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:$src)))],
1664 IIC_SSE_CVT_Scalar_RM>,
1666 Requires<[HasSSE2, OptForSize]>;
1668 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1669 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1672 let Constraints = "$src1 = $dst" in
1673 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1674 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1675 SSE_CVT_Scalar>, XS;
1677 // Convert scalar single to scalar double
1678 // SSE2 instructions with XS prefix
1679 let neverHasSideEffects = 1 in {
1680 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1681 (ins FR32:$src1, FR32:$src2),
1682 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1683 [], IIC_SSE_CVT_Scalar_RR>,
1684 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1686 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1687 (ins FR32:$src1, f32mem:$src2),
1688 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1689 [], IIC_SSE_CVT_Scalar_RM>,
1690 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1693 let Predicates = [HasAVX] in {
1694 def : Pat<(f64 (fextend FR32:$src)),
1695 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1696 def : Pat<(fextend (loadf32 addr:$src)),
1697 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1698 def : Pat<(extloadf32 addr:$src),
1699 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1702 def : Pat<(extloadf32 addr:$src),
1703 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1704 Requires<[HasAVX, OptForSpeed]>;
1706 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1707 "cvtss2sd\t{$src, $dst|$dst, $src}",
1708 [(set FR64:$dst, (fextend FR32:$src))],
1709 IIC_SSE_CVT_Scalar_RR>, XS,
1710 Requires<[HasSSE2]>;
1711 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1712 "cvtss2sd\t{$src, $dst|$dst, $src}",
1713 [(set FR64:$dst, (extloadf32 addr:$src))],
1714 IIC_SSE_CVT_Scalar_RM>, XS,
1715 Requires<[HasSSE2, OptForSize]>;
1717 // extload f32 -> f64. This matches load+fextend because we have a hack in
1718 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1720 // Since these loads aren't folded into the fextend, we have to match it
1722 def : Pat<(fextend (loadf32 addr:$src)),
1723 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1724 def : Pat<(extloadf32 addr:$src),
1725 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1727 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1728 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1729 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1730 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1732 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V,
1734 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1735 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1736 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1738 (load addr:$src2)))],
1739 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V,
1741 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1742 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1743 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1744 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1745 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1747 IIC_SSE_CVT_Scalar_RR>, XS,
1748 Requires<[HasSSE2]>;
1749 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1750 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1751 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1752 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1753 (load addr:$src2)))],
1754 IIC_SSE_CVT_Scalar_RM>, XS,
1755 Requires<[HasSSE2]>;
1758 // Convert packed single/double fp to doubleword
1759 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1760 "cvtps2dq\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1762 IIC_SSE_CVT_PS_RR>, VEX;
1763 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1764 "cvtps2dq\t{$src, $dst|$dst, $src}",
1766 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1767 IIC_SSE_CVT_PS_RM>, VEX;
1768 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1769 "cvtps2dq\t{$src, $dst|$dst, $src}",
1771 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1772 IIC_SSE_CVT_PS_RR>, VEX;
1773 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1774 "cvtps2dq\t{$src, $dst|$dst, $src}",
1776 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1777 IIC_SSE_CVT_PS_RM>, VEX;
1778 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1779 "cvtps2dq\t{$src, $dst|$dst, $src}",
1780 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1782 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1783 "cvtps2dq\t{$src, $dst|$dst, $src}",
1785 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1789 // Convert Packed Double FP to Packed DW Integers
1790 let Predicates = [HasAVX] in {
1791 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1792 // register, but the same isn't true when using memory operands instead.
1793 // Provide other assembly rr and rm forms to address this explicitly.
1794 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1796 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1800 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1801 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1802 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1803 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1805 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1808 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1809 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1811 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX;
1812 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1813 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1815 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1817 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1818 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1821 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1822 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1824 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1826 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1827 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1828 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1831 // Convert with truncation packed single/double fp to doubleword
1832 // SSE2 packed instructions with XS prefix
1833 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1834 "cvttps2dq\t{$src, $dst|$dst, $src}",
1836 (int_x86_sse2_cvttps2dq VR128:$src))],
1837 IIC_SSE_CVT_PS_RR>, VEX;
1838 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1839 "cvttps2dq\t{$src, $dst|$dst, $src}",
1840 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1841 (memopv4f32 addr:$src)))],
1842 IIC_SSE_CVT_PS_RM>, VEX;
1843 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1844 "cvttps2dq\t{$src, $dst|$dst, $src}",
1846 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1847 IIC_SSE_CVT_PS_RR>, VEX;
1848 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1849 "cvttps2dq\t{$src, $dst|$dst, $src}",
1850 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1851 (memopv8f32 addr:$src)))],
1852 IIC_SSE_CVT_PS_RM>, VEX;
1854 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1855 "cvttps2dq\t{$src, $dst|$dst, $src}",
1857 (int_x86_sse2_cvttps2dq VR128:$src))],
1859 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1860 "cvttps2dq\t{$src, $dst|$dst, $src}",
1862 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1865 let Predicates = [HasAVX] in {
1866 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1867 (VCVTDQ2PSrr VR128:$src)>;
1868 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1869 (VCVTDQ2PSrm addr:$src)>;
1871 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1872 (VCVTDQ2PSrr VR128:$src)>;
1873 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1874 (VCVTDQ2PSrm addr:$src)>;
1876 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1877 (VCVTTPS2DQrr VR128:$src)>;
1878 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1879 (VCVTTPS2DQrm addr:$src)>;
1881 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1882 (VCVTDQ2PSYrr VR256:$src)>;
1883 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1884 (VCVTDQ2PSYrm addr:$src)>;
1886 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1887 (VCVTTPS2DQYrr VR256:$src)>;
1888 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1889 (VCVTTPS2DQYrm addr:$src)>;
1892 let Predicates = [HasSSE2] in {
1893 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1894 (CVTDQ2PSrr VR128:$src)>;
1895 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1896 (CVTDQ2PSrm addr:$src)>;
1898 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1899 (CVTDQ2PSrr VR128:$src)>;
1900 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1901 (CVTDQ2PSrm addr:$src)>;
1903 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1904 (CVTTPS2DQrr VR128:$src)>;
1905 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1906 (CVTTPS2DQrm addr:$src)>;
1909 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1910 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1912 (int_x86_sse2_cvttpd2dq VR128:$src))],
1913 IIC_SSE_CVT_PD_RR>, VEX;
1915 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1916 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1917 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1919 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1920 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1921 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1922 (memopv2f64 addr:$src)))],
1925 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1926 // register, but the same isn't true when using memory operands instead.
1927 // Provide other assembly rr and rm forms to address this explicitly.
1930 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1931 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1932 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1933 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1934 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1935 (memopv2f64 addr:$src)))],
1936 IIC_SSE_CVT_PD_RM>, VEX;
1939 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1940 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1942 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
1943 IIC_SSE_CVT_PD_RR>, VEX;
1944 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1945 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1947 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
1948 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1949 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1950 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1952 let Predicates = [HasAVX] in {
1953 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1954 (VCVTTPD2DQYrr VR256:$src)>;
1955 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1956 (VCVTTPD2DQYrm addr:$src)>;
1957 } // Predicates = [HasAVX]
1959 // Convert packed single to packed double
1960 let Predicates = [HasAVX] in {
1961 // SSE2 instructions without OpSize prefix
1962 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1963 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1964 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1965 IIC_SSE_CVT_PD_RR>, TB, VEX;
1966 let neverHasSideEffects = 1, mayLoad = 1 in
1967 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1968 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1969 IIC_SSE_CVT_PD_RM>, TB, VEX;
1970 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1971 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1973 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
1974 IIC_SSE_CVT_PD_RR>, TB, VEX;
1975 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1976 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1978 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
1979 IIC_SSE_CVT_PD_RM>, TB, VEX;
1982 let Predicates = [HasSSE2] in {
1983 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1984 "cvtps2pd\t{$src, $dst|$dst, $src}",
1985 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1986 IIC_SSE_CVT_PD_RR>, TB;
1987 let neverHasSideEffects = 1, mayLoad = 1 in
1988 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1989 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
1990 IIC_SSE_CVT_PD_RM>, TB;
1993 // Convert Packed DW Integers to Packed Double FP
1994 let Predicates = [HasAVX] in {
1995 let neverHasSideEffects = 1, mayLoad = 1 in
1996 def VCVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1997 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1999 def VCVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2000 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2002 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2003 def VCVTDQ2PDYrm : SSDI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2004 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2006 (int_x86_avx_cvtdq2_pd_256
2007 (bitconvert (memopv2i64 addr:$src))))]>, VEX;
2008 def VCVTDQ2PDYrr : SSDI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2009 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2011 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX;
2014 let neverHasSideEffects = 1, mayLoad = 1 in
2015 def CVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2016 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2018 def CVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2019 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2020 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2023 // AVX 256-bit register conversion intrinsics
2024 let Predicates = [HasAVX] in {
2025 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2026 (VCVTDQ2PDYrr VR128:$src)>;
2027 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2028 (VCVTDQ2PDYrm addr:$src)>;
2029 } // Predicates = [HasAVX]
2031 // Convert packed double to packed single
2032 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2033 // register, but the same isn't true when using memory operands instead.
2034 // Provide other assembly rr and rm forms to address this explicitly.
2035 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2036 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2037 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2038 IIC_SSE_CVT_PD_RR>, VEX;
2041 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2042 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2043 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2044 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2046 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2047 IIC_SSE_CVT_PD_RM>, VEX;
2050 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2051 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2053 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2054 IIC_SSE_CVT_PD_RR>, VEX;
2055 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2056 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2058 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2059 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2060 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2061 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2063 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2064 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2065 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2067 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2068 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2070 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2074 // AVX 256-bit register conversion intrinsics
2075 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2076 // whenever possible to avoid declaring two versions of each one.
2077 let Predicates = [HasAVX] in {
2078 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2079 (VCVTDQ2PSYrr VR256:$src)>;
2080 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2081 (VCVTDQ2PSYrm addr:$src)>;
2083 // Match fround and fextend for 128/256-bit conversions
2084 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2085 (VCVTPD2PSYrr VR256:$src)>;
2086 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2087 (VCVTPD2PSYrm addr:$src)>;
2089 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2090 (VCVTPS2PDYrr VR128:$src)>;
2091 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2092 (VCVTPS2PDYrm addr:$src)>;
2095 //===----------------------------------------------------------------------===//
2096 // SSE 1 & 2 - Compare Instructions
2097 //===----------------------------------------------------------------------===//
2099 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2100 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2101 Operand CC, SDNode OpNode, ValueType VT,
2102 PatFrag ld_frag, string asm, string asm_alt,
2104 def rr : SIi8<0xC2, MRMSrcReg,
2105 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2106 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2108 def rm : SIi8<0xC2, MRMSrcMem,
2109 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2110 [(set RC:$dst, (OpNode (VT RC:$src1),
2111 (ld_frag addr:$src2), imm:$cc))],
2114 // Accept explicit immediate argument form instead of comparison code.
2115 let neverHasSideEffects = 1 in {
2116 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2117 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2118 IIC_SSE_ALU_F32S_RR>;
2120 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2121 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2122 IIC_SSE_ALU_F32S_RM>;
2126 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2127 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2128 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2130 XS, VEX_4V, VEX_LIG;
2131 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2132 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2133 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2134 SSE_ALU_F32S>, // same latency as 32 bit compare
2135 XD, VEX_4V, VEX_LIG;
2137 let Constraints = "$src1 = $dst" in {
2138 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2139 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2140 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2142 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2143 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2144 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2145 SSE_ALU_F32S>, // same latency as 32 bit compare
2149 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2150 Intrinsic Int, string asm, OpndItins itins> {
2151 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2152 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2153 [(set VR128:$dst, (Int VR128:$src1,
2154 VR128:$src, imm:$cc))],
2156 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2157 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2158 [(set VR128:$dst, (Int VR128:$src1,
2159 (load addr:$src), imm:$cc))],
2163 // Aliases to match intrinsics which expect XMM operand(s).
2164 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2165 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2168 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2169 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2170 SSE_ALU_F32S>, // same latency as f32
2172 let Constraints = "$src1 = $dst" in {
2173 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2174 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2176 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2177 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2178 SSE_ALU_F32S>, // same latency as f32
2183 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2184 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2185 ValueType vt, X86MemOperand x86memop,
2186 PatFrag ld_frag, string OpcodeStr, Domain d> {
2187 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2188 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2189 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2190 IIC_SSE_COMIS_RR, d>;
2191 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2192 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2193 [(set EFLAGS, (OpNode (vt RC:$src1),
2194 (ld_frag addr:$src2)))],
2195 IIC_SSE_COMIS_RM, d>;
2198 let Defs = [EFLAGS] in {
2199 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2200 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2201 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2202 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2204 let Pattern = []<dag> in {
2205 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2206 "comiss", SSEPackedSingle>, TB, VEX,
2208 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2209 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2213 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2214 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2215 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2216 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2218 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2219 load, "comiss", SSEPackedSingle>, TB, VEX;
2220 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2221 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2222 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2223 "ucomiss", SSEPackedSingle>, TB;
2224 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2225 "ucomisd", SSEPackedDouble>, TB, OpSize;
2227 let Pattern = []<dag> in {
2228 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2229 "comiss", SSEPackedSingle>, TB;
2230 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2231 "comisd", SSEPackedDouble>, TB, OpSize;
2234 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2235 load, "ucomiss", SSEPackedSingle>, TB;
2236 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2237 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2239 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2240 "comiss", SSEPackedSingle>, TB;
2241 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2242 "comisd", SSEPackedDouble>, TB, OpSize;
2243 } // Defs = [EFLAGS]
2245 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2246 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2247 Operand CC, Intrinsic Int, string asm,
2248 string asm_alt, Domain d> {
2249 def rri : PIi8<0xC2, MRMSrcReg,
2250 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2251 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2252 IIC_SSE_CMPP_RR, d>;
2253 def rmi : PIi8<0xC2, MRMSrcMem,
2254 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2255 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2256 IIC_SSE_CMPP_RM, d>;
2258 // Accept explicit immediate argument form instead of comparison code.
2259 let neverHasSideEffects = 1 in {
2260 def rri_alt : PIi8<0xC2, MRMSrcReg,
2261 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2262 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2263 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2264 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2265 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2269 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2270 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2271 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2272 SSEPackedSingle>, TB, VEX_4V;
2273 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2274 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2275 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2276 SSEPackedDouble>, TB, OpSize, VEX_4V;
2277 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2278 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2279 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2280 SSEPackedSingle>, TB, VEX_4V;
2281 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2282 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2283 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2284 SSEPackedDouble>, TB, OpSize, VEX_4V;
2285 let Constraints = "$src1 = $dst" in {
2286 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2287 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2288 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2289 SSEPackedSingle>, TB;
2290 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2291 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2292 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2293 SSEPackedDouble>, TB, OpSize;
2296 let Predicates = [HasAVX] in {
2297 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2298 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2299 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2300 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2301 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2302 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2303 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2304 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2306 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2307 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2308 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2309 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2310 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2311 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2312 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2313 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2316 let Predicates = [HasSSE1] in {
2317 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2318 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2319 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2320 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2323 let Predicates = [HasSSE2] in {
2324 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2325 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2326 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2327 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2330 //===----------------------------------------------------------------------===//
2331 // SSE 1 & 2 - Shuffle Instructions
2332 //===----------------------------------------------------------------------===//
2334 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2335 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2336 ValueType vt, string asm, PatFrag mem_frag,
2337 Domain d, bit IsConvertibleToThreeAddress = 0> {
2338 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2339 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2340 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2341 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2342 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2343 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2344 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2345 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2346 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2349 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2350 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2351 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2352 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2353 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2354 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2355 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2356 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2357 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2358 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2359 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2360 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2362 let Constraints = "$src1 = $dst" in {
2363 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2364 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2365 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2367 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2368 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2369 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2373 let Predicates = [HasAVX] in {
2374 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2375 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2376 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2377 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2378 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2380 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2381 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2382 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2383 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2384 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2387 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2388 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2389 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2390 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2391 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2393 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2394 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2395 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2396 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2397 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2400 let Predicates = [HasSSE1] in {
2401 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2402 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2403 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2404 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2405 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2408 let Predicates = [HasSSE2] in {
2409 // Generic SHUFPD patterns
2410 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2411 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2412 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2413 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2414 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2417 //===----------------------------------------------------------------------===//
2418 // SSE 1 & 2 - Unpack Instructions
2419 //===----------------------------------------------------------------------===//
2421 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2422 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2423 PatFrag mem_frag, RegisterClass RC,
2424 X86MemOperand x86memop, string asm,
2426 def rr : PI<opc, MRMSrcReg,
2427 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2429 (vt (OpNode RC:$src1, RC:$src2)))],
2431 def rm : PI<opc, MRMSrcMem,
2432 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2434 (vt (OpNode RC:$src1,
2435 (mem_frag addr:$src2))))],
2439 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2440 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2441 SSEPackedSingle>, TB, VEX_4V;
2442 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2443 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2444 SSEPackedDouble>, TB, OpSize, VEX_4V;
2445 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2446 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2447 SSEPackedSingle>, TB, VEX_4V;
2448 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2449 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2450 SSEPackedDouble>, TB, OpSize, VEX_4V;
2452 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2453 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2454 SSEPackedSingle>, TB, VEX_4V;
2455 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2456 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2457 SSEPackedDouble>, TB, OpSize, VEX_4V;
2458 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2459 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2460 SSEPackedSingle>, TB, VEX_4V;
2461 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2462 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2463 SSEPackedDouble>, TB, OpSize, VEX_4V;
2465 let Constraints = "$src1 = $dst" in {
2466 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2467 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2468 SSEPackedSingle>, TB;
2469 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2470 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2471 SSEPackedDouble>, TB, OpSize;
2472 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2473 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2474 SSEPackedSingle>, TB;
2475 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2476 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2477 SSEPackedDouble>, TB, OpSize;
2478 } // Constraints = "$src1 = $dst"
2480 let Predicates = [HasAVX], AddedComplexity = 1 in {
2481 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2482 // problem is during lowering, where it's not possible to recognize the load
2483 // fold cause it has two uses through a bitcast. One use disappears at isel
2484 // time and the fold opportunity reappears.
2485 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2486 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2489 let Predicates = [HasSSE2] in {
2490 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2491 // problem is during lowering, where it's not possible to recognize the load
2492 // fold cause it has two uses through a bitcast. One use disappears at isel
2493 // time and the fold opportunity reappears.
2494 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2495 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2498 //===----------------------------------------------------------------------===//
2499 // SSE 1 & 2 - Extract Floating-Point Sign mask
2500 //===----------------------------------------------------------------------===//
2502 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2503 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2505 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2506 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2507 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2508 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2509 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2510 IIC_SSE_MOVMSK, d>, REX_W;
2513 let Predicates = [HasAVX] in {
2514 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2515 "movmskps", SSEPackedSingle>, TB, VEX;
2516 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2517 "movmskpd", SSEPackedDouble>, TB,
2519 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2520 "movmskps", SSEPackedSingle>, TB, VEX;
2521 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2522 "movmskpd", SSEPackedDouble>, TB,
2525 def : Pat<(i32 (X86fgetsign FR32:$src)),
2526 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2527 def : Pat<(i64 (X86fgetsign FR32:$src)),
2528 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2529 def : Pat<(i32 (X86fgetsign FR64:$src)),
2530 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2531 def : Pat<(i64 (X86fgetsign FR64:$src)),
2532 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2535 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2536 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2537 SSEPackedSingle>, TB, VEX;
2538 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2539 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2540 SSEPackedDouble>, TB,
2542 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2543 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2544 SSEPackedSingle>, TB, VEX;
2545 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2546 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2547 SSEPackedDouble>, TB,
2551 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2552 SSEPackedSingle>, TB;
2553 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2554 SSEPackedDouble>, TB, OpSize;
2556 def : Pat<(i32 (X86fgetsign FR32:$src)),
2557 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2558 Requires<[HasSSE1]>;
2559 def : Pat<(i64 (X86fgetsign FR32:$src)),
2560 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2561 Requires<[HasSSE1]>;
2562 def : Pat<(i32 (X86fgetsign FR64:$src)),
2563 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2564 Requires<[HasSSE2]>;
2565 def : Pat<(i64 (X86fgetsign FR64:$src)),
2566 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2567 Requires<[HasSSE2]>;
2569 //===---------------------------------------------------------------------===//
2570 // SSE2 - Packed Integer Logical Instructions
2571 //===---------------------------------------------------------------------===//
2573 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2575 /// PDI_binop_rm - Simple SSE2 binary operator.
2576 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2577 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2578 X86MemOperand x86memop,
2580 bit IsCommutable = 0,
2582 let isCommutable = IsCommutable in
2583 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2584 (ins RC:$src1, RC:$src2),
2586 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2587 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2588 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2589 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2590 (ins RC:$src1, x86memop:$src2),
2592 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2593 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2594 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2595 (bitconvert (memop_frag addr:$src2)))))],
2598 } // ExeDomain = SSEPackedInt
2600 // These are ordered here for pattern ordering requirements with the fp versions
2602 let Predicates = [HasAVX] in {
2603 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2604 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2605 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2606 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2607 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2608 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2609 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2610 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2613 let Constraints = "$src1 = $dst" in {
2614 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2615 i128mem, SSE_BIT_ITINS_P, 1>;
2616 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2617 i128mem, SSE_BIT_ITINS_P, 1>;
2618 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2619 i128mem, SSE_BIT_ITINS_P, 1>;
2620 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2621 i128mem, SSE_BIT_ITINS_P, 0>;
2622 } // Constraints = "$src1 = $dst"
2624 let Predicates = [HasAVX2] in {
2625 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2626 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2627 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2628 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2629 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2630 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2631 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2632 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2635 //===----------------------------------------------------------------------===//
2636 // SSE 1 & 2 - Logical Instructions
2637 //===----------------------------------------------------------------------===//
2639 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2641 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2642 SDNode OpNode, OpndItins itins> {
2643 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2644 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2647 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2648 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2651 let Constraints = "$src1 = $dst" in {
2652 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2653 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2656 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2657 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2662 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2663 let mayLoad = 0 in {
2664 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2666 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2668 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2672 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2673 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2676 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2678 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2680 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2681 // are all promoted to v2i64, and the patterns are covered by the int
2682 // version. This is needed in SSE only, because v2i64 isn't supported on
2683 // SSE1, but only on SSE2.
2684 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2685 !strconcat(OpcodeStr, "ps"), f128mem, [],
2686 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2687 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2689 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2690 !strconcat(OpcodeStr, "pd"), f128mem,
2691 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2692 (bc_v2i64 (v2f64 VR128:$src2))))],
2693 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2694 (memopv2i64 addr:$src2)))], 0>,
2696 let Constraints = "$src1 = $dst" in {
2697 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2698 !strconcat(OpcodeStr, "ps"), f128mem,
2699 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2700 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2701 (memopv2i64 addr:$src2)))]>, TB;
2703 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2704 !strconcat(OpcodeStr, "pd"), f128mem,
2705 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2706 (bc_v2i64 (v2f64 VR128:$src2))))],
2707 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2708 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2712 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2714 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2716 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2717 !strconcat(OpcodeStr, "ps"), f256mem,
2718 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2719 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2720 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2722 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2723 !strconcat(OpcodeStr, "pd"), f256mem,
2724 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2725 (bc_v4i64 (v4f64 VR256:$src2))))],
2726 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2727 (memopv4i64 addr:$src2)))], 0>,
2731 // AVX 256-bit packed logical ops forms
2732 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2733 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2734 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2735 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2737 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2738 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2739 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2740 let isCommutable = 0 in
2741 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2743 //===----------------------------------------------------------------------===//
2744 // SSE 1 & 2 - Arithmetic Instructions
2745 //===----------------------------------------------------------------------===//
2747 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2750 /// In addition, we also have a special variant of the scalar form here to
2751 /// represent the associated intrinsic operation. This form is unlike the
2752 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2753 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2755 /// These three forms can each be reg+reg or reg+mem.
2758 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2760 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2763 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2764 OpNode, FR32, f32mem,
2765 itins.s, Is2Addr>, XS;
2766 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2767 OpNode, FR64, f64mem,
2768 itins.d, Is2Addr>, XD;
2771 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2774 let mayLoad = 0 in {
2775 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2776 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2778 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2779 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2784 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2787 let mayLoad = 0 in {
2788 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2789 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2791 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2792 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2797 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2800 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2801 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2802 itins.s, Is2Addr>, XS;
2803 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2804 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2805 itins.d, Is2Addr>, XD;
2808 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2811 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2812 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2813 SSEPackedSingle, itins.s, Is2Addr>,
2816 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2817 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2818 SSEPackedDouble, itins.d, Is2Addr>,
2822 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2824 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2825 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2826 SSEPackedSingle, itins.s, 0>, TB;
2828 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2829 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2830 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2833 // Binary Arithmetic instructions
2834 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2835 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2837 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2838 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2840 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2841 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2843 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2844 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2847 let isCommutable = 0 in {
2848 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2849 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2851 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2852 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2853 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2854 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2856 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2857 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2859 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2860 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2862 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2863 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2864 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2865 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2867 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2868 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2870 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2871 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2872 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2873 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2877 let Constraints = "$src1 = $dst" in {
2878 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2879 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2880 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2881 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2882 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2883 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2885 let isCommutable = 0 in {
2886 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2887 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2888 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2889 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2890 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2891 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2892 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2893 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2894 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2895 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2896 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2897 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2898 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2899 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2904 /// In addition, we also have a special variant of the scalar form here to
2905 /// represent the associated intrinsic operation. This form is unlike the
2906 /// plain scalar form, in that it takes an entire vector (instead of a
2907 /// scalar) and leaves the top elements undefined.
2909 /// And, we have a special variant form for a full-vector intrinsic form.
2911 def SSE_SQRTP : OpndItins<
2912 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2915 def SSE_SQRTS : OpndItins<
2916 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2919 def SSE_RCPP : OpndItins<
2920 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2923 def SSE_RCPS : OpndItins<
2924 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2927 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2928 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2929 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2930 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2931 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2932 [(set FR32:$dst, (OpNode FR32:$src))]>;
2933 // For scalar unary operations, fold a load into the operation
2934 // only in OptForSize mode. It eliminates an instruction, but it also
2935 // eliminates a whole-register clobber (the load), so it introduces a
2936 // partial register update condition.
2937 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2938 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2939 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2940 Requires<[HasSSE1, OptForSize]>;
2941 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2942 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2943 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2944 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2945 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2946 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2949 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2950 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2951 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2952 !strconcat(OpcodeStr,
2953 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2955 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2956 !strconcat(OpcodeStr,
2957 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2958 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2959 (ins VR128:$src1, ssmem:$src2),
2960 !strconcat(OpcodeStr,
2961 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2964 /// sse1_fp_unop_p - SSE1 unops in packed form.
2965 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2967 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2968 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2969 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
2970 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2971 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2972 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
2975 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2976 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
2978 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2979 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2980 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
2982 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2983 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2984 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
2988 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2989 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2990 Intrinsic V4F32Int, OpndItins itins> {
2991 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2992 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2993 [(set VR128:$dst, (V4F32Int VR128:$src))],
2995 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2996 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2997 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3001 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3002 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3003 Intrinsic V4F32Int, OpndItins itins> {
3004 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3005 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3006 [(set VR256:$dst, (V4F32Int VR256:$src))],
3008 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3009 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3010 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3014 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3015 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3016 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3017 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3018 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3019 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3020 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3021 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3022 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3023 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3024 Requires<[HasSSE2, OptForSize]>;
3025 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3026 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3027 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3028 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3029 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3030 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3033 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3034 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3035 let neverHasSideEffects = 1 in {
3036 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3037 !strconcat(OpcodeStr,
3038 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3040 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3041 !strconcat(OpcodeStr,
3042 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3044 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3045 (ins VR128:$src1, sdmem:$src2),
3046 !strconcat(OpcodeStr,
3047 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3050 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3051 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3052 SDNode OpNode, OpndItins itins> {
3053 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3054 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3055 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3056 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3057 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3058 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3061 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3062 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3064 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3065 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3066 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3068 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3069 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3070 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3074 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3075 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3076 Intrinsic V2F64Int, OpndItins itins> {
3077 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3078 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3079 [(set VR128:$dst, (V2F64Int VR128:$src))],
3081 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3082 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3083 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3087 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3088 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3089 Intrinsic V2F64Int, OpndItins itins> {
3090 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3091 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3092 [(set VR256:$dst, (V2F64Int VR256:$src))],
3094 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3095 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3096 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3100 let Predicates = [HasAVX] in {
3102 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3103 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3105 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3106 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3107 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3108 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3109 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3111 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3113 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3115 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3119 // Reciprocal approximations. Note that these typically require refinement
3120 // in order to obtain suitable precision.
3121 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3122 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3123 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3124 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3126 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3129 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3130 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3131 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3132 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3134 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3138 let AddedComplexity = 1 in {
3139 def : Pat<(f32 (fsqrt FR32:$src)),
3140 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3141 def : Pat<(f32 (fsqrt (load addr:$src))),
3142 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3143 Requires<[HasAVX, OptForSize]>;
3144 def : Pat<(f64 (fsqrt FR64:$src)),
3145 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3146 def : Pat<(f64 (fsqrt (load addr:$src))),
3147 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3148 Requires<[HasAVX, OptForSize]>;
3150 def : Pat<(f32 (X86frsqrt FR32:$src)),
3151 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3152 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3153 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3154 Requires<[HasAVX, OptForSize]>;
3156 def : Pat<(f32 (X86frcp FR32:$src)),
3157 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3158 def : Pat<(f32 (X86frcp (load addr:$src))),
3159 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3160 Requires<[HasAVX, OptForSize]>;
3163 let Predicates = [HasAVX], AddedComplexity = 1 in {
3164 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3165 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3166 (COPY_TO_REGCLASS VR128:$src, FR32)),
3168 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3169 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3171 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3172 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3173 (COPY_TO_REGCLASS VR128:$src, FR64)),
3175 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3176 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3178 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3179 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3180 (COPY_TO_REGCLASS VR128:$src, FR32)),
3182 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3183 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3185 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3186 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3187 (COPY_TO_REGCLASS VR128:$src, FR32)),
3189 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3190 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3194 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3196 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3197 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3198 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3200 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3201 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3203 // Reciprocal approximations. Note that these typically require refinement
3204 // in order to obtain suitable precision.
3205 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3207 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3208 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3210 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3212 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3213 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3215 // There is no f64 version of the reciprocal approximation instructions.
3217 //===----------------------------------------------------------------------===//
3218 // SSE 1 & 2 - Non-temporal stores
3219 //===----------------------------------------------------------------------===//
3221 let AddedComplexity = 400 in { // Prefer non-temporal versions
3222 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3223 (ins f128mem:$dst, VR128:$src),
3224 "movntps\t{$src, $dst|$dst, $src}",
3225 [(alignednontemporalstore (v4f32 VR128:$src),
3227 IIC_SSE_MOVNT>, VEX;
3228 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3229 (ins f128mem:$dst, VR128:$src),
3230 "movntpd\t{$src, $dst|$dst, $src}",
3231 [(alignednontemporalstore (v2f64 VR128:$src),
3233 IIC_SSE_MOVNT>, VEX;
3235 let ExeDomain = SSEPackedInt in
3236 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3237 (ins f128mem:$dst, VR128:$src),
3238 "movntdq\t{$src, $dst|$dst, $src}",
3239 [(alignednontemporalstore (v2i64 VR128:$src),
3241 IIC_SSE_MOVNT>, VEX;
3243 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3244 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3246 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3247 (ins f256mem:$dst, VR256:$src),
3248 "movntps\t{$src, $dst|$dst, $src}",
3249 [(alignednontemporalstore (v8f32 VR256:$src),
3251 IIC_SSE_MOVNT>, VEX;
3252 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3253 (ins f256mem:$dst, VR256:$src),
3254 "movntpd\t{$src, $dst|$dst, $src}",
3255 [(alignednontemporalstore (v4f64 VR256:$src),
3257 IIC_SSE_MOVNT>, VEX;
3258 let ExeDomain = SSEPackedInt in
3259 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3260 (ins f256mem:$dst, VR256:$src),
3261 "movntdq\t{$src, $dst|$dst, $src}",
3262 [(alignednontemporalstore (v4i64 VR256:$src),
3264 IIC_SSE_MOVNT>, VEX;
3267 let AddedComplexity = 400 in { // Prefer non-temporal versions
3268 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3269 "movntps\t{$src, $dst|$dst, $src}",
3270 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3272 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3273 "movntpd\t{$src, $dst|$dst, $src}",
3274 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3277 let ExeDomain = SSEPackedInt in
3278 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3279 "movntdq\t{$src, $dst|$dst, $src}",
3280 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3283 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3284 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3286 // There is no AVX form for instructions below this point
3287 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3288 "movnti{l}\t{$src, $dst|$dst, $src}",
3289 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3291 TB, Requires<[HasSSE2]>;
3292 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3293 "movnti{q}\t{$src, $dst|$dst, $src}",
3294 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3296 TB, Requires<[HasSSE2]>;
3299 //===----------------------------------------------------------------------===//
3300 // SSE 1 & 2 - Prefetch and memory fence
3301 //===----------------------------------------------------------------------===//
3303 // Prefetch intrinsic.
3304 let Predicates = [HasSSE1] in {
3305 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3306 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3307 IIC_SSE_PREFETCH>, TB;
3308 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3309 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3310 IIC_SSE_PREFETCH>, TB;
3311 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3312 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3313 IIC_SSE_PREFETCH>, TB;
3314 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3315 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3316 IIC_SSE_PREFETCH>, TB;
3320 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3321 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3322 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3324 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3325 // was introduced with SSE2, it's backward compatible.
3326 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3328 // Load, store, and memory fence
3329 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3330 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3331 TB, Requires<[HasSSE1]>;
3332 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3333 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3334 TB, Requires<[HasSSE2]>;
3335 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3336 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3337 TB, Requires<[HasSSE2]>;
3339 def : Pat<(X86SFence), (SFENCE)>;
3340 def : Pat<(X86LFence), (LFENCE)>;
3341 def : Pat<(X86MFence), (MFENCE)>;
3343 //===----------------------------------------------------------------------===//
3344 // SSE 1 & 2 - Load/Store XCSR register
3345 //===----------------------------------------------------------------------===//
3347 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3348 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3349 IIC_SSE_LDMXCSR>, VEX;
3350 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3351 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3352 IIC_SSE_STMXCSR>, VEX;
3354 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3355 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3357 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3358 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3361 //===---------------------------------------------------------------------===//
3362 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3363 //===---------------------------------------------------------------------===//
3365 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3367 let neverHasSideEffects = 1 in {
3368 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3369 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3371 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3372 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3375 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3376 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3378 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3379 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3383 let isCodeGenOnly = 1 in {
3384 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3385 "movdqa\t{$src, $dst|$dst, $src}", [],
3388 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3389 "movdqa\t{$src, $dst|$dst, $src}", [],
3392 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3393 "movdqu\t{$src, $dst|$dst, $src}", [],
3396 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3397 "movdqu\t{$src, $dst|$dst, $src}", [],
3402 let canFoldAsLoad = 1, mayLoad = 1 in {
3403 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3404 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3406 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3407 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3409 let Predicates = [HasAVX] in {
3410 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3411 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3413 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3414 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3419 let mayStore = 1 in {
3420 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3421 (ins i128mem:$dst, VR128:$src),
3422 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3424 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3425 (ins i256mem:$dst, VR256:$src),
3426 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3428 let Predicates = [HasAVX] in {
3429 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3430 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3432 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3433 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3438 let neverHasSideEffects = 1 in
3439 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3440 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3442 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3443 "movdqu\t{$src, $dst|$dst, $src}",
3444 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3447 let isCodeGenOnly = 1 in {
3448 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3449 "movdqa\t{$src, $dst|$dst, $src}", [],
3452 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3453 "movdqu\t{$src, $dst|$dst, $src}",
3454 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3457 let canFoldAsLoad = 1, mayLoad = 1 in {
3458 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3459 "movdqa\t{$src, $dst|$dst, $src}",
3460 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3462 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3463 "movdqu\t{$src, $dst|$dst, $src}",
3464 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3466 XS, Requires<[HasSSE2]>;
3469 let mayStore = 1 in {
3470 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3471 "movdqa\t{$src, $dst|$dst, $src}",
3472 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3474 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3475 "movdqu\t{$src, $dst|$dst, $src}",
3476 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3478 XS, Requires<[HasSSE2]>;
3481 // Intrinsic forms of MOVDQU load and store
3482 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3483 "vmovdqu\t{$src, $dst|$dst, $src}",
3484 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3486 XS, VEX, Requires<[HasAVX]>;
3488 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3489 "movdqu\t{$src, $dst|$dst, $src}",
3490 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3492 XS, Requires<[HasSSE2]>;
3494 } // ExeDomain = SSEPackedInt
3496 let Predicates = [HasAVX] in {
3497 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3498 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3501 //===---------------------------------------------------------------------===//
3502 // SSE2 - Packed Integer Arithmetic Instructions
3503 //===---------------------------------------------------------------------===//
3505 def SSE_PMADD : OpndItins<
3506 IIC_SSE_PMADD, IIC_SSE_PMADD
3509 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3511 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3512 RegisterClass RC, PatFrag memop_frag,
3513 X86MemOperand x86memop,
3515 bit IsCommutable = 0,
3517 let isCommutable = IsCommutable in
3518 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3519 (ins RC:$src1, RC:$src2),
3521 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3522 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3523 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3524 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3525 (ins RC:$src1, x86memop:$src2),
3527 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3528 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3529 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3533 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3534 string OpcodeStr, SDNode OpNode,
3535 SDNode OpNode2, RegisterClass RC,
3536 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3537 ShiftOpndItins itins,
3539 // src2 is always 128-bit
3540 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3541 (ins RC:$src1, VR128:$src2),
3543 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3544 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3545 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3547 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3548 (ins RC:$src1, i128mem:$src2),
3550 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3551 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3552 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3553 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3554 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3555 (ins RC:$src1, i32i8imm:$src2),
3557 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3558 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3559 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3562 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3563 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3564 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3565 PatFrag memop_frag, X86MemOperand x86memop,
3567 bit IsCommutable = 0, bit Is2Addr = 1> {
3568 let isCommutable = IsCommutable in
3569 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3570 (ins RC:$src1, RC:$src2),
3572 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3573 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3574 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3575 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3576 (ins RC:$src1, x86memop:$src2),
3578 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3579 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3580 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3581 (bitconvert (memop_frag addr:$src2)))))]>;
3583 } // ExeDomain = SSEPackedInt
3585 // 128-bit Integer Arithmetic
3587 let Predicates = [HasAVX] in {
3588 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3589 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3591 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3592 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3593 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3594 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3595 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3596 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3597 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3598 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3599 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3600 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3601 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3602 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3603 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3604 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3605 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3606 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3607 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3608 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3612 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3613 VR128, memopv2i64, i128mem,
3614 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3615 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3616 VR128, memopv2i64, i128mem,
3617 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3618 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3619 VR128, memopv2i64, i128mem,
3620 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3621 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3622 VR128, memopv2i64, i128mem,
3623 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3624 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3625 VR128, memopv2i64, i128mem,
3626 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3627 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3628 VR128, memopv2i64, i128mem,
3629 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3630 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3631 VR128, memopv2i64, i128mem,
3632 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3633 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3634 VR128, memopv2i64, i128mem,
3635 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3636 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3637 VR128, memopv2i64, i128mem,
3638 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3639 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3640 VR128, memopv2i64, i128mem,
3641 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3642 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3643 VR128, memopv2i64, i128mem,
3644 SSE_PMADD, 1, 0>, VEX_4V;
3645 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3646 VR128, memopv2i64, i128mem,
3647 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3648 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3649 VR128, memopv2i64, i128mem,
3650 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3651 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3652 VR128, memopv2i64, i128mem,
3653 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3654 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3655 VR128, memopv2i64, i128mem,
3656 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3657 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3658 VR128, memopv2i64, i128mem,
3659 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3660 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3661 VR128, memopv2i64, i128mem,
3662 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3663 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3664 VR128, memopv2i64, i128mem,
3665 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3668 let Predicates = [HasAVX2] in {
3669 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3670 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3671 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3672 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3673 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3674 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3675 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3676 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3677 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3678 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3679 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3680 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3681 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3682 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3683 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3684 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3685 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3686 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3687 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3688 VR256, memopv4i64, i256mem,
3689 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3692 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3693 VR256, memopv4i64, i256mem,
3694 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3695 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3696 VR256, memopv4i64, i256mem,
3697 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3698 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3699 VR256, memopv4i64, i256mem,
3700 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3701 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3702 VR256, memopv4i64, i256mem,
3703 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3704 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3705 VR256, memopv4i64, i256mem,
3706 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3707 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3708 VR256, memopv4i64, i256mem,
3709 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3710 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3711 VR256, memopv4i64, i256mem,
3712 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3713 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3714 VR256, memopv4i64, i256mem,
3715 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3716 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3717 VR256, memopv4i64, i256mem,
3718 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3719 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3720 VR256, memopv4i64, i256mem,
3721 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3722 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3723 VR256, memopv4i64, i256mem,
3724 SSE_PMADD, 1, 0>, VEX_4V;
3725 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3726 VR256, memopv4i64, i256mem,
3727 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3728 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3729 VR256, memopv4i64, i256mem,
3730 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3731 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3732 VR256, memopv4i64, i256mem,
3733 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3734 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3735 VR256, memopv4i64, i256mem,
3736 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3737 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3738 VR256, memopv4i64, i256mem,
3739 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3740 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3741 VR256, memopv4i64, i256mem,
3742 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3743 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3744 VR256, memopv4i64, i256mem,
3745 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3748 let Constraints = "$src1 = $dst" in {
3749 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3750 i128mem, SSE_INTALU_ITINS_P, 1>;
3751 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3752 i128mem, SSE_INTALU_ITINS_P, 1>;
3753 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3754 i128mem, SSE_INTALU_ITINS_P, 1>;
3755 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3756 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3757 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3758 i128mem, SSE_INTMUL_ITINS_P, 1>;
3759 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3760 i128mem, SSE_INTALU_ITINS_P>;
3761 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3762 i128mem, SSE_INTALU_ITINS_P>;
3763 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3764 i128mem, SSE_INTALU_ITINS_P>;
3765 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3766 i128mem, SSE_INTALUQ_ITINS_P>;
3767 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3768 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3771 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3772 VR128, memopv2i64, i128mem,
3773 SSE_INTALU_ITINS_P>;
3774 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3775 VR128, memopv2i64, i128mem,
3776 SSE_INTALU_ITINS_P>;
3777 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3778 VR128, memopv2i64, i128mem,
3779 SSE_INTALU_ITINS_P>;
3780 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3781 VR128, memopv2i64, i128mem,
3782 SSE_INTALU_ITINS_P>;
3783 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3784 VR128, memopv2i64, i128mem,
3785 SSE_INTALU_ITINS_P, 1>;
3786 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3787 VR128, memopv2i64, i128mem,
3788 SSE_INTALU_ITINS_P, 1>;
3789 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3790 VR128, memopv2i64, i128mem,
3791 SSE_INTALU_ITINS_P, 1>;
3792 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3793 VR128, memopv2i64, i128mem,
3794 SSE_INTALU_ITINS_P, 1>;
3795 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3796 VR128, memopv2i64, i128mem,
3797 SSE_INTMUL_ITINS_P, 1>;
3798 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3799 VR128, memopv2i64, i128mem,
3800 SSE_INTMUL_ITINS_P, 1>;
3801 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3802 VR128, memopv2i64, i128mem,
3804 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3805 VR128, memopv2i64, i128mem,
3806 SSE_INTALU_ITINS_P, 1>;
3807 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3808 VR128, memopv2i64, i128mem,
3809 SSE_INTALU_ITINS_P, 1>;
3810 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3811 VR128, memopv2i64, i128mem,
3812 SSE_INTALU_ITINS_P, 1>;
3813 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3814 VR128, memopv2i64, i128mem,
3815 SSE_INTALU_ITINS_P, 1>;
3816 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3817 VR128, memopv2i64, i128mem,
3818 SSE_INTALU_ITINS_P, 1>;
3819 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3820 VR128, memopv2i64, i128mem,
3821 SSE_INTALU_ITINS_P, 1>;
3822 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3823 VR128, memopv2i64, i128mem,
3824 SSE_INTALU_ITINS_P, 1>;
3826 } // Constraints = "$src1 = $dst"
3828 //===---------------------------------------------------------------------===//
3829 // SSE2 - Packed Integer Logical Instructions
3830 //===---------------------------------------------------------------------===//
3832 let Predicates = [HasAVX] in {
3833 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3834 VR128, v8i16, v8i16, bc_v8i16,
3835 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3836 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3837 VR128, v4i32, v4i32, bc_v4i32,
3838 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3839 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3840 VR128, v2i64, v2i64, bc_v2i64,
3841 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3843 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3844 VR128, v8i16, v8i16, bc_v8i16,
3845 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3846 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3847 VR128, v4i32, v4i32, bc_v4i32,
3848 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3849 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3850 VR128, v2i64, v2i64, bc_v2i64,
3851 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3853 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3854 VR128, v8i16, v8i16, bc_v8i16,
3855 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3856 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3857 VR128, v4i32, v4i32, bc_v4i32,
3858 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3860 let ExeDomain = SSEPackedInt in {
3861 // 128-bit logical shifts.
3862 def VPSLLDQri : PDIi8<0x73, MRM7r,
3863 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3864 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3866 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3868 def VPSRLDQri : PDIi8<0x73, MRM3r,
3869 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3870 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3872 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3874 // PSRADQri doesn't exist in SSE[1-3].
3876 } // Predicates = [HasAVX]
3878 let Predicates = [HasAVX2] in {
3879 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3880 VR256, v16i16, v8i16, bc_v8i16,
3881 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3882 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3883 VR256, v8i32, v4i32, bc_v4i32,
3884 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3885 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3886 VR256, v4i64, v2i64, bc_v2i64,
3887 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3889 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3890 VR256, v16i16, v8i16, bc_v8i16,
3891 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3892 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3893 VR256, v8i32, v4i32, bc_v4i32,
3894 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3895 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3896 VR256, v4i64, v2i64, bc_v2i64,
3897 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3899 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3900 VR256, v16i16, v8i16, bc_v8i16,
3901 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3902 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3903 VR256, v8i32, v4i32, bc_v4i32,
3904 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3906 let ExeDomain = SSEPackedInt in {
3907 // 256-bit logical shifts.
3908 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3909 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3910 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3912 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3914 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3915 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3916 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3918 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3920 // PSRADQYri doesn't exist in SSE[1-3].
3922 } // Predicates = [HasAVX2]
3924 let Constraints = "$src1 = $dst" in {
3925 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3926 VR128, v8i16, v8i16, bc_v8i16,
3927 SSE_INTSHIFT_ITINS_P>;
3928 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3929 VR128, v4i32, v4i32, bc_v4i32,
3930 SSE_INTSHIFT_ITINS_P>;
3931 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3932 VR128, v2i64, v2i64, bc_v2i64,
3933 SSE_INTSHIFT_ITINS_P>;
3935 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3936 VR128, v8i16, v8i16, bc_v8i16,
3937 SSE_INTSHIFT_ITINS_P>;
3938 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3939 VR128, v4i32, v4i32, bc_v4i32,
3940 SSE_INTSHIFT_ITINS_P>;
3941 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3942 VR128, v2i64, v2i64, bc_v2i64,
3943 SSE_INTSHIFT_ITINS_P>;
3945 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3946 VR128, v8i16, v8i16, bc_v8i16,
3947 SSE_INTSHIFT_ITINS_P>;
3948 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3949 VR128, v4i32, v4i32, bc_v4i32,
3950 SSE_INTSHIFT_ITINS_P>;
3952 let ExeDomain = SSEPackedInt in {
3953 // 128-bit logical shifts.
3954 def PSLLDQri : PDIi8<0x73, MRM7r,
3955 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3956 "pslldq\t{$src2, $dst|$dst, $src2}",
3958 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3959 def PSRLDQri : PDIi8<0x73, MRM3r,
3960 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3961 "psrldq\t{$src2, $dst|$dst, $src2}",
3963 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3964 // PSRADQri doesn't exist in SSE[1-3].
3966 } // Constraints = "$src1 = $dst"
3968 let Predicates = [HasAVX] in {
3969 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3970 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3971 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3972 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3973 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3974 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3976 // Shift up / down and insert zero's.
3977 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3978 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3979 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3980 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3983 let Predicates = [HasAVX2] in {
3984 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3985 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3986 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3987 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3990 let Predicates = [HasSSE2] in {
3991 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3992 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3993 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3994 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3995 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3996 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3998 // Shift up / down and insert zero's.
3999 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4000 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4001 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4002 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4005 //===---------------------------------------------------------------------===//
4006 // SSE2 - Packed Integer Comparison Instructions
4007 //===---------------------------------------------------------------------===//
4009 let Predicates = [HasAVX] in {
4010 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4011 VR128, memopv2i64, i128mem,
4012 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4013 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4014 VR128, memopv2i64, i128mem,
4015 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4016 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4017 VR128, memopv2i64, i128mem,
4018 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4019 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4020 VR128, memopv2i64, i128mem,
4021 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4022 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4023 VR128, memopv2i64, i128mem,
4024 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4025 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4026 VR128, memopv2i64, i128mem,
4027 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4030 let Predicates = [HasAVX2] in {
4031 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4032 VR256, memopv4i64, i256mem,
4033 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4034 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4035 VR256, memopv4i64, i256mem,
4036 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4037 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4038 VR256, memopv4i64, i256mem,
4039 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4040 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4041 VR256, memopv4i64, i256mem,
4042 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4043 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4044 VR256, memopv4i64, i256mem,
4045 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4046 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4047 VR256, memopv4i64, i256mem,
4048 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4051 let Constraints = "$src1 = $dst" in {
4052 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4053 VR128, memopv2i64, i128mem,
4054 SSE_INTALU_ITINS_P, 1>;
4055 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4056 VR128, memopv2i64, i128mem,
4057 SSE_INTALU_ITINS_P, 1>;
4058 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4059 VR128, memopv2i64, i128mem,
4060 SSE_INTALU_ITINS_P, 1>;
4061 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4062 VR128, memopv2i64, i128mem,
4063 SSE_INTALU_ITINS_P>;
4064 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4065 VR128, memopv2i64, i128mem,
4066 SSE_INTALU_ITINS_P>;
4067 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4068 VR128, memopv2i64, i128mem,
4069 SSE_INTALU_ITINS_P>;
4070 } // Constraints = "$src1 = $dst"
4072 //===---------------------------------------------------------------------===//
4073 // SSE2 - Packed Integer Pack Instructions
4074 //===---------------------------------------------------------------------===//
4076 let Predicates = [HasAVX] in {
4077 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4078 VR128, memopv2i64, i128mem,
4079 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4080 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4081 VR128, memopv2i64, i128mem,
4082 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4083 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4084 VR128, memopv2i64, i128mem,
4085 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4088 let Predicates = [HasAVX2] in {
4089 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4090 VR256, memopv4i64, i256mem,
4091 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4092 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4093 VR256, memopv4i64, i256mem,
4094 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4095 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4096 VR256, memopv4i64, i256mem,
4097 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4100 let Constraints = "$src1 = $dst" in {
4101 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4102 VR128, memopv2i64, i128mem,
4103 SSE_INTALU_ITINS_P>;
4104 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4105 VR128, memopv2i64, i128mem,
4106 SSE_INTALU_ITINS_P>;
4107 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4108 VR128, memopv2i64, i128mem,
4109 SSE_INTALU_ITINS_P>;
4110 } // Constraints = "$src1 = $dst"
4112 //===---------------------------------------------------------------------===//
4113 // SSE2 - Packed Integer Shuffle Instructions
4114 //===---------------------------------------------------------------------===//
4116 let ExeDomain = SSEPackedInt in {
4117 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4118 def ri : Ii8<0x70, MRMSrcReg,
4119 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4120 !strconcat(OpcodeStr,
4121 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4122 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4124 def mi : Ii8<0x70, MRMSrcMem,
4125 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4126 !strconcat(OpcodeStr,
4127 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4129 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4134 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4135 def Yri : Ii8<0x70, MRMSrcReg,
4136 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4137 !strconcat(OpcodeStr,
4138 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4139 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4140 def Ymi : Ii8<0x70, MRMSrcMem,
4141 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4142 !strconcat(OpcodeStr,
4143 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4145 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4146 (i8 imm:$src2))))]>;
4148 } // ExeDomain = SSEPackedInt
4150 let Predicates = [HasAVX] in {
4151 let AddedComplexity = 5 in
4152 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4154 // SSE2 with ImmT == Imm8 and XS prefix.
4155 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4157 // SSE2 with ImmT == Imm8 and XD prefix.
4158 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4160 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4161 (VPSHUFDmi addr:$src1, imm:$imm)>;
4162 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4163 (VPSHUFDri VR128:$src1, imm:$imm)>;
4166 let Predicates = [HasAVX2] in {
4167 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4168 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4169 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4172 let Predicates = [HasSSE2] in {
4173 let AddedComplexity = 5 in
4174 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4176 // SSE2 with ImmT == Imm8 and XS prefix.
4177 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4179 // SSE2 with ImmT == Imm8 and XD prefix.
4180 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4182 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4183 (PSHUFDmi addr:$src1, imm:$imm)>;
4184 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4185 (PSHUFDri VR128:$src1, imm:$imm)>;
4188 //===---------------------------------------------------------------------===//
4189 // SSE2 - Packed Integer Unpack Instructions
4190 //===---------------------------------------------------------------------===//
4192 let ExeDomain = SSEPackedInt in {
4193 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4194 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4195 def rr : PDI<opc, MRMSrcReg,
4196 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4198 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4199 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4200 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4202 def rm : PDI<opc, MRMSrcMem,
4203 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4205 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4206 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4207 [(set VR128:$dst, (OpNode VR128:$src1,
4208 (bc_frag (memopv2i64
4213 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4214 SDNode OpNode, PatFrag bc_frag> {
4215 def Yrr : PDI<opc, MRMSrcReg,
4216 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4217 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4218 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4219 def Yrm : PDI<opc, MRMSrcMem,
4220 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4221 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4222 [(set VR256:$dst, (OpNode VR256:$src1,
4223 (bc_frag (memopv4i64 addr:$src2))))]>;
4226 let Predicates = [HasAVX] in {
4227 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4228 bc_v16i8, 0>, VEX_4V;
4229 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4230 bc_v8i16, 0>, VEX_4V;
4231 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4232 bc_v4i32, 0>, VEX_4V;
4233 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4234 bc_v2i64, 0>, VEX_4V;
4236 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4237 bc_v16i8, 0>, VEX_4V;
4238 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4239 bc_v8i16, 0>, VEX_4V;
4240 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4241 bc_v4i32, 0>, VEX_4V;
4242 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4243 bc_v2i64, 0>, VEX_4V;
4246 let Predicates = [HasAVX2] in {
4247 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4249 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4251 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4253 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4256 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4258 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4260 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4262 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4266 let Constraints = "$src1 = $dst" in {
4267 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4269 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4271 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4273 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4276 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4278 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4280 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4282 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4285 } // ExeDomain = SSEPackedInt
4287 // Patterns for using AVX1 instructions with integer vectors
4288 // Here to give AVX2 priority
4289 let Predicates = [HasAVX] in {
4290 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4291 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4292 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4293 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4294 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4295 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4296 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4297 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4299 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4300 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4301 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4302 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4303 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4304 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4305 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4306 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4309 //===---------------------------------------------------------------------===//
4310 // SSE2 - Packed Integer Extract and Insert
4311 //===---------------------------------------------------------------------===//
4313 let ExeDomain = SSEPackedInt in {
4314 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4315 def rri : Ii8<0xC4, MRMSrcReg,
4316 (outs VR128:$dst), (ins VR128:$src1,
4317 GR32:$src2, i32i8imm:$src3),
4319 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4320 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4322 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4323 def rmi : Ii8<0xC4, MRMSrcMem,
4324 (outs VR128:$dst), (ins VR128:$src1,
4325 i16mem:$src2, i32i8imm:$src3),
4327 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4328 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4330 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4331 imm:$src3))], IIC_SSE_PINSRW>;
4335 let Predicates = [HasAVX] in
4336 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4337 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4338 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4339 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4340 imm:$src2))]>, TB, OpSize, VEX;
4341 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4342 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4343 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4344 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4345 imm:$src2))], IIC_SSE_PEXTRW>;
4348 let Predicates = [HasAVX] in {
4349 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4350 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4351 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4352 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4353 []>, TB, OpSize, VEX_4V;
4356 let Constraints = "$src1 = $dst" in
4357 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4359 } // ExeDomain = SSEPackedInt
4361 //===---------------------------------------------------------------------===//
4362 // SSE2 - Packed Mask Creation
4363 //===---------------------------------------------------------------------===//
4365 let ExeDomain = SSEPackedInt in {
4367 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4368 "pmovmskb\t{$src, $dst|$dst, $src}",
4369 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4370 IIC_SSE_MOVMSK>, VEX;
4371 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4372 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4374 let Predicates = [HasAVX2] in {
4375 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4376 "pmovmskb\t{$src, $dst|$dst, $src}",
4377 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4378 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4379 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4382 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4383 "pmovmskb\t{$src, $dst|$dst, $src}",
4384 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4387 } // ExeDomain = SSEPackedInt
4389 //===---------------------------------------------------------------------===//
4390 // SSE2 - Conditional Store
4391 //===---------------------------------------------------------------------===//
4393 let ExeDomain = SSEPackedInt in {
4396 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4397 (ins VR128:$src, VR128:$mask),
4398 "maskmovdqu\t{$mask, $src|$src, $mask}",
4399 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4400 IIC_SSE_MASKMOV>, VEX;
4402 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4403 (ins VR128:$src, VR128:$mask),
4404 "maskmovdqu\t{$mask, $src|$src, $mask}",
4405 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4406 IIC_SSE_MASKMOV>, VEX;
4409 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4410 "maskmovdqu\t{$mask, $src|$src, $mask}",
4411 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4414 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4415 "maskmovdqu\t{$mask, $src|$src, $mask}",
4416 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4419 } // ExeDomain = SSEPackedInt
4421 //===---------------------------------------------------------------------===//
4422 // SSE2 - Move Doubleword
4423 //===---------------------------------------------------------------------===//
4425 //===---------------------------------------------------------------------===//
4426 // Move Int Doubleword to Packed Double Int
4428 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4429 "movd\t{$src, $dst|$dst, $src}",
4431 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4433 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4436 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4439 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4440 "mov{d|q}\t{$src, $dst|$dst, $src}",
4442 (v2i64 (scalar_to_vector GR64:$src)))],
4443 IIC_SSE_MOVDQ>, VEX;
4444 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4445 "mov{d|q}\t{$src, $dst|$dst, $src}",
4446 [(set FR64:$dst, (bitconvert GR64:$src))],
4447 IIC_SSE_MOVDQ>, VEX;
4449 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4450 "movd\t{$src, $dst|$dst, $src}",
4452 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4453 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4454 "movd\t{$src, $dst|$dst, $src}",
4456 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4458 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4459 "mov{d|q}\t{$src, $dst|$dst, $src}",
4461 (v2i64 (scalar_to_vector GR64:$src)))],
4463 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4464 "mov{d|q}\t{$src, $dst|$dst, $src}",
4465 [(set FR64:$dst, (bitconvert GR64:$src))],
4468 //===---------------------------------------------------------------------===//
4469 // Move Int Doubleword to Single Scalar
4471 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4472 "movd\t{$src, $dst|$dst, $src}",
4473 [(set FR32:$dst, (bitconvert GR32:$src))],
4474 IIC_SSE_MOVDQ>, VEX;
4476 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4477 "movd\t{$src, $dst|$dst, $src}",
4478 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4481 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4482 "movd\t{$src, $dst|$dst, $src}",
4483 [(set FR32:$dst, (bitconvert GR32:$src))],
4486 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4487 "movd\t{$src, $dst|$dst, $src}",
4488 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4491 //===---------------------------------------------------------------------===//
4492 // Move Packed Doubleword Int to Packed Double Int
4494 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4495 "movd\t{$src, $dst|$dst, $src}",
4496 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4497 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4498 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4499 (ins i32mem:$dst, VR128:$src),
4500 "movd\t{$src, $dst|$dst, $src}",
4501 [(store (i32 (vector_extract (v4i32 VR128:$src),
4502 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4504 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4505 "movd\t{$src, $dst|$dst, $src}",
4506 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4507 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4508 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4509 "movd\t{$src, $dst|$dst, $src}",
4510 [(store (i32 (vector_extract (v4i32 VR128:$src),
4511 (iPTR 0))), addr:$dst)],
4514 //===---------------------------------------------------------------------===//
4515 // Move Packed Doubleword Int first element to Doubleword Int
4517 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4518 "mov{d|q}\t{$src, $dst|$dst, $src}",
4519 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4522 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4524 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4525 "mov{d|q}\t{$src, $dst|$dst, $src}",
4526 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4530 //===---------------------------------------------------------------------===//
4531 // Bitcast FR64 <-> GR64
4533 let Predicates = [HasAVX] in
4534 def VMOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4535 "vmovq\t{$src, $dst|$dst, $src}",
4536 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4538 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4539 "mov{d|q}\t{$src, $dst|$dst, $src}",
4540 [(set GR64:$dst, (bitconvert FR64:$src))],
4541 IIC_SSE_MOVDQ>, VEX;
4542 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4543 "movq\t{$src, $dst|$dst, $src}",
4544 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4545 IIC_SSE_MOVDQ>, VEX;
4547 def MOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4548 "movq\t{$src, $dst|$dst, $src}",
4549 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4551 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4552 "mov{d|q}\t{$src, $dst|$dst, $src}",
4553 [(set GR64:$dst, (bitconvert FR64:$src))],
4555 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4556 "movq\t{$src, $dst|$dst, $src}",
4557 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4560 //===---------------------------------------------------------------------===//
4561 // Move Scalar Single to Double Int
4563 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4564 "movd\t{$src, $dst|$dst, $src}",
4565 [(set GR32:$dst, (bitconvert FR32:$src))],
4566 IIC_SSE_MOVD_ToGP>, VEX;
4567 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4568 "movd\t{$src, $dst|$dst, $src}",
4569 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4570 IIC_SSE_MOVDQ>, VEX;
4571 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4572 "movd\t{$src, $dst|$dst, $src}",
4573 [(set GR32:$dst, (bitconvert FR32:$src))],
4575 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4576 "movd\t{$src, $dst|$dst, $src}",
4577 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4580 //===---------------------------------------------------------------------===//
4581 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4583 let AddedComplexity = 15 in {
4584 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4585 "movd\t{$src, $dst|$dst, $src}",
4586 [(set VR128:$dst, (v4i32 (X86vzmovl
4587 (v4i32 (scalar_to_vector GR32:$src)))))],
4588 IIC_SSE_MOVDQ>, VEX;
4589 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4590 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4591 [(set VR128:$dst, (v2i64 (X86vzmovl
4592 (v2i64 (scalar_to_vector GR64:$src)))))],
4596 let AddedComplexity = 15 in {
4597 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4598 "movd\t{$src, $dst|$dst, $src}",
4599 [(set VR128:$dst, (v4i32 (X86vzmovl
4600 (v4i32 (scalar_to_vector GR32:$src)))))],
4602 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4603 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4604 [(set VR128:$dst, (v2i64 (X86vzmovl
4605 (v2i64 (scalar_to_vector GR64:$src)))))],
4609 let AddedComplexity = 20 in {
4610 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4611 "movd\t{$src, $dst|$dst, $src}",
4613 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4614 (loadi32 addr:$src))))))],
4615 IIC_SSE_MOVDQ>, VEX;
4616 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4617 "movd\t{$src, $dst|$dst, $src}",
4619 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4620 (loadi32 addr:$src))))))],
4624 let Predicates = [HasAVX] in {
4625 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4626 let AddedComplexity = 20 in {
4627 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4628 (VMOVZDI2PDIrm addr:$src)>;
4629 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4630 (VMOVZDI2PDIrm addr:$src)>;
4632 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4633 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4634 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4635 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4636 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4637 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4638 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4641 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4642 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4643 (MOVZDI2PDIrm addr:$src)>;
4644 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4645 (MOVZDI2PDIrm addr:$src)>;
4648 // These are the correct encodings of the instructions so that we know how to
4649 // read correct assembly, even though we continue to emit the wrong ones for
4650 // compatibility with Darwin's buggy assembler.
4651 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4652 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4653 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4654 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4655 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4656 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4657 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4658 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4659 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4660 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4661 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4662 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4664 //===---------------------------------------------------------------------===//
4665 // SSE2 - Move Quadword
4666 //===---------------------------------------------------------------------===//
4668 //===---------------------------------------------------------------------===//
4669 // Move Quadword Int to Packed Quadword Int
4671 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4672 "vmovq\t{$src, $dst|$dst, $src}",
4674 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4675 VEX, Requires<[HasAVX]>;
4676 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4677 "movq\t{$src, $dst|$dst, $src}",
4679 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4681 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4683 //===---------------------------------------------------------------------===//
4684 // Move Packed Quadword Int to Quadword Int
4686 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4687 "movq\t{$src, $dst|$dst, $src}",
4688 [(store (i64 (vector_extract (v2i64 VR128:$src),
4689 (iPTR 0))), addr:$dst)],
4690 IIC_SSE_MOVDQ>, VEX;
4691 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4692 "movq\t{$src, $dst|$dst, $src}",
4693 [(store (i64 (vector_extract (v2i64 VR128:$src),
4694 (iPTR 0))), addr:$dst)],
4697 //===---------------------------------------------------------------------===//
4698 // Store / copy lower 64-bits of a XMM register.
4700 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4701 "movq\t{$src, $dst|$dst, $src}",
4702 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4703 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4704 "movq\t{$src, $dst|$dst, $src}",
4705 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4708 let AddedComplexity = 20 in
4709 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4710 "vmovq\t{$src, $dst|$dst, $src}",
4712 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4713 (loadi64 addr:$src))))))],
4715 XS, VEX, Requires<[HasAVX]>;
4717 let AddedComplexity = 20 in
4718 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4719 "movq\t{$src, $dst|$dst, $src}",
4721 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4722 (loadi64 addr:$src))))))],
4724 XS, Requires<[HasSSE2]>;
4726 let Predicates = [HasAVX], AddedComplexity = 20 in {
4727 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4728 (VMOVZQI2PQIrm addr:$src)>;
4729 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4730 (VMOVZQI2PQIrm addr:$src)>;
4731 def : Pat<(v2i64 (X86vzload addr:$src)),
4732 (VMOVZQI2PQIrm addr:$src)>;
4735 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4736 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4737 (MOVZQI2PQIrm addr:$src)>;
4738 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4739 (MOVZQI2PQIrm addr:$src)>;
4740 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4743 let Predicates = [HasAVX] in {
4744 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4745 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4746 def : Pat<(v4i64 (X86vzload addr:$src)),
4747 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4750 //===---------------------------------------------------------------------===//
4751 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4752 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4754 let AddedComplexity = 15 in
4755 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4756 "vmovq\t{$src, $dst|$dst, $src}",
4757 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4759 XS, VEX, Requires<[HasAVX]>;
4760 let AddedComplexity = 15 in
4761 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4762 "movq\t{$src, $dst|$dst, $src}",
4763 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4765 XS, Requires<[HasSSE2]>;
4767 let AddedComplexity = 20 in
4768 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4769 "vmovq\t{$src, $dst|$dst, $src}",
4770 [(set VR128:$dst, (v2i64 (X86vzmovl
4771 (loadv2i64 addr:$src))))],
4773 XS, VEX, Requires<[HasAVX]>;
4774 let AddedComplexity = 20 in {
4775 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4776 "movq\t{$src, $dst|$dst, $src}",
4777 [(set VR128:$dst, (v2i64 (X86vzmovl
4778 (loadv2i64 addr:$src))))],
4780 XS, Requires<[HasSSE2]>;
4783 let AddedComplexity = 20 in {
4784 let Predicates = [HasAVX] in {
4785 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4786 (VMOVZPQILo2PQIrm addr:$src)>;
4787 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4788 (VMOVZPQILo2PQIrr VR128:$src)>;
4790 let Predicates = [HasSSE2] in {
4791 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4792 (MOVZPQILo2PQIrm addr:$src)>;
4793 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4794 (MOVZPQILo2PQIrr VR128:$src)>;
4798 // Instructions to match in the assembler
4799 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4800 "movq\t{$src, $dst|$dst, $src}", [],
4801 IIC_SSE_MOVDQ>, VEX, VEX_W;
4802 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4803 "movq\t{$src, $dst|$dst, $src}", [],
4804 IIC_SSE_MOVDQ>, VEX, VEX_W;
4805 // Recognize "movd" with GR64 destination, but encode as a "movq"
4806 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4807 "movd\t{$src, $dst|$dst, $src}", [],
4808 IIC_SSE_MOVDQ>, VEX, VEX_W;
4810 // Instructions for the disassembler
4811 // xr = XMM register
4814 let Predicates = [HasAVX] in
4815 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4816 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4817 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4818 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4820 //===---------------------------------------------------------------------===//
4821 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4822 //===---------------------------------------------------------------------===//
4823 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4824 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4825 X86MemOperand x86memop> {
4826 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4827 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4828 [(set RC:$dst, (vt (OpNode RC:$src)))],
4830 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4831 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4832 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4836 let Predicates = [HasAVX] in {
4837 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4838 v4f32, VR128, memopv4f32, f128mem>, VEX;
4839 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4840 v4f32, VR128, memopv4f32, f128mem>, VEX;
4841 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4842 v8f32, VR256, memopv8f32, f256mem>, VEX;
4843 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4844 v8f32, VR256, memopv8f32, f256mem>, VEX;
4846 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4847 memopv4f32, f128mem>;
4848 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4849 memopv4f32, f128mem>;
4851 let Predicates = [HasAVX] in {
4852 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4853 (VMOVSHDUPrr VR128:$src)>;
4854 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4855 (VMOVSHDUPrm addr:$src)>;
4856 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4857 (VMOVSLDUPrr VR128:$src)>;
4858 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4859 (VMOVSLDUPrm addr:$src)>;
4860 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4861 (VMOVSHDUPYrr VR256:$src)>;
4862 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4863 (VMOVSHDUPYrm addr:$src)>;
4864 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4865 (VMOVSLDUPYrr VR256:$src)>;
4866 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4867 (VMOVSLDUPYrm addr:$src)>;
4870 let Predicates = [HasSSE3] in {
4871 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4872 (MOVSHDUPrr VR128:$src)>;
4873 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4874 (MOVSHDUPrm addr:$src)>;
4875 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4876 (MOVSLDUPrr VR128:$src)>;
4877 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4878 (MOVSLDUPrm addr:$src)>;
4881 //===---------------------------------------------------------------------===//
4882 // SSE3 - Replicate Double FP - MOVDDUP
4883 //===---------------------------------------------------------------------===//
4885 multiclass sse3_replicate_dfp<string OpcodeStr> {
4886 let neverHasSideEffects = 1 in
4887 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4888 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4889 [], IIC_SSE_MOV_LH>;
4890 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4891 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4894 (scalar_to_vector (loadf64 addr:$src)))))],
4898 // FIXME: Merge with above classe when there're patterns for the ymm version
4899 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4900 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4901 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4902 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4903 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4904 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4907 (scalar_to_vector (loadf64 addr:$src)))))]>;
4910 let Predicates = [HasAVX] in {
4911 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4912 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4915 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4917 let Predicates = [HasAVX] in {
4918 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4919 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4920 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4921 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4922 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4923 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4924 def : Pat<(X86Movddup (bc_v2f64
4925 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4926 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4929 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4930 (VMOVDDUPYrm addr:$src)>;
4931 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4932 (VMOVDDUPYrm addr:$src)>;
4933 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4934 (VMOVDDUPYrm addr:$src)>;
4935 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4936 (VMOVDDUPYrr VR256:$src)>;
4939 let Predicates = [HasSSE3] in {
4940 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4941 (MOVDDUPrm addr:$src)>;
4942 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4943 (MOVDDUPrm addr:$src)>;
4944 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4945 (MOVDDUPrm addr:$src)>;
4946 def : Pat<(X86Movddup (bc_v2f64
4947 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4948 (MOVDDUPrm addr:$src)>;
4951 //===---------------------------------------------------------------------===//
4952 // SSE3 - Move Unaligned Integer
4953 //===---------------------------------------------------------------------===//
4955 let Predicates = [HasAVX] in {
4956 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4957 "vlddqu\t{$src, $dst|$dst, $src}",
4958 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4959 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4960 "vlddqu\t{$src, $dst|$dst, $src}",
4961 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4963 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4964 "lddqu\t{$src, $dst|$dst, $src}",
4965 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4968 //===---------------------------------------------------------------------===//
4969 // SSE3 - Arithmetic
4970 //===---------------------------------------------------------------------===//
4972 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4973 X86MemOperand x86memop, OpndItins itins,
4975 def rr : I<0xD0, MRMSrcReg,
4976 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4978 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4980 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4981 def rm : I<0xD0, MRMSrcMem,
4982 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4984 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4985 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4986 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4989 let Predicates = [HasAVX] in {
4990 let ExeDomain = SSEPackedSingle in {
4991 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4992 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4993 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4994 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4996 let ExeDomain = SSEPackedDouble in {
4997 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4998 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4999 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5000 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5003 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5004 let ExeDomain = SSEPackedSingle in
5005 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5006 f128mem, SSE_ALU_F32P>, TB, XD;
5007 let ExeDomain = SSEPackedDouble in
5008 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5009 f128mem, SSE_ALU_F64P>, TB, OpSize;
5012 //===---------------------------------------------------------------------===//
5013 // SSE3 Instructions
5014 //===---------------------------------------------------------------------===//
5017 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5018 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5019 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5021 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5022 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5023 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5025 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5027 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5029 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5030 IIC_SSE_HADDSUB_RM>;
5032 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5033 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5034 def rr : S3I<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 : S3I<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>;
5048 let Predicates = [HasAVX] in {
5049 let ExeDomain = SSEPackedSingle in {
5050 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5051 X86fhadd, 0>, VEX_4V;
5052 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5053 X86fhsub, 0>, VEX_4V;
5054 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5055 X86fhadd, 0>, VEX_4V;
5056 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5057 X86fhsub, 0>, VEX_4V;
5059 let ExeDomain = SSEPackedDouble in {
5060 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5061 X86fhadd, 0>, VEX_4V;
5062 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5063 X86fhsub, 0>, VEX_4V;
5064 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5065 X86fhadd, 0>, VEX_4V;
5066 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5067 X86fhsub, 0>, VEX_4V;
5071 let Constraints = "$src1 = $dst" in {
5072 let ExeDomain = SSEPackedSingle in {
5073 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5074 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5076 let ExeDomain = SSEPackedDouble in {
5077 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5078 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5082 //===---------------------------------------------------------------------===//
5083 // SSSE3 - Packed Absolute Instructions
5084 //===---------------------------------------------------------------------===//
5087 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5088 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5089 Intrinsic IntId128> {
5090 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5092 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5093 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5096 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5098 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5101 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5105 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5106 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5107 Intrinsic IntId256> {
5108 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5110 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5111 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5114 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5116 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5119 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5122 let Predicates = [HasAVX] in {
5123 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5124 int_x86_ssse3_pabs_b_128>, VEX;
5125 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5126 int_x86_ssse3_pabs_w_128>, VEX;
5127 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5128 int_x86_ssse3_pabs_d_128>, VEX;
5131 let Predicates = [HasAVX2] in {
5132 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5133 int_x86_avx2_pabs_b>, VEX;
5134 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5135 int_x86_avx2_pabs_w>, VEX;
5136 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5137 int_x86_avx2_pabs_d>, VEX;
5140 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5141 int_x86_ssse3_pabs_b_128>;
5142 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5143 int_x86_ssse3_pabs_w_128>;
5144 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5145 int_x86_ssse3_pabs_d_128>;
5147 //===---------------------------------------------------------------------===//
5148 // SSSE3 - Packed Binary Operator Instructions
5149 //===---------------------------------------------------------------------===//
5151 def SSE_PHADDSUBD : OpndItins<
5152 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5154 def SSE_PHADDSUBSW : OpndItins<
5155 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5157 def SSE_PHADDSUBW : OpndItins<
5158 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5160 def SSE_PSHUFB : OpndItins<
5161 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5163 def SSE_PSIGN : OpndItins<
5164 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5166 def SSE_PMULHRSW : OpndItins<
5167 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5170 /// SS3I_binop_rm - Simple SSSE3 bin op
5171 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5172 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5173 X86MemOperand x86memop, OpndItins itins,
5175 let isCommutable = 1 in
5176 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5177 (ins RC:$src1, RC:$src2),
5179 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5180 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5181 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5183 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5184 (ins RC:$src1, x86memop:$src2),
5186 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5187 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5189 (OpVT (OpNode RC:$src1,
5190 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5193 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5194 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5195 Intrinsic IntId128, OpndItins itins,
5197 let isCommutable = 1 in
5198 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5199 (ins VR128:$src1, VR128:$src2),
5201 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5202 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5203 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5205 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5206 (ins VR128:$src1, i128mem:$src2),
5208 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5209 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5211 (IntId128 VR128:$src1,
5212 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5215 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5216 Intrinsic IntId256> {
5217 let isCommutable = 1 in
5218 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5219 (ins VR256:$src1, VR256:$src2),
5220 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5221 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5223 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5224 (ins VR256:$src1, i256mem:$src2),
5225 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5227 (IntId256 VR256:$src1,
5228 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5231 let ImmT = NoImm, Predicates = [HasAVX] in {
5232 let isCommutable = 0 in {
5233 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5234 memopv2i64, i128mem,
5235 SSE_PHADDSUBW, 0>, VEX_4V;
5236 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5237 memopv2i64, i128mem,
5238 SSE_PHADDSUBD, 0>, VEX_4V;
5239 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5240 memopv2i64, i128mem,
5241 SSE_PHADDSUBW, 0>, VEX_4V;
5242 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5243 memopv2i64, i128mem,
5244 SSE_PHADDSUBD, 0>, VEX_4V;
5245 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5246 memopv2i64, i128mem,
5247 SSE_PSIGN, 0>, VEX_4V;
5248 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5249 memopv2i64, i128mem,
5250 SSE_PSIGN, 0>, VEX_4V;
5251 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5252 memopv2i64, i128mem,
5253 SSE_PSIGN, 0>, VEX_4V;
5254 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5255 memopv2i64, i128mem,
5256 SSE_PSHUFB, 0>, VEX_4V;
5257 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5258 int_x86_ssse3_phadd_sw_128,
5259 SSE_PHADDSUBSW, 0>, VEX_4V;
5260 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5261 int_x86_ssse3_phsub_sw_128,
5262 SSE_PHADDSUBSW, 0>, VEX_4V;
5263 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5264 int_x86_ssse3_pmadd_ub_sw_128,
5265 SSE_PMADD, 0>, VEX_4V;
5267 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5268 int_x86_ssse3_pmul_hr_sw_128,
5269 SSE_PMULHRSW, 0>, VEX_4V;
5272 let ImmT = NoImm, Predicates = [HasAVX2] in {
5273 let isCommutable = 0 in {
5274 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5275 memopv4i64, i256mem,
5276 SSE_PHADDSUBW, 0>, VEX_4V;
5277 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5278 memopv4i64, i256mem,
5279 SSE_PHADDSUBW, 0>, VEX_4V;
5280 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5281 memopv4i64, i256mem,
5282 SSE_PHADDSUBW, 0>, VEX_4V;
5283 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5284 memopv4i64, i256mem,
5285 SSE_PHADDSUBW, 0>, VEX_4V;
5286 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5287 memopv4i64, i256mem,
5288 SSE_PHADDSUBW, 0>, VEX_4V;
5289 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5290 memopv4i64, i256mem,
5291 SSE_PHADDSUBW, 0>, VEX_4V;
5292 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5293 memopv4i64, i256mem,
5294 SSE_PHADDSUBW, 0>, VEX_4V;
5295 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5296 memopv4i64, i256mem,
5297 SSE_PHADDSUBW, 0>, VEX_4V;
5298 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5299 int_x86_avx2_phadd_sw>, VEX_4V;
5300 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5301 int_x86_avx2_phsub_sw>, VEX_4V;
5302 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5303 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5305 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5306 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5309 // None of these have i8 immediate fields.
5310 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5311 let isCommutable = 0 in {
5312 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5313 memopv2i64, i128mem, SSE_PHADDSUBW>;
5314 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5315 memopv2i64, i128mem, SSE_PHADDSUBD>;
5316 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5317 memopv2i64, i128mem, SSE_PHADDSUBW>;
5318 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5319 memopv2i64, i128mem, SSE_PHADDSUBD>;
5320 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5321 memopv2i64, i128mem, SSE_PSIGN>;
5322 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5323 memopv2i64, i128mem, SSE_PSIGN>;
5324 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5325 memopv2i64, i128mem, SSE_PSIGN>;
5326 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5327 memopv2i64, i128mem, SSE_PSHUFB>;
5328 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5329 int_x86_ssse3_phadd_sw_128,
5331 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5332 int_x86_ssse3_phsub_sw_128,
5334 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5335 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5337 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5338 int_x86_ssse3_pmul_hr_sw_128,
5342 //===---------------------------------------------------------------------===//
5343 // SSSE3 - Packed Align Instruction Patterns
5344 //===---------------------------------------------------------------------===//
5346 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5347 let neverHasSideEffects = 1 in {
5348 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5349 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5351 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5353 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5354 [], IIC_SSE_PALIGNR>, OpSize;
5356 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5357 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5359 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5361 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5362 [], IIC_SSE_PALIGNR>, OpSize;
5366 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5367 let neverHasSideEffects = 1 in {
5368 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5369 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5371 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5374 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5375 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5377 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5382 let Predicates = [HasAVX] in
5383 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5384 let Predicates = [HasAVX2] in
5385 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5386 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5387 defm PALIGN : ssse3_palign<"palignr">;
5389 let Predicates = [HasAVX2] in {
5390 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5391 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5392 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5393 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5394 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5395 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5396 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5397 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5400 let Predicates = [HasAVX] in {
5401 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5402 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5403 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5404 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5405 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5406 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5407 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5408 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5411 let Predicates = [HasSSSE3] in {
5412 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5413 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5414 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5415 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5416 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5417 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5418 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5419 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5422 //===---------------------------------------------------------------------===//
5423 // SSSE3 - Thread synchronization
5424 //===---------------------------------------------------------------------===//
5426 let usesCustomInserter = 1 in {
5427 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5428 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5429 Requires<[HasSSE3]>;
5430 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5431 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5432 Requires<[HasSSE3]>;
5435 let Uses = [EAX, ECX, EDX] in
5436 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5437 TB, Requires<[HasSSE3]>;
5438 let Uses = [ECX, EAX] in
5439 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5440 TB, Requires<[HasSSE3]>;
5442 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5443 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5445 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5446 Requires<[In32BitMode]>;
5447 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5448 Requires<[In64BitMode]>;
5450 //===----------------------------------------------------------------------===//
5451 // SSE4.1 - Packed Move with Sign/Zero Extend
5452 //===----------------------------------------------------------------------===//
5454 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5455 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5456 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5457 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5459 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5460 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5462 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5466 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5468 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5469 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5470 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5472 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5473 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5474 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5477 let Predicates = [HasAVX] in {
5478 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5480 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5482 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5484 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5486 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5488 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5492 let Predicates = [HasAVX2] in {
5493 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5494 int_x86_avx2_pmovsxbw>, VEX;
5495 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5496 int_x86_avx2_pmovsxwd>, VEX;
5497 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5498 int_x86_avx2_pmovsxdq>, VEX;
5499 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5500 int_x86_avx2_pmovzxbw>, VEX;
5501 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5502 int_x86_avx2_pmovzxwd>, VEX;
5503 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5504 int_x86_avx2_pmovzxdq>, VEX;
5507 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5508 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5509 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5510 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5511 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5512 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5514 let Predicates = [HasAVX] in {
5515 // Common patterns involving scalar load.
5516 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5517 (VPMOVSXBWrm addr:$src)>;
5518 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5519 (VPMOVSXBWrm addr:$src)>;
5521 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5522 (VPMOVSXWDrm addr:$src)>;
5523 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5524 (VPMOVSXWDrm addr:$src)>;
5526 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5527 (VPMOVSXDQrm addr:$src)>;
5528 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5529 (VPMOVSXDQrm addr:$src)>;
5531 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5532 (VPMOVZXBWrm addr:$src)>;
5533 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5534 (VPMOVZXBWrm addr:$src)>;
5536 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5537 (VPMOVZXWDrm addr:$src)>;
5538 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5539 (VPMOVZXWDrm addr:$src)>;
5541 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5542 (VPMOVZXDQrm addr:$src)>;
5543 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5544 (VPMOVZXDQrm addr:$src)>;
5547 let Predicates = [HasSSE41] in {
5548 // Common patterns involving scalar load.
5549 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5550 (PMOVSXBWrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5552 (PMOVSXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5555 (PMOVSXWDrm addr:$src)>;
5556 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5557 (PMOVSXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5560 (PMOVSXDQrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5562 (PMOVSXDQrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5565 (PMOVZXBWrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5567 (PMOVZXBWrm addr:$src)>;
5569 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5570 (PMOVZXWDrm addr:$src)>;
5571 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5572 (PMOVZXWDrm addr:$src)>;
5574 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5575 (PMOVZXDQrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5577 (PMOVZXDQrm addr:$src)>;
5580 let Predicates = [HasAVX2] in {
5581 let AddedComplexity = 15 in {
5582 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5583 (VPMOVZXDQYrr VR128:$src)>;
5584 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5585 (VPMOVZXWDYrr VR128:$src)>;
5588 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5589 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5592 let Predicates = [HasAVX] in {
5593 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5594 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5597 let Predicates = [HasSSE41] in {
5598 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5599 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5603 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5604 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5605 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5606 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5608 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5611 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5615 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5617 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5618 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5619 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5621 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5622 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5624 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5628 let Predicates = [HasAVX] in {
5629 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5631 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5633 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5635 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5639 let Predicates = [HasAVX2] in {
5640 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5641 int_x86_avx2_pmovsxbd>, VEX;
5642 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5643 int_x86_avx2_pmovsxwq>, VEX;
5644 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5645 int_x86_avx2_pmovzxbd>, VEX;
5646 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5647 int_x86_avx2_pmovzxwq>, VEX;
5650 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5651 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5652 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5653 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5655 let Predicates = [HasAVX] in {
5656 // Common patterns involving scalar load
5657 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5658 (VPMOVSXBDrm addr:$src)>;
5659 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5660 (VPMOVSXWQrm addr:$src)>;
5662 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5663 (VPMOVZXBDrm addr:$src)>;
5664 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5665 (VPMOVZXWQrm addr:$src)>;
5668 let Predicates = [HasSSE41] in {
5669 // Common patterns involving scalar load
5670 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5671 (PMOVSXBDrm addr:$src)>;
5672 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5673 (PMOVSXWQrm addr:$src)>;
5675 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5676 (PMOVZXBDrm addr:$src)>;
5677 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5678 (PMOVZXWQrm addr:$src)>;
5681 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5682 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5683 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5684 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5686 // Expecting a i16 load any extended to i32 value.
5687 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5688 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5689 [(set VR128:$dst, (IntId (bitconvert
5690 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5694 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5696 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5698 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5700 // Expecting a i16 load any extended to i32 value.
5701 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5702 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5703 [(set VR256:$dst, (IntId (bitconvert
5704 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5708 let Predicates = [HasAVX] in {
5709 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5711 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5714 let Predicates = [HasAVX2] in {
5715 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5716 int_x86_avx2_pmovsxbq>, VEX;
5717 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5718 int_x86_avx2_pmovzxbq>, VEX;
5720 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5721 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5723 let Predicates = [HasAVX] in {
5724 // Common patterns involving scalar load
5725 def : Pat<(int_x86_sse41_pmovsxbq
5726 (bitconvert (v4i32 (X86vzmovl
5727 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5728 (VPMOVSXBQrm addr:$src)>;
5730 def : Pat<(int_x86_sse41_pmovzxbq
5731 (bitconvert (v4i32 (X86vzmovl
5732 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5733 (VPMOVZXBQrm addr:$src)>;
5736 let Predicates = [HasSSE41] in {
5737 // Common patterns involving scalar load
5738 def : Pat<(int_x86_sse41_pmovsxbq
5739 (bitconvert (v4i32 (X86vzmovl
5740 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5741 (PMOVSXBQrm addr:$src)>;
5743 def : Pat<(int_x86_sse41_pmovzxbq
5744 (bitconvert (v4i32 (X86vzmovl
5745 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5746 (PMOVZXBQrm addr:$src)>;
5749 //===----------------------------------------------------------------------===//
5750 // SSE4.1 - Extract Instructions
5751 //===----------------------------------------------------------------------===//
5753 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5754 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5755 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5756 (ins VR128:$src1, i32i8imm:$src2),
5757 !strconcat(OpcodeStr,
5758 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5759 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5761 let neverHasSideEffects = 1, mayStore = 1 in
5762 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5763 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5764 !strconcat(OpcodeStr,
5765 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5768 // There's an AssertZext in the way of writing the store pattern
5769 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5772 let Predicates = [HasAVX] in {
5773 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5774 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5775 (ins VR128:$src1, i32i8imm:$src2),
5776 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5779 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5782 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5783 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5784 let neverHasSideEffects = 1, mayStore = 1 in
5785 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5786 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5787 !strconcat(OpcodeStr,
5788 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5791 // There's an AssertZext in the way of writing the store pattern
5792 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5795 let Predicates = [HasAVX] in
5796 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5798 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5801 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5802 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5803 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5804 (ins VR128:$src1, i32i8imm:$src2),
5805 !strconcat(OpcodeStr,
5806 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5808 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5809 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5810 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5811 !strconcat(OpcodeStr,
5812 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5813 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5814 addr:$dst)]>, OpSize;
5817 let Predicates = [HasAVX] in
5818 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5820 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5822 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5823 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5824 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5825 (ins VR128:$src1, i32i8imm:$src2),
5826 !strconcat(OpcodeStr,
5827 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5829 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5830 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5831 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5832 !strconcat(OpcodeStr,
5833 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5834 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5835 addr:$dst)]>, OpSize, REX_W;
5838 let Predicates = [HasAVX] in
5839 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5841 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5843 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5845 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5846 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5847 (ins VR128:$src1, i32i8imm:$src2),
5848 !strconcat(OpcodeStr,
5849 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5851 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5853 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5854 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5855 !strconcat(OpcodeStr,
5856 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5857 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5858 addr:$dst)]>, OpSize;
5861 let ExeDomain = SSEPackedSingle in {
5862 let Predicates = [HasAVX] in {
5863 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5864 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5865 (ins VR128:$src1, i32i8imm:$src2),
5866 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5869 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5872 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5873 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5876 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5878 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5881 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5882 Requires<[HasSSE41]>;
5884 //===----------------------------------------------------------------------===//
5885 // SSE4.1 - Insert Instructions
5886 //===----------------------------------------------------------------------===//
5888 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5889 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5890 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5892 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5894 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5896 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5897 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5898 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5900 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5902 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5904 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5905 imm:$src3))]>, OpSize;
5908 let Predicates = [HasAVX] in
5909 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5910 let Constraints = "$src1 = $dst" in
5911 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5913 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5914 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5915 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5917 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5919 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5921 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5923 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5924 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5926 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5928 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5930 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5931 imm:$src3)))]>, OpSize;
5934 let Predicates = [HasAVX] in
5935 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5936 let Constraints = "$src1 = $dst" in
5937 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5939 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5940 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5941 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5943 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5945 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5947 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5949 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5950 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5952 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5954 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5956 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5957 imm:$src3)))]>, OpSize;
5960 let Predicates = [HasAVX] in
5961 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5962 let Constraints = "$src1 = $dst" in
5963 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5965 // insertps has a few different modes, there's the first two here below which
5966 // are optimized inserts that won't zero arbitrary elements in the destination
5967 // vector. The next one matches the intrinsic and could zero arbitrary elements
5968 // in the target vector.
5969 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5970 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5971 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5973 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5975 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5977 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5979 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5980 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5982 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5984 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5986 (X86insrtps VR128:$src1,
5987 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5988 imm:$src3))]>, OpSize;
5991 let ExeDomain = SSEPackedSingle in {
5992 let Predicates = [HasAVX] in
5993 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5994 let Constraints = "$src1 = $dst" in
5995 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5998 //===----------------------------------------------------------------------===//
5999 // SSE4.1 - Round Instructions
6000 //===----------------------------------------------------------------------===//
6002 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6003 X86MemOperand x86memop, RegisterClass RC,
6004 PatFrag mem_frag32, PatFrag mem_frag64,
6005 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6006 let ExeDomain = SSEPackedSingle in {
6007 // Intrinsic operation, reg.
6008 // Vector intrinsic operation, reg
6009 def PSr : SS4AIi8<opcps, MRMSrcReg,
6010 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6011 !strconcat(OpcodeStr,
6012 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6013 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6016 // Vector intrinsic operation, mem
6017 def PSm : SS4AIi8<opcps, MRMSrcMem,
6018 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6019 !strconcat(OpcodeStr,
6020 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6022 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6024 } // ExeDomain = SSEPackedSingle
6026 let ExeDomain = SSEPackedDouble in {
6027 // Vector intrinsic operation, reg
6028 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6029 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6030 !strconcat(OpcodeStr,
6031 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6032 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6035 // Vector intrinsic operation, mem
6036 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6037 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6038 !strconcat(OpcodeStr,
6039 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6041 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6043 } // ExeDomain = SSEPackedDouble
6046 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6049 Intrinsic F64Int, bit Is2Addr = 1> {
6050 let ExeDomain = GenericDomain in {
6052 def SSr : SS4AIi8<opcss, MRMSrcReg,
6053 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6055 !strconcat(OpcodeStr,
6056 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6057 !strconcat(OpcodeStr,
6058 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6061 // Intrinsic operation, reg.
6062 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6063 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6065 !strconcat(OpcodeStr,
6066 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6067 !strconcat(OpcodeStr,
6068 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6069 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6072 // Intrinsic operation, mem.
6073 def SSm : SS4AIi8<opcss, MRMSrcMem,
6074 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6076 !strconcat(OpcodeStr,
6077 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6078 !strconcat(OpcodeStr,
6079 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6081 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6085 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6086 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6088 !strconcat(OpcodeStr,
6089 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6090 !strconcat(OpcodeStr,
6091 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6094 // Intrinsic operation, reg.
6095 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6096 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6098 !strconcat(OpcodeStr,
6099 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6100 !strconcat(OpcodeStr,
6101 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6102 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6105 // Intrinsic operation, mem.
6106 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6107 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6109 !strconcat(OpcodeStr,
6110 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6111 !strconcat(OpcodeStr,
6112 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6114 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6116 } // ExeDomain = GenericDomain
6119 // FP round - roundss, roundps, roundsd, roundpd
6120 let Predicates = [HasAVX] in {
6122 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6123 memopv4f32, memopv2f64,
6124 int_x86_sse41_round_ps,
6125 int_x86_sse41_round_pd>, VEX;
6126 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6127 memopv8f32, memopv4f64,
6128 int_x86_avx_round_ps_256,
6129 int_x86_avx_round_pd_256>, VEX;
6130 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6131 int_x86_sse41_round_ss,
6132 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6134 def : Pat<(ffloor FR32:$src),
6135 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6136 def : Pat<(f64 (ffloor FR64:$src)),
6137 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6138 def : Pat<(f32 (fnearbyint FR32:$src)),
6139 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6140 def : Pat<(f64 (fnearbyint FR64:$src)),
6141 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6142 def : Pat<(f32 (fceil FR32:$src)),
6143 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6144 def : Pat<(f64 (fceil FR64:$src)),
6145 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6146 def : Pat<(f32 (frint FR32:$src)),
6147 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6148 def : Pat<(f64 (frint FR64:$src)),
6149 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6150 def : Pat<(f32 (ftrunc FR32:$src)),
6151 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6152 def : Pat<(f64 (ftrunc FR64:$src)),
6153 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6156 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6157 memopv4f32, memopv2f64,
6158 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6159 let Constraints = "$src1 = $dst" in
6160 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6161 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6163 def : Pat<(ffloor FR32:$src),
6164 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6165 def : Pat<(f64 (ffloor FR64:$src)),
6166 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6167 def : Pat<(f32 (fnearbyint FR32:$src)),
6168 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6169 def : Pat<(f64 (fnearbyint FR64:$src)),
6170 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6171 def : Pat<(f32 (fceil FR32:$src)),
6172 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6173 def : Pat<(f64 (fceil FR64:$src)),
6174 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6175 def : Pat<(f32 (frint FR32:$src)),
6176 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6177 def : Pat<(f64 (frint FR64:$src)),
6178 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6179 def : Pat<(f32 (ftrunc FR32:$src)),
6180 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6181 def : Pat<(f64 (ftrunc FR64:$src)),
6182 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6184 //===----------------------------------------------------------------------===//
6185 // SSE4.1 - Packed Bit Test
6186 //===----------------------------------------------------------------------===//
6188 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6189 // the intel intrinsic that corresponds to this.
6190 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6191 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6192 "vptest\t{$src2, $src1|$src1, $src2}",
6193 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6195 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6196 "vptest\t{$src2, $src1|$src1, $src2}",
6197 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6200 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6201 "vptest\t{$src2, $src1|$src1, $src2}",
6202 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6204 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6205 "vptest\t{$src2, $src1|$src1, $src2}",
6206 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6210 let Defs = [EFLAGS] in {
6211 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6212 "ptest\t{$src2, $src1|$src1, $src2}",
6213 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6215 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6216 "ptest\t{$src2, $src1|$src1, $src2}",
6217 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6221 // The bit test instructions below are AVX only
6222 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6223 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6224 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6225 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6226 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6227 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6228 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6229 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6233 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6234 let ExeDomain = SSEPackedSingle in {
6235 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6236 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6238 let ExeDomain = SSEPackedDouble in {
6239 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6240 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6244 //===----------------------------------------------------------------------===//
6245 // SSE4.1 - Misc Instructions
6246 //===----------------------------------------------------------------------===//
6248 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6249 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6250 "popcnt{w}\t{$src, $dst|$dst, $src}",
6251 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6253 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6254 "popcnt{w}\t{$src, $dst|$dst, $src}",
6255 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6256 (implicit EFLAGS)]>, OpSize, XS;
6258 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6259 "popcnt{l}\t{$src, $dst|$dst, $src}",
6260 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6262 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6263 "popcnt{l}\t{$src, $dst|$dst, $src}",
6264 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6265 (implicit EFLAGS)]>, XS;
6267 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6268 "popcnt{q}\t{$src, $dst|$dst, $src}",
6269 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6271 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6272 "popcnt{q}\t{$src, $dst|$dst, $src}",
6273 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6274 (implicit EFLAGS)]>, XS;
6279 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6280 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6281 Intrinsic IntId128> {
6282 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6284 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6285 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6286 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6288 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6291 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6294 let Predicates = [HasAVX] in
6295 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6296 int_x86_sse41_phminposuw>, VEX;
6297 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6298 int_x86_sse41_phminposuw>;
6300 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6301 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6302 Intrinsic IntId128, bit Is2Addr = 1> {
6303 let isCommutable = 1 in
6304 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6305 (ins VR128:$src1, VR128:$src2),
6307 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6308 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6309 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6310 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6311 (ins VR128:$src1, i128mem:$src2),
6313 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6314 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6316 (IntId128 VR128:$src1,
6317 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6320 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6321 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6322 Intrinsic IntId256> {
6323 let isCommutable = 1 in
6324 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6325 (ins VR256:$src1, VR256:$src2),
6326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6327 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6328 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6329 (ins VR256:$src1, i256mem:$src2),
6330 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6332 (IntId256 VR256:$src1,
6333 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6336 let Predicates = [HasAVX] in {
6337 let isCommutable = 0 in
6338 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6340 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6342 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6344 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6346 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6348 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6350 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6352 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6354 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6356 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6360 let Predicates = [HasAVX2] in {
6361 let isCommutable = 0 in
6362 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6363 int_x86_avx2_packusdw>, VEX_4V;
6364 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6365 int_x86_avx2_pmins_b>, VEX_4V;
6366 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6367 int_x86_avx2_pmins_d>, VEX_4V;
6368 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6369 int_x86_avx2_pminu_d>, VEX_4V;
6370 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6371 int_x86_avx2_pminu_w>, VEX_4V;
6372 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6373 int_x86_avx2_pmaxs_b>, VEX_4V;
6374 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6375 int_x86_avx2_pmaxs_d>, VEX_4V;
6376 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6377 int_x86_avx2_pmaxu_d>, VEX_4V;
6378 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6379 int_x86_avx2_pmaxu_w>, VEX_4V;
6380 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6381 int_x86_avx2_pmul_dq>, VEX_4V;
6384 let Constraints = "$src1 = $dst" in {
6385 let isCommutable = 0 in
6386 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6387 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6388 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6389 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6390 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6391 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6392 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6393 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6394 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6395 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6398 /// SS48I_binop_rm - Simple SSE41 binary operator.
6399 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6400 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6401 X86MemOperand x86memop, bit Is2Addr = 1> {
6402 let isCommutable = 1 in
6403 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6404 (ins RC:$src1, RC:$src2),
6406 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6408 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6409 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6410 (ins RC:$src1, x86memop:$src2),
6412 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6413 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6415 (OpVT (OpNode RC:$src1,
6416 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6419 let Predicates = [HasAVX] in {
6420 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6421 memopv2i64, i128mem, 0>, VEX_4V;
6422 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6423 memopv2i64, i128mem, 0>, VEX_4V;
6425 let Predicates = [HasAVX2] in {
6426 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6427 memopv4i64, i256mem, 0>, VEX_4V;
6428 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6429 memopv4i64, i256mem, 0>, VEX_4V;
6432 let Constraints = "$src1 = $dst" in {
6433 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6434 memopv2i64, i128mem>;
6435 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6436 memopv2i64, i128mem>;
6439 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6440 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6441 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6442 X86MemOperand x86memop, bit Is2Addr = 1> {
6443 let isCommutable = 1 in
6444 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6445 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6447 !strconcat(OpcodeStr,
6448 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6449 !strconcat(OpcodeStr,
6450 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6451 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6453 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6454 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6456 !strconcat(OpcodeStr,
6457 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6458 !strconcat(OpcodeStr,
6459 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6462 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6466 let Predicates = [HasAVX] in {
6467 let isCommutable = 0 in {
6468 let ExeDomain = SSEPackedSingle in {
6469 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6470 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6471 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6472 int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
6474 let ExeDomain = SSEPackedDouble in {
6475 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6476 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6477 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6478 int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
6480 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6481 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6482 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6483 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6485 let ExeDomain = SSEPackedSingle in
6486 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6487 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6488 let ExeDomain = SSEPackedDouble in
6489 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6490 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6491 let ExeDomain = SSEPackedSingle in
6492 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6493 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6496 let Predicates = [HasAVX2] in {
6497 let isCommutable = 0 in {
6498 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6499 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6500 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6501 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6505 let Constraints = "$src1 = $dst" in {
6506 let isCommutable = 0 in {
6507 let ExeDomain = SSEPackedSingle in
6508 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6509 VR128, memopv4f32, f128mem>;
6510 let ExeDomain = SSEPackedDouble in
6511 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6512 VR128, memopv2f64, f128mem>;
6513 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6514 VR128, memopv2i64, i128mem>;
6515 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6516 VR128, memopv2i64, i128mem>;
6518 let ExeDomain = SSEPackedSingle in
6519 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6520 VR128, memopv4f32, f128mem>;
6521 let ExeDomain = SSEPackedDouble in
6522 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6523 VR128, memopv2f64, f128mem>;
6526 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6527 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6528 RegisterClass RC, X86MemOperand x86memop,
6529 PatFrag mem_frag, Intrinsic IntId> {
6530 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6531 (ins RC:$src1, RC:$src2, RC:$src3),
6532 !strconcat(OpcodeStr,
6533 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6534 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6535 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6537 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6538 (ins RC:$src1, x86memop:$src2, RC:$src3),
6539 !strconcat(OpcodeStr,
6540 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6542 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6544 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6547 let Predicates = [HasAVX] in {
6548 let ExeDomain = SSEPackedDouble in {
6549 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6550 memopv2f64, int_x86_sse41_blendvpd>;
6551 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6552 memopv4f64, int_x86_avx_blendv_pd_256>;
6553 } // ExeDomain = SSEPackedDouble
6554 let ExeDomain = SSEPackedSingle in {
6555 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6556 memopv4f32, int_x86_sse41_blendvps>;
6557 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6558 memopv8f32, int_x86_avx_blendv_ps_256>;
6559 } // ExeDomain = SSEPackedSingle
6560 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6561 memopv2i64, int_x86_sse41_pblendvb>;
6564 let Predicates = [HasAVX2] in {
6565 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6566 memopv4i64, int_x86_avx2_pblendvb>;
6569 let Predicates = [HasAVX] in {
6570 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6571 (v16i8 VR128:$src2))),
6572 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6573 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6574 (v4i32 VR128:$src2))),
6575 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6576 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6577 (v4f32 VR128:$src2))),
6578 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6579 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6580 (v2i64 VR128:$src2))),
6581 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6582 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6583 (v2f64 VR128:$src2))),
6584 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6585 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6586 (v8i32 VR256:$src2))),
6587 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6588 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6589 (v8f32 VR256:$src2))),
6590 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6591 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6592 (v4i64 VR256:$src2))),
6593 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6594 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6595 (v4f64 VR256:$src2))),
6596 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6598 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6600 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6601 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6603 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6605 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6607 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6608 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6610 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6611 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6613 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6616 let Predicates = [HasAVX2] in {
6617 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6618 (v32i8 VR256:$src2))),
6619 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6620 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6622 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6625 /// SS41I_ternary_int - SSE 4.1 ternary operator
6626 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6627 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6628 X86MemOperand x86memop, Intrinsic IntId> {
6629 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6630 (ins VR128:$src1, VR128:$src2),
6631 !strconcat(OpcodeStr,
6632 "\t{$src2, $dst|$dst, $src2}"),
6633 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6636 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6637 (ins VR128:$src1, x86memop:$src2),
6638 !strconcat(OpcodeStr,
6639 "\t{$src2, $dst|$dst, $src2}"),
6642 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6646 let ExeDomain = SSEPackedDouble in
6647 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6648 int_x86_sse41_blendvpd>;
6649 let ExeDomain = SSEPackedSingle in
6650 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6651 int_x86_sse41_blendvps>;
6652 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6653 int_x86_sse41_pblendvb>;
6655 // Aliases with the implicit xmm0 argument
6656 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6657 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6658 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6659 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6660 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6661 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6662 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6663 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6664 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6665 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6666 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6667 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6669 let Predicates = [HasSSE41] in {
6670 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6671 (v16i8 VR128:$src2))),
6672 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6673 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6674 (v4i32 VR128:$src2))),
6675 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6676 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6677 (v4f32 VR128:$src2))),
6678 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6679 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6680 (v2i64 VR128:$src2))),
6681 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6682 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6683 (v2f64 VR128:$src2))),
6684 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6686 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6688 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6689 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6691 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6692 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6694 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6698 let Predicates = [HasAVX] in
6699 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6700 "vmovntdqa\t{$src, $dst|$dst, $src}",
6701 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6703 let Predicates = [HasAVX2] in
6704 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6705 "vmovntdqa\t{$src, $dst|$dst, $src}",
6706 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6708 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6709 "movntdqa\t{$src, $dst|$dst, $src}",
6710 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6713 //===----------------------------------------------------------------------===//
6714 // SSE4.2 - Compare Instructions
6715 //===----------------------------------------------------------------------===//
6717 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6718 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6719 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6720 X86MemOperand x86memop, bit Is2Addr = 1> {
6721 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6722 (ins RC:$src1, RC:$src2),
6724 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6725 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6726 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6728 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6729 (ins RC:$src1, x86memop:$src2),
6731 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6732 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6734 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6737 let Predicates = [HasAVX] in
6738 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6739 memopv2i64, i128mem, 0>, VEX_4V;
6741 let Predicates = [HasAVX2] in
6742 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6743 memopv4i64, i256mem, 0>, VEX_4V;
6745 let Constraints = "$src1 = $dst" in
6746 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6747 memopv2i64, i128mem>;
6749 //===----------------------------------------------------------------------===//
6750 // SSE4.2 - String/text Processing Instructions
6751 //===----------------------------------------------------------------------===//
6753 // Packed Compare Implicit Length Strings, Return Mask
6754 multiclass pseudo_pcmpistrm<string asm> {
6755 def REG : PseudoI<(outs VR128:$dst),
6756 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6757 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6759 def MEM : PseudoI<(outs VR128:$dst),
6760 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6761 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6762 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6765 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6766 let AddedComplexity = 1 in
6767 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6768 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6771 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6772 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6773 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6774 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6776 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6777 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6778 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6781 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6782 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6783 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6784 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6786 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6787 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6788 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6791 // Packed Compare Explicit Length Strings, Return Mask
6792 multiclass pseudo_pcmpestrm<string asm> {
6793 def REG : PseudoI<(outs VR128:$dst),
6794 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6795 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6796 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6797 def MEM : PseudoI<(outs VR128:$dst),
6798 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6799 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6800 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6803 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6804 let AddedComplexity = 1 in
6805 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6806 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6809 let Predicates = [HasAVX],
6810 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6811 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6812 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6813 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6815 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6816 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6817 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6820 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6821 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6822 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6823 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6825 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6826 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6827 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6830 // Packed Compare Implicit Length Strings, Return Index
6831 let Defs = [ECX, EFLAGS] in {
6832 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6833 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6834 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6835 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6836 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6837 (implicit EFLAGS)]>, OpSize;
6838 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6839 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6840 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6841 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6842 (implicit EFLAGS)]>, OpSize;
6846 let Predicates = [HasAVX] in {
6847 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6849 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6851 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6853 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6855 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6857 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6861 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6862 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6863 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6864 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6865 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6866 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6868 // Packed Compare Explicit Length Strings, Return Index
6869 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6870 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6871 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6872 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6873 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6874 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6875 (implicit EFLAGS)]>, OpSize;
6876 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6877 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6878 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6880 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6881 (implicit EFLAGS)]>, OpSize;
6885 let Predicates = [HasAVX] in {
6886 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6888 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6890 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6892 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6894 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6896 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6900 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6901 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6902 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6903 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6904 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6905 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6907 //===----------------------------------------------------------------------===//
6908 // SSE4.2 - CRC Instructions
6909 //===----------------------------------------------------------------------===//
6911 // No CRC instructions have AVX equivalents
6913 // crc intrinsic instruction
6914 // This set of instructions are only rm, the only difference is the size
6916 let Constraints = "$src1 = $dst" in {
6917 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6918 (ins GR32:$src1, i8mem:$src2),
6919 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6921 (int_x86_sse42_crc32_32_8 GR32:$src1,
6922 (load addr:$src2)))]>;
6923 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6924 (ins GR32:$src1, GR8:$src2),
6925 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6927 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6928 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6929 (ins GR32:$src1, i16mem:$src2),
6930 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6932 (int_x86_sse42_crc32_32_16 GR32:$src1,
6933 (load addr:$src2)))]>,
6935 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6936 (ins GR32:$src1, GR16:$src2),
6937 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6939 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6941 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6942 (ins GR32:$src1, i32mem:$src2),
6943 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6945 (int_x86_sse42_crc32_32_32 GR32:$src1,
6946 (load addr:$src2)))]>;
6947 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6948 (ins GR32:$src1, GR32:$src2),
6949 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6951 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6952 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6953 (ins GR64:$src1, i8mem:$src2),
6954 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6956 (int_x86_sse42_crc32_64_8 GR64:$src1,
6957 (load addr:$src2)))]>,
6959 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6960 (ins GR64:$src1, GR8:$src2),
6961 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6963 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6965 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6966 (ins GR64:$src1, i64mem:$src2),
6967 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6969 (int_x86_sse42_crc32_64_64 GR64:$src1,
6970 (load addr:$src2)))]>,
6972 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6973 (ins GR64:$src1, GR64:$src2),
6974 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6976 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6980 //===----------------------------------------------------------------------===//
6981 // AES-NI Instructions
6982 //===----------------------------------------------------------------------===//
6984 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6985 Intrinsic IntId128, bit Is2Addr = 1> {
6986 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6987 (ins VR128:$src1, VR128:$src2),
6989 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6990 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6991 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6993 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6994 (ins VR128:$src1, i128mem:$src2),
6996 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6997 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6999 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7002 // Perform One Round of an AES Encryption/Decryption Flow
7003 let Predicates = [HasAVX, HasAES] in {
7004 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7005 int_x86_aesni_aesenc, 0>, VEX_4V;
7006 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7007 int_x86_aesni_aesenclast, 0>, VEX_4V;
7008 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7009 int_x86_aesni_aesdec, 0>, VEX_4V;
7010 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7011 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7014 let Constraints = "$src1 = $dst" in {
7015 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7016 int_x86_aesni_aesenc>;
7017 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7018 int_x86_aesni_aesenclast>;
7019 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7020 int_x86_aesni_aesdec>;
7021 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7022 int_x86_aesni_aesdeclast>;
7025 // Perform the AES InvMixColumn Transformation
7026 let Predicates = [HasAVX, HasAES] in {
7027 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7029 "vaesimc\t{$src1, $dst|$dst, $src1}",
7031 (int_x86_aesni_aesimc VR128:$src1))]>,
7033 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7034 (ins i128mem:$src1),
7035 "vaesimc\t{$src1, $dst|$dst, $src1}",
7036 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7039 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7041 "aesimc\t{$src1, $dst|$dst, $src1}",
7043 (int_x86_aesni_aesimc VR128:$src1))]>,
7045 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7046 (ins i128mem:$src1),
7047 "aesimc\t{$src1, $dst|$dst, $src1}",
7048 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7051 // AES Round Key Generation Assist
7052 let Predicates = [HasAVX, HasAES] in {
7053 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7054 (ins VR128:$src1, i8imm:$src2),
7055 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7057 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7059 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7060 (ins i128mem:$src1, i8imm:$src2),
7061 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7063 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7066 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7067 (ins VR128:$src1, i8imm:$src2),
7068 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7070 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7072 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7073 (ins i128mem:$src1, i8imm:$src2),
7074 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7076 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7079 //===----------------------------------------------------------------------===//
7080 // PCLMUL Instructions
7081 //===----------------------------------------------------------------------===//
7083 // AVX carry-less Multiplication instructions
7084 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7085 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7086 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7088 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7090 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7091 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7092 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7093 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7094 (memopv2i64 addr:$src2), imm:$src3))]>;
7096 // Carry-less Multiplication instructions
7097 let Constraints = "$src1 = $dst" in {
7098 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7099 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7100 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7102 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7104 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7105 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7106 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7107 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7108 (memopv2i64 addr:$src2), imm:$src3))]>;
7109 } // Constraints = "$src1 = $dst"
7112 multiclass pclmul_alias<string asm, int immop> {
7113 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7114 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7116 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7117 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7119 def : InstAlias<!strconcat("vpclmul", asm,
7120 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7121 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7123 def : InstAlias<!strconcat("vpclmul", asm,
7124 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7125 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7127 defm : pclmul_alias<"hqhq", 0x11>;
7128 defm : pclmul_alias<"hqlq", 0x01>;
7129 defm : pclmul_alias<"lqhq", 0x10>;
7130 defm : pclmul_alias<"lqlq", 0x00>;
7132 //===----------------------------------------------------------------------===//
7133 // SSE4A Instructions
7134 //===----------------------------------------------------------------------===//
7136 let Predicates = [HasSSE4A] in {
7138 let Constraints = "$src = $dst" in {
7139 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7140 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7141 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7142 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7143 imm:$idx))]>, TB, OpSize;
7144 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7145 (ins VR128:$src, VR128:$mask),
7146 "extrq\t{$mask, $src|$src, $mask}",
7147 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7148 VR128:$mask))]>, TB, OpSize;
7150 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7151 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7152 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7153 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7154 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7155 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7156 (ins VR128:$src, VR128:$mask),
7157 "insertq\t{$mask, $src|$src, $mask}",
7158 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7159 VR128:$mask))]>, XD;
7162 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7163 "movntss\t{$src, $dst|$dst, $src}",
7164 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7166 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7167 "movntsd\t{$src, $dst|$dst, $src}",
7168 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7171 //===----------------------------------------------------------------------===//
7173 //===----------------------------------------------------------------------===//
7175 //===----------------------------------------------------------------------===//
7176 // VBROADCAST - Load from memory and broadcast to all elements of the
7177 // destination operand
7179 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7180 X86MemOperand x86memop, Intrinsic Int> :
7181 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7182 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7183 [(set RC:$dst, (Int addr:$src))]>, VEX;
7185 // AVX2 adds register forms
7186 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7188 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7189 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7190 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7192 let ExeDomain = SSEPackedSingle in {
7193 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7194 int_x86_avx_vbroadcast_ss>;
7195 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7196 int_x86_avx_vbroadcast_ss_256>;
7198 let ExeDomain = SSEPackedDouble in
7199 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7200 int_x86_avx_vbroadcast_sd_256>;
7201 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7202 int_x86_avx_vbroadcastf128_pd_256>;
7204 let ExeDomain = SSEPackedSingle in {
7205 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7206 int_x86_avx2_vbroadcast_ss_ps>;
7207 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7208 int_x86_avx2_vbroadcast_ss_ps_256>;
7210 let ExeDomain = SSEPackedDouble in
7211 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7212 int_x86_avx2_vbroadcast_sd_pd_256>;
7214 let Predicates = [HasAVX2] in
7215 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7216 int_x86_avx2_vbroadcasti128>;
7218 let Predicates = [HasAVX] in
7219 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7220 (VBROADCASTF128 addr:$src)>;
7223 //===----------------------------------------------------------------------===//
7224 // VINSERTF128 - Insert packed floating-point values
7226 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7227 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7228 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7229 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7232 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7233 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7234 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7238 let Predicates = [HasAVX] in {
7239 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7241 (VINSERTF128rr VR256:$src1, VR128:$src2,
7242 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7243 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7245 (VINSERTF128rr VR256:$src1, VR128:$src2,
7246 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7247 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7249 (VINSERTF128rr VR256:$src1, VR128:$src2,
7250 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7251 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7253 (VINSERTF128rr VR256:$src1, VR128:$src2,
7254 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7255 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7257 (VINSERTF128rr VR256:$src1, VR128:$src2,
7258 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7259 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7261 (VINSERTF128rr VR256:$src1, VR128:$src2,
7262 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7264 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7266 (VINSERTF128rm VR256:$src1, addr:$src2,
7267 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7268 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7270 (VINSERTF128rm VR256:$src1, addr:$src2,
7271 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7272 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7274 (VINSERTF128rm VR256:$src1, addr:$src2,
7275 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7278 //===----------------------------------------------------------------------===//
7279 // VEXTRACTF128 - Extract packed floating-point values
7281 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7282 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7283 (ins VR256:$src1, i8imm:$src2),
7284 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7287 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7288 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7289 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7293 // Extract and store.
7294 let Predicates = [HasAVX] in {
7295 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7296 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7297 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7298 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7299 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7300 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7302 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7303 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7304 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7305 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7306 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7307 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7311 let Predicates = [HasAVX] in {
7312 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7313 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7314 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7315 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7316 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7317 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7319 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7320 (v4f32 (VEXTRACTF128rr
7321 (v8f32 VR256:$src1),
7322 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7323 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7324 (v2f64 (VEXTRACTF128rr
7325 (v4f64 VR256:$src1),
7326 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7327 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7328 (v2i64 (VEXTRACTF128rr
7329 (v4i64 VR256:$src1),
7330 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7331 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7332 (v4i32 (VEXTRACTF128rr
7333 (v8i32 VR256:$src1),
7334 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7335 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7336 (v8i16 (VEXTRACTF128rr
7337 (v16i16 VR256:$src1),
7338 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7339 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7340 (v16i8 (VEXTRACTF128rr
7341 (v32i8 VR256:$src1),
7342 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7345 //===----------------------------------------------------------------------===//
7346 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7348 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7349 Intrinsic IntLd, Intrinsic IntLd256,
7350 Intrinsic IntSt, Intrinsic IntSt256> {
7351 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7352 (ins VR128:$src1, f128mem:$src2),
7353 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7354 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7356 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7357 (ins VR256:$src1, f256mem:$src2),
7358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7359 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7361 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7362 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7363 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7364 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7365 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7366 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7367 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7368 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7371 let ExeDomain = SSEPackedSingle in
7372 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7373 int_x86_avx_maskload_ps,
7374 int_x86_avx_maskload_ps_256,
7375 int_x86_avx_maskstore_ps,
7376 int_x86_avx_maskstore_ps_256>;
7377 let ExeDomain = SSEPackedDouble in
7378 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7379 int_x86_avx_maskload_pd,
7380 int_x86_avx_maskload_pd_256,
7381 int_x86_avx_maskstore_pd,
7382 int_x86_avx_maskstore_pd_256>;
7384 //===----------------------------------------------------------------------===//
7385 // VPERMIL - Permute Single and Double Floating-Point Values
7387 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7388 RegisterClass RC, X86MemOperand x86memop_f,
7389 X86MemOperand x86memop_i, PatFrag i_frag,
7390 Intrinsic IntVar, ValueType vt> {
7391 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7392 (ins RC:$src1, RC:$src2),
7393 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7394 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7395 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7396 (ins RC:$src1, x86memop_i:$src2),
7397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7398 [(set RC:$dst, (IntVar RC:$src1,
7399 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7401 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7402 (ins RC:$src1, i8imm:$src2),
7403 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7404 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7405 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7406 (ins x86memop_f:$src1, i8imm:$src2),
7407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7409 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7412 let ExeDomain = SSEPackedSingle in {
7413 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7414 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7415 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7416 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7418 let ExeDomain = SSEPackedDouble in {
7419 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7420 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7421 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7422 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7425 let Predicates = [HasAVX] in {
7426 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7427 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7428 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7429 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7430 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7432 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7433 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7434 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7436 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7437 (VPERMILPDri VR128:$src1, imm:$imm)>;
7438 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7439 (VPERMILPDmi addr:$src1, imm:$imm)>;
7442 //===----------------------------------------------------------------------===//
7443 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7445 let ExeDomain = SSEPackedSingle in {
7446 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7447 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7448 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7449 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7450 (i8 imm:$src3))))]>, VEX_4V;
7451 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7452 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7453 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7454 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7455 (i8 imm:$src3)))]>, VEX_4V;
7458 let Predicates = [HasAVX] in {
7459 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7460 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7461 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7462 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7463 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7464 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7465 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7466 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7467 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7468 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7470 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7471 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7472 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7473 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7474 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7475 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7476 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7477 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7478 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7479 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7480 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7481 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7482 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7483 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7484 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7485 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7486 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7487 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7490 //===----------------------------------------------------------------------===//
7491 // VZERO - Zero YMM registers
7493 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7494 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7495 // Zero All YMM registers
7496 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7497 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7499 // Zero Upper bits of YMM registers
7500 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7501 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7504 //===----------------------------------------------------------------------===//
7505 // Half precision conversion instructions
7506 //===----------------------------------------------------------------------===//
7507 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7508 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7509 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7510 [(set RC:$dst, (Int VR128:$src))]>,
7512 let neverHasSideEffects = 1, mayLoad = 1 in
7513 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7514 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7517 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7518 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7519 (ins RC:$src1, i32i8imm:$src2),
7520 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7521 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7523 let neverHasSideEffects = 1, mayStore = 1 in
7524 def mr : Ii8<0x1D, MRMDestMem, (outs),
7525 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7526 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7530 let Predicates = [HasAVX, HasF16C] in {
7531 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7532 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7533 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7534 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7537 //===----------------------------------------------------------------------===//
7538 // AVX2 Instructions
7539 //===----------------------------------------------------------------------===//
7541 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7542 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7543 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7544 X86MemOperand x86memop> {
7545 let isCommutable = 1 in
7546 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7547 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7548 !strconcat(OpcodeStr,
7549 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7550 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7552 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7553 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7554 !strconcat(OpcodeStr,
7555 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7558 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7562 let isCommutable = 0 in {
7563 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7564 VR128, memopv2i64, i128mem>;
7565 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7566 VR256, memopv4i64, i256mem>;
7569 //===----------------------------------------------------------------------===//
7570 // VPBROADCAST - Load from memory and broadcast to all elements of the
7571 // destination operand
7573 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7574 X86MemOperand x86memop, PatFrag ld_frag,
7575 Intrinsic Int128, Intrinsic Int256> {
7576 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7577 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7578 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7579 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7580 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7582 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7583 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7584 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7585 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7586 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7587 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7589 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7592 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7593 int_x86_avx2_pbroadcastb_128,
7594 int_x86_avx2_pbroadcastb_256>;
7595 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7596 int_x86_avx2_pbroadcastw_128,
7597 int_x86_avx2_pbroadcastw_256>;
7598 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7599 int_x86_avx2_pbroadcastd_128,
7600 int_x86_avx2_pbroadcastd_256>;
7601 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7602 int_x86_avx2_pbroadcastq_128,
7603 int_x86_avx2_pbroadcastq_256>;
7605 let Predicates = [HasAVX2] in {
7606 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7607 (VPBROADCASTBrm addr:$src)>;
7608 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7609 (VPBROADCASTBYrm addr:$src)>;
7610 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7611 (VPBROADCASTWrm addr:$src)>;
7612 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7613 (VPBROADCASTWYrm addr:$src)>;
7614 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7615 (VPBROADCASTDrm addr:$src)>;
7616 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7617 (VPBROADCASTDYrm addr:$src)>;
7618 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7619 (VPBROADCASTQrm addr:$src)>;
7620 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7621 (VPBROADCASTQYrm addr:$src)>;
7623 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7624 (VPBROADCASTBrr VR128:$src)>;
7625 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7626 (VPBROADCASTBYrr VR128:$src)>;
7627 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7628 (VPBROADCASTWrr VR128:$src)>;
7629 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7630 (VPBROADCASTWYrr VR128:$src)>;
7631 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7632 (VPBROADCASTDrr VR128:$src)>;
7633 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7634 (VPBROADCASTDYrr VR128:$src)>;
7635 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7636 (VPBROADCASTQrr VR128:$src)>;
7637 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7638 (VPBROADCASTQYrr VR128:$src)>;
7639 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7640 (VBROADCASTSSrr VR128:$src)>;
7641 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7642 (VBROADCASTSSYrr VR128:$src)>;
7643 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7644 (VPBROADCASTQrr VR128:$src)>;
7645 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7646 (VBROADCASTSDYrr VR128:$src)>;
7648 // Provide fallback in case the load node that is used in the patterns above
7649 // is used by additional users, which prevents the pattern selection.
7650 let AddedComplexity = 20 in {
7651 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7652 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7653 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7654 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7655 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7656 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7658 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7659 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7660 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7661 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7662 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7663 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7667 // AVX1 broadcast patterns
7668 let Predicates = [HasAVX] in {
7669 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7670 (VBROADCASTSSYrm addr:$src)>;
7671 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7672 (VBROADCASTSDYrm addr:$src)>;
7673 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7674 (VBROADCASTSSYrm addr:$src)>;
7675 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7676 (VBROADCASTSDYrm addr:$src)>;
7677 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7678 (VBROADCASTSSrm addr:$src)>;
7679 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7680 (VBROADCASTSSrm addr:$src)>;
7682 // Provide fallback in case the load node that is used in the patterns above
7683 // is used by additional users, which prevents the pattern selection.
7684 let AddedComplexity = 20 in {
7685 // 128bit broadcasts:
7686 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7687 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7688 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7689 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7690 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7691 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7692 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7693 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7694 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7695 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7697 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7698 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7699 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7700 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7701 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7702 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7703 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7704 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7705 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7706 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7710 //===----------------------------------------------------------------------===//
7711 // VPERM - Permute instructions
7714 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7716 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7717 (ins VR256:$src1, VR256:$src2),
7718 !strconcat(OpcodeStr,
7719 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7721 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7722 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7723 (ins VR256:$src1, i256mem:$src2),
7724 !strconcat(OpcodeStr,
7725 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7727 (OpVT (X86VPermv VR256:$src1,
7728 (bitconvert (mem_frag addr:$src2)))))]>,
7732 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7733 let ExeDomain = SSEPackedSingle in
7734 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7736 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7738 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7739 (ins VR256:$src1, i8imm:$src2),
7740 !strconcat(OpcodeStr,
7741 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7743 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7744 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7745 (ins i256mem:$src1, i8imm:$src2),
7746 !strconcat(OpcodeStr,
7747 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7749 (OpVT (X86VPermi (mem_frag addr:$src1),
7750 (i8 imm:$src2))))]>, VEX;
7753 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7754 let ExeDomain = SSEPackedDouble in
7755 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7757 //===----------------------------------------------------------------------===//
7758 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7760 let AddedComplexity = 1 in {
7761 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7762 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7763 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7764 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7765 (i8 imm:$src3))))]>, VEX_4V;
7766 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7767 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7768 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7769 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7770 (i8 imm:$src3)))]>, VEX_4V;
7773 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7774 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7775 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7776 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7777 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7778 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7779 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7781 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7783 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7784 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7785 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7786 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7787 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7789 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7793 //===----------------------------------------------------------------------===//
7794 // VINSERTI128 - Insert packed integer values
7796 let neverHasSideEffects = 1 in {
7797 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7798 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7799 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7802 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7803 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7804 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7808 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7809 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7811 (VINSERTI128rr VR256:$src1, VR128:$src2,
7812 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7813 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7815 (VINSERTI128rr VR256:$src1, VR128:$src2,
7816 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7817 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7819 (VINSERTI128rr VR256:$src1, VR128:$src2,
7820 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7821 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7823 (VINSERTI128rr VR256:$src1, VR128:$src2,
7824 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7827 //===----------------------------------------------------------------------===//
7828 // VEXTRACTI128 - Extract packed integer values
7830 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7831 (ins VR256:$src1, i8imm:$src2),
7832 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7834 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7836 let neverHasSideEffects = 1, mayStore = 1 in
7837 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7838 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7839 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7841 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7842 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7843 (v2i64 (VEXTRACTI128rr
7844 (v4i64 VR256:$src1),
7845 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7846 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7847 (v4i32 (VEXTRACTI128rr
7848 (v8i32 VR256:$src1),
7849 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7850 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7851 (v8i16 (VEXTRACTI128rr
7852 (v16i16 VR256:$src1),
7853 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7854 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7855 (v16i8 (VEXTRACTI128rr
7856 (v32i8 VR256:$src1),
7857 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7860 //===----------------------------------------------------------------------===//
7861 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7863 multiclass avx2_pmovmask<string OpcodeStr,
7864 Intrinsic IntLd128, Intrinsic IntLd256,
7865 Intrinsic IntSt128, Intrinsic IntSt256> {
7866 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7867 (ins VR128:$src1, i128mem:$src2),
7868 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7869 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7870 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7871 (ins VR256:$src1, i256mem:$src2),
7872 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7873 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7874 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7875 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7876 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7877 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7878 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7879 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7880 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7881 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7884 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7885 int_x86_avx2_maskload_d,
7886 int_x86_avx2_maskload_d_256,
7887 int_x86_avx2_maskstore_d,
7888 int_x86_avx2_maskstore_d_256>;
7889 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7890 int_x86_avx2_maskload_q,
7891 int_x86_avx2_maskload_q_256,
7892 int_x86_avx2_maskstore_q,
7893 int_x86_avx2_maskstore_q_256>, VEX_W;
7896 //===----------------------------------------------------------------------===//
7897 // Variable Bit Shifts
7899 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7900 ValueType vt128, ValueType vt256> {
7901 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7902 (ins VR128:$src1, VR128:$src2),
7903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7905 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7907 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7908 (ins VR128:$src1, i128mem:$src2),
7909 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7911 (vt128 (OpNode VR128:$src1,
7912 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7914 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7915 (ins VR256:$src1, VR256:$src2),
7916 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7918 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7920 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7921 (ins VR256:$src1, i256mem:$src2),
7922 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7924 (vt256 (OpNode VR256:$src1,
7925 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7929 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7930 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7931 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7932 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7933 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
7935 //===----------------------------------------------------------------------===//
7936 // VGATHER - GATHER Operations
7937 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
7938 X86MemOperand memop128, X86MemOperand memop256> {
7939 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
7940 (ins VR128:$src1, memop128:$src2, VR128:$mask),
7941 !strconcat(OpcodeStr,
7942 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7944 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
7945 (ins RC256:$src1, memop256:$src2, RC256:$mask),
7946 !strconcat(OpcodeStr,
7947 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7948 []>, VEX_4VOp3, VEX_L;
7951 let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
7952 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
7953 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
7954 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
7955 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
7956 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
7957 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
7958 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
7959 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;