconcat(OpcodeStr,
6092 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6093 !strconcat(OpcodeStr,
6094 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6096 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6100 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6101 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6103 !strconcat(OpcodeStr,
6104 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6105 !strconcat(OpcodeStr,
6106 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6109 // Intrinsic operation, reg.
6110 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6111 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6113 !strconcat(OpcodeStr,
6114 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6115 !strconcat(OpcodeStr,
6116 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6117 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6120 // Intrinsic operation, mem.
6121 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6122 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6124 !strconcat(OpcodeStr,
6125 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6126 !strconcat(OpcodeStr,
6127 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6129 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6131 } // ExeDomain = GenericDomain
6134 // FP round - roundss, roundps, roundsd, roundpd
6135 let Predicates = [HasAVX] in {
6137 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6138 memopv4f32, memopv2f64,
6139 int_x86_sse41_round_ps,
6140 int_x86_sse41_round_pd>, VEX;
6141 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6142 memopv8f32, memopv4f64,
6143 int_x86_avx_round_ps_256,
6144 int_x86_avx_round_pd_256>, VEX;
6145 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6146 int_x86_sse41_round_ss,
6147 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6149 def : Pat<(ffloor FR32:$src),
6150 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6151 def : Pat<(f64 (ffloor FR64:$src)),
6152 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6153 def : Pat<(f32 (fnearbyint FR32:$src)),
6154 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6155 def : Pat<(f64 (fnearbyint FR64:$src)),
6156 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6157 def : Pat<(f32 (fceil FR32:$src)),
6158 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6159 def : Pat<(f64 (fceil FR64:$src)),
6160 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6161 def : Pat<(f32 (frint FR32:$src)),
6162 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6163 def : Pat<(f64 (frint FR64:$src)),
6164 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6165 def : Pat<(f32 (ftrunc FR32:$src)),
6166 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6167 def : Pat<(f64 (ftrunc FR64:$src)),
6168 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6171 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6172 memopv4f32, memopv2f64,
6173 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6174 let Constraints = "$src1 = $dst" in
6175 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6176 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6178 def : Pat<(ffloor FR32:$src),
6179 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6180 def : Pat<(f64 (ffloor FR64:$src)),
6181 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6182 def : Pat<(f32 (fnearbyint FR32:$src)),
6183 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6184 def : Pat<(f64 (fnearbyint FR64:$src)),
6185 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6186 def : Pat<(f32 (fceil FR32:$src)),
6187 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6188 def : Pat<(f64 (fceil FR64:$src)),
6189 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6190 def : Pat<(f32 (frint FR32:$src)),
6191 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6192 def : Pat<(f64 (frint FR64:$src)),
6193 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6194 def : Pat<(f32 (ftrunc FR32:$src)),
6195 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6196 def : Pat<(f64 (ftrunc FR64:$src)),
6197 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6199 //===----------------------------------------------------------------------===//
6200 // SSE4.1 - Packed Bit Test
6201 //===----------------------------------------------------------------------===//
6203 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6204 // the intel intrinsic that corresponds to this.
6205 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6206 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6207 "vptest\t{$src2, $src1|$src1, $src2}",
6208 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6210 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6211 "vptest\t{$src2, $src1|$src1, $src2}",
6212 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6215 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6216 "vptest\t{$src2, $src1|$src1, $src2}",
6217 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6219 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6220 "vptest\t{$src2, $src1|$src1, $src2}",
6221 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6225 let Defs = [EFLAGS] in {
6226 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6227 "ptest\t{$src2, $src1|$src1, $src2}",
6228 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6230 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6231 "ptest\t{$src2, $src1|$src1, $src2}",
6232 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6236 // The bit test instructions below are AVX only
6237 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6238 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6239 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6240 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6241 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6242 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6243 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6244 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6248 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6249 let ExeDomain = SSEPackedSingle in {
6250 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6251 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6253 let ExeDomain = SSEPackedDouble in {
6254 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6255 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6259 //===----------------------------------------------------------------------===//
6260 // SSE4.1 - Misc Instructions
6261 //===----------------------------------------------------------------------===//
6263 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6264 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6265 "popcnt{w}\t{$src, $dst|$dst, $src}",
6266 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6268 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6269 "popcnt{w}\t{$src, $dst|$dst, $src}",
6270 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6271 (implicit EFLAGS)]>, OpSize, XS;
6273 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6274 "popcnt{l}\t{$src, $dst|$dst, $src}",
6275 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6277 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6278 "popcnt{l}\t{$src, $dst|$dst, $src}",
6279 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6280 (implicit EFLAGS)]>, XS;
6282 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6283 "popcnt{q}\t{$src, $dst|$dst, $src}",
6284 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6286 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6287 "popcnt{q}\t{$src, $dst|$dst, $src}",
6288 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6289 (implicit EFLAGS)]>, XS;
6294 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6295 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6296 Intrinsic IntId128> {
6297 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6299 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6300 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6301 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6303 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6306 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6309 let Predicates = [HasAVX] in
6310 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6311 int_x86_sse41_phminposuw>, VEX;
6312 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6313 int_x86_sse41_phminposuw>;
6315 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6316 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6317 Intrinsic IntId128, bit Is2Addr = 1> {
6318 let isCommutable = 1 in
6319 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6320 (ins VR128:$src1, VR128:$src2),
6322 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6323 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6324 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6325 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6326 (ins VR128:$src1, i128mem:$src2),
6328 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6331 (IntId128 VR128:$src1,
6332 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6335 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6336 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6337 Intrinsic IntId256> {
6338 let isCommutable = 1 in
6339 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6340 (ins VR256:$src1, VR256:$src2),
6341 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6342 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6343 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6344 (ins VR256:$src1, i256mem:$src2),
6345 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6347 (IntId256 VR256:$src1,
6348 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6351 let Predicates = [HasAVX] in {
6352 let isCommutable = 0 in
6353 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6355 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6357 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6359 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6361 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6363 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6365 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6367 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6369 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6371 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6375 let Predicates = [HasAVX2] in {
6376 let isCommutable = 0 in
6377 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6378 int_x86_avx2_packusdw>, VEX_4V;
6379 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6380 int_x86_avx2_pmins_b>, VEX_4V;
6381 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6382 int_x86_avx2_pmins_d>, VEX_4V;
6383 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6384 int_x86_avx2_pminu_d>, VEX_4V;
6385 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6386 int_x86_avx2_pminu_w>, VEX_4V;
6387 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6388 int_x86_avx2_pmaxs_b>, VEX_4V;
6389 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6390 int_x86_avx2_pmaxs_d>, VEX_4V;
6391 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6392 int_x86_avx2_pmaxu_d>, VEX_4V;
6393 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6394 int_x86_avx2_pmaxu_w>, VEX_4V;
6395 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6396 int_x86_avx2_pmul_dq>, VEX_4V;
6399 let Constraints = "$src1 = $dst" in {
6400 let isCommutable = 0 in
6401 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6402 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6403 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6404 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6405 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6406 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6407 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6408 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6409 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6410 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6413 /// SS48I_binop_rm - Simple SSE41 binary operator.
6414 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6415 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6416 X86MemOperand x86memop, bit Is2Addr = 1> {
6417 let isCommutable = 1 in
6418 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6419 (ins RC:$src1, RC:$src2),
6421 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6422 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6423 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6424 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6425 (ins RC:$src1, x86memop:$src2),
6427 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6428 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6430 (OpVT (OpNode RC:$src1,
6431 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6434 let Predicates = [HasAVX] in {
6435 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6436 memopv2i64, i128mem, 0>, VEX_4V;
6437 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6438 memopv2i64, i128mem, 0>, VEX_4V;
6440 let Predicates = [HasAVX2] in {
6441 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6442 memopv4i64, i256mem, 0>, VEX_4V;
6443 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6444 memopv4i64, i256mem, 0>, VEX_4V;
6447 let Constraints = "$src1 = $dst" in {
6448 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6449 memopv2i64, i128mem>;
6450 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6451 memopv2i64, i128mem>;
6454 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6455 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6456 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6457 X86MemOperand x86memop, bit Is2Addr = 1> {
6458 let isCommutable = 1 in
6459 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6460 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6462 !strconcat(OpcodeStr,
6463 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6464 !strconcat(OpcodeStr,
6465 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6466 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6468 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6469 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6471 !strconcat(OpcodeStr,
6472 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6473 !strconcat(OpcodeStr,
6474 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6477 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6481 let Predicates = [HasAVX] in {
6482 let isCommutable = 0 in {
6483 let ExeDomain = SSEPackedSingle in {
6484 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6485 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6486 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6487 int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
6489 let ExeDomain = SSEPackedDouble in {
6490 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6491 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6492 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6493 int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
6495 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6496 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6497 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6498 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6500 let ExeDomain = SSEPackedSingle in
6501 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6502 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6503 let ExeDomain = SSEPackedDouble in
6504 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6505 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6506 let ExeDomain = SSEPackedSingle in
6507 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6508 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6511 let Predicates = [HasAVX2] in {
6512 let isCommutable = 0 in {
6513 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6514 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6515 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6516 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6520 let Constraints = "$src1 = $dst" in {
6521 let isCommutable = 0 in {
6522 let ExeDomain = SSEPackedSingle in
6523 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6524 VR128, memopv4f32, f128mem>;
6525 let ExeDomain = SSEPackedDouble in
6526 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6527 VR128, memopv2f64, f128mem>;
6528 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6529 VR128, memopv2i64, i128mem>;
6530 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6531 VR128, memopv2i64, i128mem>;
6533 let ExeDomain = SSEPackedSingle in
6534 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6535 VR128, memopv4f32, f128mem>;
6536 let ExeDomain = SSEPackedDouble in
6537 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6538 VR128, memopv2f64, f128mem>;
6541 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6542 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6543 RegisterClass RC, X86MemOperand x86memop,
6544 PatFrag mem_frag, Intrinsic IntId> {
6545 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6546 (ins RC:$src1, RC:$src2, RC:$src3),
6547 !strconcat(OpcodeStr,
6548 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6549 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6550 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6552 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6553 (ins RC:$src1, x86memop:$src2, RC:$src3),
6554 !strconcat(OpcodeStr,
6555 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6557 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6559 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6562 let Predicates = [HasAVX] in {
6563 let ExeDomain = SSEPackedDouble in {
6564 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6565 memopv2f64, int_x86_sse41_blendvpd>;
6566 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6567 memopv4f64, int_x86_avx_blendv_pd_256>;
6568 } // ExeDomain = SSEPackedDouble
6569 let ExeDomain = SSEPackedSingle in {
6570 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6571 memopv4f32, int_x86_sse41_blendvps>;
6572 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6573 memopv8f32, int_x86_avx_blendv_ps_256>;
6574 } // ExeDomain = SSEPackedSingle
6575 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6576 memopv2i64, int_x86_sse41_pblendvb>;
6579 let Predicates = [HasAVX2] in {
6580 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6581 memopv4i64, int_x86_avx2_pblendvb>;
6584 let Predicates = [HasAVX] in {
6585 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6586 (v16i8 VR128:$src2))),
6587 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6588 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6589 (v4i32 VR128:$src2))),
6590 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6591 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6592 (v4f32 VR128:$src2))),
6593 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6594 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6595 (v2i64 VR128:$src2))),
6596 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6597 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6598 (v2f64 VR128:$src2))),
6599 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6600 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6601 (v8i32 VR256:$src2))),
6602 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6603 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6604 (v8f32 VR256:$src2))),
6605 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6606 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6607 (v4i64 VR256:$src2))),
6608 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6609 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6610 (v4f64 VR256:$src2))),
6611 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6613 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6615 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6616 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6618 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6620 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6622 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6623 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6625 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6626 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6628 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6631 let Predicates = [HasAVX2] in {
6632 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6633 (v32i8 VR256:$src2))),
6634 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6635 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6637 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6640 /// SS41I_ternary_int - SSE 4.1 ternary operator
6641 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6642 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6643 X86MemOperand x86memop, Intrinsic IntId> {
6644 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6645 (ins VR128:$src1, VR128:$src2),
6646 !strconcat(OpcodeStr,
6647 "\t{$src2, $dst|$dst, $src2}"),
6648 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6651 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6652 (ins VR128:$src1, x86memop:$src2),
6653 !strconcat(OpcodeStr,
6654 "\t{$src2, $dst|$dst, $src2}"),
6657 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6661 let ExeDomain = SSEPackedDouble in
6662 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6663 int_x86_sse41_blendvpd>;
6664 let ExeDomain = SSEPackedSingle in
6665 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6666 int_x86_sse41_blendvps>;
6667 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6668 int_x86_sse41_pblendvb>;
6670 // Aliases with the implicit xmm0 argument
6671 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6672 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6673 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6674 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6675 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6676 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6677 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6678 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6679 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6680 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6681 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6682 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6684 let Predicates = [HasSSE41] in {
6685 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6686 (v16i8 VR128:$src2))),
6687 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6688 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6689 (v4i32 VR128:$src2))),
6690 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6691 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6692 (v4f32 VR128:$src2))),
6693 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6694 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6695 (v2i64 VR128:$src2))),
6696 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6697 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6698 (v2f64 VR128:$src2))),
6699 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6701 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6703 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6704 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6706 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6707 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6709 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6713 let Predicates = [HasAVX] in
6714 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6715 "vmovntdqa\t{$src, $dst|$dst, $src}",
6716 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6718 let Predicates = [HasAVX2] in
6719 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6720 "vmovntdqa\t{$src, $dst|$dst, $src}",
6721 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6723 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6724 "movntdqa\t{$src, $dst|$dst, $src}",
6725 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6728 //===----------------------------------------------------------------------===//
6729 // SSE4.2 - Compare Instructions
6730 //===----------------------------------------------------------------------===//
6732 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6733 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6734 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6735 X86MemOperand x86memop, bit Is2Addr = 1> {
6736 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6737 (ins RC:$src1, RC:$src2),
6739 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6740 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6741 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6743 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6744 (ins RC:$src1, x86memop:$src2),
6746 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6747 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6749 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6752 let Predicates = [HasAVX] in
6753 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6754 memopv2i64, i128mem, 0>, VEX_4V;
6756 let Predicates = [HasAVX2] in
6757 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6758 memopv4i64, i256mem, 0>, VEX_4V;
6760 let Constraints = "$src1 = $dst" in
6761 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6762 memopv2i64, i128mem>;
6764 //===----------------------------------------------------------------------===//
6765 // SSE4.2 - String/text Processing Instructions
6766 //===----------------------------------------------------------------------===//
6768 // Packed Compare Implicit Length Strings, Return Mask
6769 multiclass pseudo_pcmpistrm<string asm> {
6770 def REG : PseudoI<(outs VR128:$dst),
6771 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6772 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6774 def MEM : PseudoI<(outs VR128:$dst),
6775 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6776 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6777 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6780 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6781 let AddedComplexity = 1 in
6782 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6783 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6786 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6787 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6788 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6789 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6791 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6792 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6793 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6796 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6797 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6798 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6799 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6801 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6802 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6803 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6806 // Packed Compare Explicit Length Strings, Return Mask
6807 multiclass pseudo_pcmpestrm<string asm> {
6808 def REG : PseudoI<(outs VR128:$dst),
6809 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6810 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6811 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6812 def MEM : PseudoI<(outs VR128:$dst),
6813 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6814 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6815 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6818 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6819 let AddedComplexity = 1 in
6820 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6821 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6824 let Predicates = [HasAVX],
6825 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6826 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6827 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6828 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6830 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6831 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6832 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6835 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6836 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6837 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6838 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6840 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6841 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6842 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6845 // Packed Compare Implicit Length Strings, Return Index
6846 let Defs = [ECX, EFLAGS] in {
6847 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6848 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6849 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6850 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6851 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6852 (implicit EFLAGS)]>, OpSize;
6853 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6854 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6855 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6856 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6857 (implicit EFLAGS)]>, OpSize;
6861 let Predicates = [HasAVX] in {
6862 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6864 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6866 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6868 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6870 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6872 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6876 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6877 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6878 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6879 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6880 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6881 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6883 // Packed Compare Explicit Length Strings, Return Index
6884 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6885 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6886 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6887 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6888 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6889 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6890 (implicit EFLAGS)]>, OpSize;
6891 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6892 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6893 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6895 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6896 (implicit EFLAGS)]>, OpSize;
6900 let Predicates = [HasAVX] in {
6901 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6903 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6905 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6907 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6909 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6911 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6915 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6916 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6917 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6918 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6919 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6920 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6922 //===----------------------------------------------------------------------===//
6923 // SSE4.2 - CRC Instructions
6924 //===----------------------------------------------------------------------===//
6926 // No CRC instructions have AVX equivalents
6928 // crc intrinsic instruction
6929 // This set of instructions are only rm, the only difference is the size
6931 let Constraints = "$src1 = $dst" in {
6932 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6933 (ins GR32:$src1, i8mem:$src2),
6934 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6936 (int_x86_sse42_crc32_32_8 GR32:$src1,
6937 (load addr:$src2)))]>;
6938 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6939 (ins GR32:$src1, GR8:$src2),
6940 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6942 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6943 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6944 (ins GR32:$src1, i16mem:$src2),
6945 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6947 (int_x86_sse42_crc32_32_16 GR32:$src1,
6948 (load addr:$src2)))]>,
6950 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6951 (ins GR32:$src1, GR16:$src2),
6952 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6954 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6956 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6957 (ins GR32:$src1, i32mem:$src2),
6958 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6960 (int_x86_sse42_crc32_32_32 GR32:$src1,
6961 (load addr:$src2)))]>;
6962 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6963 (ins GR32:$src1, GR32:$src2),
6964 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6966 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6967 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6968 (ins GR64:$src1, i8mem:$src2),
6969 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6971 (int_x86_sse42_crc32_64_8 GR64:$src1,
6972 (load addr:$src2)))]>,
6974 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6975 (ins GR64:$src1, GR8:$src2),
6976 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6978 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6980 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6981 (ins GR64:$src1, i64mem:$src2),
6982 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_64_64 GR64:$src1,
6985 (load addr:$src2)))]>,
6987 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6988 (ins GR64:$src1, GR64:$src2),
6989 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6991 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6995 //===----------------------------------------------------------------------===//
6996 // AES-NI Instructions
6997 //===----------------------------------------------------------------------===//
6999 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7000 Intrinsic IntId128, bit Is2Addr = 1> {
7001 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7002 (ins VR128:$src1, VR128:$src2),
7004 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7005 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7006 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7008 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7009 (ins VR128:$src1, i128mem:$src2),
7011 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7012 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7014 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7017 // Perform One Round of an AES Encryption/Decryption Flow
7018 let Predicates = [HasAVX, HasAES] in {
7019 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7020 int_x86_aesni_aesenc, 0>, VEX_4V;
7021 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7022 int_x86_aesni_aesenclast, 0>, VEX_4V;
7023 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7024 int_x86_aesni_aesdec, 0>, VEX_4V;
7025 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7026 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7029 let Constraints = "$src1 = $dst" in {
7030 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7031 int_x86_aesni_aesenc>;
7032 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7033 int_x86_aesni_aesenclast>;
7034 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7035 int_x86_aesni_aesdec>;
7036 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7037 int_x86_aesni_aesdeclast>;
7040 // Perform the AES InvMixColumn Transformation
7041 let Predicates = [HasAVX, HasAES] in {
7042 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7044 "vaesimc\t{$src1, $dst|$dst, $src1}",
7046 (int_x86_aesni_aesimc VR128:$src1))]>,
7048 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7049 (ins i128mem:$src1),
7050 "vaesimc\t{$src1, $dst|$dst, $src1}",
7051 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7054 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7056 "aesimc\t{$src1, $dst|$dst, $src1}",
7058 (int_x86_aesni_aesimc VR128:$src1))]>,
7060 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7061 (ins i128mem:$src1),
7062 "aesimc\t{$src1, $dst|$dst, $src1}",
7063 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7066 // AES Round Key Generation Assist
7067 let Predicates = [HasAVX, HasAES] in {
7068 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7069 (ins VR128:$src1, i8imm:$src2),
7070 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7072 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7074 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7075 (ins i128mem:$src1, i8imm:$src2),
7076 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7078 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7081 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7082 (ins VR128:$src1, i8imm:$src2),
7083 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7085 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7087 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7088 (ins i128mem:$src1, i8imm:$src2),
7089 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7091 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7094 //===----------------------------------------------------------------------===//
7095 // PCLMUL Instructions
7096 //===----------------------------------------------------------------------===//
7098 // AVX carry-less Multiplication instructions
7099 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7100 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7101 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7103 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7105 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7106 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7107 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7108 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7109 (memopv2i64 addr:$src2), imm:$src3))]>;
7111 // Carry-less Multiplication instructions
7112 let Constraints = "$src1 = $dst" in {
7113 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7114 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7115 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7117 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7119 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7120 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7121 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7122 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7123 (memopv2i64 addr:$src2), imm:$src3))]>;
7124 } // Constraints = "$src1 = $dst"
7127 multiclass pclmul_alias<string asm, int immop> {
7128 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7129 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7131 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7132 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7134 def : InstAlias<!strconcat("vpclmul", asm,
7135 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7136 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7138 def : InstAlias<!strconcat("vpclmul", asm,
7139 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7140 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7142 defm : pclmul_alias<"hqhq", 0x11>;
7143 defm : pclmul_alias<"hqlq", 0x01>;
7144 defm : pclmul_alias<"lqhq", 0x10>;
7145 defm : pclmul_alias<"lqlq", 0x00>;
7147 //===----------------------------------------------------------------------===//
7148 // SSE4A Instructions
7149 //===----------------------------------------------------------------------===//
7151 let Predicates = [HasSSE4A] in {
7153 let Constraints = "$src = $dst" in {
7154 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7155 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7156 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7157 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7158 imm:$idx))]>, TB, OpSize;
7159 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7160 (ins VR128:$src, VR128:$mask),
7161 "extrq\t{$mask, $src|$src, $mask}",
7162 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7163 VR128:$mask))]>, TB, OpSize;
7165 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7166 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7167 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7168 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7169 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7170 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7171 (ins VR128:$src, VR128:$mask),
7172 "insertq\t{$mask, $src|$src, $mask}",
7173 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7174 VR128:$mask))]>, XD;
7177 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7178 "movntss\t{$src, $dst|$dst, $src}",
7179 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7181 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7182 "movntsd\t{$src, $dst|$dst, $src}",
7183 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7186 //===----------------------------------------------------------------------===//
7188 //===----------------------------------------------------------------------===//
7190 //===----------------------------------------------------------------------===//
7191 // VBROADCAST - Load from memory and broadcast to all elements of the
7192 // destination operand
7194 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7195 X86MemOperand x86memop, Intrinsic Int> :
7196 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7197 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7198 [(set RC:$dst, (Int addr:$src))]>, VEX;
7200 // AVX2 adds register forms
7201 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7203 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7204 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7205 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7207 let ExeDomain = SSEPackedSingle in {
7208 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7209 int_x86_avx_vbroadcast_ss>;
7210 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7211 int_x86_avx_vbroadcast_ss_256>;
7213 let ExeDomain = SSEPackedDouble in
7214 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7215 int_x86_avx_vbroadcast_sd_256>;
7216 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7217 int_x86_avx_vbroadcastf128_pd_256>;
7219 let ExeDomain = SSEPackedSingle in {
7220 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7221 int_x86_avx2_vbroadcast_ss_ps>;
7222 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7223 int_x86_avx2_vbroadcast_ss_ps_256>;
7225 let ExeDomain = SSEPackedDouble in
7226 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7227 int_x86_avx2_vbroadcast_sd_pd_256>;
7229 let Predicates = [HasAVX2] in
7230 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7231 int_x86_avx2_vbroadcasti128>;
7233 let Predicates = [HasAVX] in
7234 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7235 (VBROADCASTF128 addr:$src)>;
7238 //===----------------------------------------------------------------------===//
7239 // VINSERTF128 - Insert packed floating-point values
7241 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7242 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7243 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7244 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7247 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7248 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7249 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7253 let Predicates = [HasAVX] in {
7254 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7256 (VINSERTF128rr VR256:$src1, VR128:$src2,
7257 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7258 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7260 (VINSERTF128rr VR256:$src1, VR128:$src2,
7261 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7262 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7264 (VINSERTF128rr VR256:$src1, VR128:$src2,
7265 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7266 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7268 (VINSERTF128rr VR256:$src1, VR128:$src2,
7269 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7270 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7272 (VINSERTF128rr VR256:$src1, VR128:$src2,
7273 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7274 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7276 (VINSERTF128rr VR256:$src1, VR128:$src2,
7277 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7279 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7281 (VINSERTF128rm VR256:$src1, addr:$src2,
7282 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7283 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7285 (VINSERTF128rm VR256:$src1, addr:$src2,
7286 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7287 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7289 (VINSERTF128rm VR256:$src1, addr:$src2,
7290 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7293 //===----------------------------------------------------------------------===//
7294 // VEXTRACTF128 - Extract packed floating-point values
7296 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7297 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7298 (ins VR256:$src1, i8imm:$src2),
7299 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7302 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7303 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7304 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7308 // Extract and store.
7309 let Predicates = [HasAVX] in {
7310 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7311 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7312 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7313 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7314 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7315 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7317 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7318 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7319 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7320 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7321 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7322 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7326 let Predicates = [HasAVX] in {
7327 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7328 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7329 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7330 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7331 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7332 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7334 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7335 (v4f32 (VEXTRACTF128rr
7336 (v8f32 VR256:$src1),
7337 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7338 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7339 (v2f64 (VEXTRACTF128rr
7340 (v4f64 VR256:$src1),
7341 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7342 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7343 (v2i64 (VEXTRACTF128rr
7344 (v4i64 VR256:$src1),
7345 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7346 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7347 (v4i32 (VEXTRACTF128rr
7348 (v8i32 VR256:$src1),
7349 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7350 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7351 (v8i16 (VEXTRACTF128rr
7352 (v16i16 VR256:$src1),
7353 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7354 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7355 (v16i8 (VEXTRACTF128rr
7356 (v32i8 VR256:$src1),
7357 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7360 //===----------------------------------------------------------------------===//
7361 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7363 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7364 Intrinsic IntLd, Intrinsic IntLd256,
7365 Intrinsic IntSt, Intrinsic IntSt256> {
7366 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7367 (ins VR128:$src1, f128mem:$src2),
7368 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7369 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7371 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7372 (ins VR256:$src1, f256mem:$src2),
7373 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7374 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7376 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7377 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7379 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7380 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7381 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7382 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7383 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7386 let ExeDomain = SSEPackedSingle in
7387 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7388 int_x86_avx_maskload_ps,
7389 int_x86_avx_maskload_ps_256,
7390 int_x86_avx_maskstore_ps,
7391 int_x86_avx_maskstore_ps_256>;
7392 let ExeDomain = SSEPackedDouble in
7393 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7394 int_x86_avx_maskload_pd,
7395 int_x86_avx_maskload_pd_256,
7396 int_x86_avx_maskstore_pd,
7397 int_x86_avx_maskstore_pd_256>;
7399 //===----------------------------------------------------------------------===//
7400 // VPERMIL - Permute Single and Double Floating-Point Values
7402 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7403 RegisterClass RC, X86MemOperand x86memop_f,
7404 X86MemOperand x86memop_i, PatFrag i_frag,
7405 Intrinsic IntVar, ValueType vt> {
7406 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7407 (ins RC:$src1, RC:$src2),
7408 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7409 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7410 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7411 (ins RC:$src1, x86memop_i:$src2),
7412 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7413 [(set RC:$dst, (IntVar RC:$src1,
7414 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7416 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7417 (ins RC:$src1, i8imm:$src2),
7418 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7419 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7420 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7421 (ins x86memop_f:$src1, i8imm:$src2),
7422 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7424 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7427 let ExeDomain = SSEPackedSingle in {
7428 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7429 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7430 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7431 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7433 let ExeDomain = SSEPackedDouble in {
7434 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7435 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7436 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7437 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7440 let Predicates = [HasAVX] in {
7441 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7442 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7443 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7444 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7445 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7447 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7448 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7449 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7451 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7452 (VPERMILPDri VR128:$src1, imm:$imm)>;
7453 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7454 (VPERMILPDmi addr:$src1, imm:$imm)>;
7457 //===----------------------------------------------------------------------===//
7458 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7460 let ExeDomain = SSEPackedSingle in {
7461 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7462 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7463 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7464 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7465 (i8 imm:$src3))))]>, VEX_4V;
7466 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7467 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7468 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7469 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7470 (i8 imm:$src3)))]>, VEX_4V;
7473 let Predicates = [HasAVX] in {
7474 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7475 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7476 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7477 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7478 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7479 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7480 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7481 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7482 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7483 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7485 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7486 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7487 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7488 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7489 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7490 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7491 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7492 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7493 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7494 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7495 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7496 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7497 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7498 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7499 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7500 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7501 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7502 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7505 //===----------------------------------------------------------------------===//
7506 // VZERO - Zero YMM registers
7508 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7509 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7510 // Zero All YMM registers
7511 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7512 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7514 // Zero Upper bits of YMM registers
7515 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7516 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7519 //===----------------------------------------------------------------------===//
7520 // Half precision conversion instructions
7521 //===----------------------------------------------------------------------===//
7522 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7523 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7524 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7525 [(set RC:$dst, (Int VR128:$src))]>,
7527 let neverHasSideEffects = 1, mayLoad = 1 in
7528 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7529 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7532 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7533 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7534 (ins RC:$src1, i32i8imm:$src2),
7535 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7536 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7538 let neverHasSideEffects = 1, mayStore = 1 in
7539 def mr : Ii8<0x1D, MRMDestMem, (outs),
7540 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7541 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7545 let Predicates = [HasAVX, HasF16C] in {
7546 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7547 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7548 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7549 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7552 //===----------------------------------------------------------------------===//
7553 // AVX2 Instructions
7554 //===----------------------------------------------------------------------===//
7556 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7557 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7558 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7559 X86MemOperand x86memop> {
7560 let isCommutable = 1 in
7561 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7562 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7563 !strconcat(OpcodeStr,
7564 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7565 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7567 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7568 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7569 !strconcat(OpcodeStr,
7570 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7573 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7577 let isCommutable = 0 in {
7578 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7579 VR128, memopv2i64, i128mem>;
7580 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7581 VR256, memopv4i64, i256mem>;
7584 //===----------------------------------------------------------------------===//
7585 // VPBROADCAST - Load from memory and broadcast to all elements of the
7586 // destination operand
7588 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7589 X86MemOperand x86memop, PatFrag ld_frag,
7590 Intrinsic Int128, Intrinsic Int256> {
7591 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7592 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7593 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7594 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7595 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7597 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7598 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7599 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7600 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7601 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7602 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7604 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7607 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7608 int_x86_avx2_pbroadcastb_128,
7609 int_x86_avx2_pbroadcastb_256>;
7610 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7611 int_x86_avx2_pbroadcastw_128,
7612 int_x86_avx2_pbroadcastw_256>;
7613 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7614 int_x86_avx2_pbroadcastd_128,
7615 int_x86_avx2_pbroadcastd_256>;
7616 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7617 int_x86_avx2_pbroadcastq_128,
7618 int_x86_avx2_pbroadcastq_256>;
7620 let Predicates = [HasAVX2] in {
7621 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7622 (VPBROADCASTBrm addr:$src)>;
7623 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7624 (VPBROADCASTBYrm addr:$src)>;
7625 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7626 (VPBROADCASTWrm addr:$src)>;
7627 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7628 (VPBROADCASTWYrm addr:$src)>;
7629 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7630 (VPBROADCASTDrm addr:$src)>;
7631 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7632 (VPBROADCASTDYrm addr:$src)>;
7633 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7634 (VPBROADCASTQrm addr:$src)>;
7635 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7636 (VPBROADCASTQYrm addr:$src)>;
7638 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7639 (VPBROADCASTBrr VR128:$src)>;
7640 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7641 (VPBROADCASTBYrr VR128:$src)>;
7642 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7643 (VPBROADCASTWrr VR128:$src)>;
7644 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7645 (VPBROADCASTWYrr VR128:$src)>;
7646 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7647 (VPBROADCASTDrr VR128:$src)>;
7648 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7649 (VPBROADCASTDYrr VR128:$src)>;
7650 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7651 (VPBROADCASTQrr VR128:$src)>;
7652 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7653 (VPBROADCASTQYrr VR128:$src)>;
7654 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7655 (VBROADCASTSSrr VR128:$src)>;
7656 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7657 (VBROADCASTSSYrr VR128:$src)>;
7658 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7659 (VPBROADCASTQrr VR128:$src)>;
7660 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7661 (VBROADCASTSDYrr VR128:$src)>;
7663 // Provide fallback in case the load node that is used in the patterns above
7664 // is used by additional users, which prevents the pattern selection.
7665 let AddedComplexity = 20 in {
7666 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7667 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7668 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7669 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7670 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7671 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7673 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7674 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7675 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7676 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7677 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7678 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7682 // AVX1 broadcast patterns
7683 let Predicates = [HasAVX] in {
7684 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7685 (VBROADCASTSSYrm addr:$src)>;
7686 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7687 (VBROADCASTSDYrm addr:$src)>;
7688 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7689 (VBROADCASTSSYrm addr:$src)>;
7690 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7691 (VBROADCASTSDYrm addr:$src)>;
7692 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7693 (VBROADCASTSSrm addr:$src)>;
7694 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7695 (VBROADCASTSSrm addr:$src)>;
7697 // Provide fallback in case the load node that is used in the patterns above
7698 // is used by additional users, which prevents the pattern selection.
7699 let AddedComplexity = 20 in {
7700 // 128bit broadcasts:
7701 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7702 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7703 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7704 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7705 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7706 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7707 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7708 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7709 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7710 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7712 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7713 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7714 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7715 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7716 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7717 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7718 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7719 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7720 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7721 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7725 //===----------------------------------------------------------------------===//
7726 // VPERM - Permute instructions
7729 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7731 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7732 (ins VR256:$src1, VR256:$src2),
7733 !strconcat(OpcodeStr,
7734 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7736 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7737 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7738 (ins VR256:$src1, i256mem:$src2),
7739 !strconcat(OpcodeStr,
7740 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7742 (OpVT (X86VPermv VR256:$src1,
7743 (bitconvert (mem_frag addr:$src2)))))]>,
7747 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7748 let ExeDomain = SSEPackedSingle in
7749 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7751 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7753 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7754 (ins VR256:$src1, i8imm:$src2),
7755 !strconcat(OpcodeStr,
7756 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7758 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7759 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7760 (ins i256mem:$src1, i8imm:$src2),
7761 !strconcat(OpcodeStr,
7762 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7764 (OpVT (X86VPermi (mem_frag addr:$src1),
7765 (i8 imm:$src2))))]>, VEX;
7768 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7769 let ExeDomain = SSEPackedDouble in
7770 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7772 //===----------------------------------------------------------------------===//
7773 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7775 let AddedComplexity = 1 in {
7776 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7777 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7778 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7779 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7780 (i8 imm:$src3))))]>, VEX_4V;
7781 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7782 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7783 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7784 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7785 (i8 imm:$src3)))]>, VEX_4V;
7788 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7789 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7790 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7791 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7792 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7793 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7794 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7796 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7798 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7799 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7800 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7801 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7802 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7804 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7808 //===----------------------------------------------------------------------===//
7809 // VINSERTI128 - Insert packed integer values
7811 let neverHasSideEffects = 1 in {
7812 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7813 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7814 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7817 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7818 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7819 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7823 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7824 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7826 (VINSERTI128rr VR256:$src1, VR128:$src2,
7827 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7828 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7830 (VINSERTI128rr VR256:$src1, VR128:$src2,
7831 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7832 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7834 (VINSERTI128rr VR256:$src1, VR128:$src2,
7835 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7836 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7838 (VINSERTI128rr VR256:$src1, VR128:$src2,
7839 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7842 //===----------------------------------------------------------------------===//
7843 // VEXTRACTI128 - Extract packed integer values
7845 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7846 (ins VR256:$src1, i8imm:$src2),
7847 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7849 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7851 let neverHasSideEffects = 1, mayStore = 1 in
7852 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7853 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7854 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7856 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7857 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7858 (v2i64 (VEXTRACTI128rr
7859 (v4i64 VR256:$src1),
7860 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7861 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7862 (v4i32 (VEXTRACTI128rr
7863 (v8i32 VR256:$src1),
7864 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7865 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7866 (v8i16 (VEXTRACTI128rr
7867 (v16i16 VR256:$src1),
7868 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7869 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7870 (v16i8 (VEXTRACTI128rr
7871 (v32i8 VR256:$src1),
7872 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7875 //===----------------------------------------------------------------------===//
7876 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7878 multiclass avx2_pmovmask<string OpcodeStr,
7879 Intrinsic IntLd128, Intrinsic IntLd256,
7880 Intrinsic IntSt128, Intrinsic IntSt256> {
7881 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7882 (ins VR128:$src1, i128mem:$src2),
7883 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7884 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7885 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7886 (ins VR256:$src1, i256mem:$src2),
7887 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7888 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7889 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7890 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7891 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7892 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7893 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7894 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7895 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7896 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7899 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7900 int_x86_avx2_maskload_d,
7901 int_x86_avx2_maskload_d_256,
7902 int_x86_avx2_maskstore_d,
7903 int_x86_avx2_maskstore_d_256>;
7904 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7905 int_x86_avx2_maskload_q,
7906 int_x86_avx2_maskload_q_256,
7907 int_x86_avx2_maskstore_q,
7908 int_x86_avx2_maskstore_q_256>, VEX_W;
7911 //===----------------------------------------------------------------------===//
7912 // Variable Bit Shifts
7914 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7915 ValueType vt128, ValueType vt256> {
7916 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7917 (ins VR128:$src1, VR128:$src2),
7918 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7920 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7922 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7923 (ins VR128:$src1, i128mem:$src2),
7924 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7926 (vt128 (OpNode VR128:$src1,
7927 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7929 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7930 (ins VR256:$src1, VR256:$src2),
7931 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7933 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7935 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7936 (ins VR256:$src1, i256mem:$src2),
7937 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7939 (vt256 (OpNode VR256:$src1,
7940 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7944 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7945 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7946 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7947 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7948 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
7950 //===----------------------------------------------------------------------===//
7951 // VGATHER - GATHER Operations
7952 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
7953 X86MemOperand memop128, X86MemOperand memop256> {
7954 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
7955 (ins VR128:$src1, memop128:$src2, VR128:$mask),
7956 !strconcat(OpcodeStr,
7957 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7959 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
7960 (ins RC256:$src1, memop256:$src2, RC256:$mask),
7961 !strconcat(OpcodeStr,
7962 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7963 []>, VEX_4VOp3, VEX_L;
7966 let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
7967 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
7968 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
7969 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
7970 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
7971 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
7972 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
7973 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
7974 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;