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:$src)))],
1400 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1401 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1402 string asm, Domain d, OpndItins itins> {
1403 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1404 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1406 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1407 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1411 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1412 X86MemOperand x86memop, string asm> {
1413 let neverHasSideEffects = 1 in {
1414 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1415 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1417 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1418 (ins DstRC:$src1, x86memop:$src),
1419 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1420 } // neverHasSideEffects = 1
1423 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1424 "cvttss2si\t{$src, $dst|$dst, $src}",
1427 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1428 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1430 XS, VEX, VEX_W, VEX_LIG;
1431 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1432 "cvttsd2si\t{$src, $dst|$dst, $src}",
1435 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1436 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1438 XD, VEX, VEX_W, VEX_LIG;
1440 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1441 // register, but the same isn't true when only using memory operands,
1442 // provide other assembly "l" and "q" forms to address this explicitly
1443 // where appropriate to do so.
1444 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1445 XS, VEX_4V, VEX_LIG;
1446 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1447 XS, VEX_4V, VEX_W, VEX_LIG;
1448 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1449 XD, VEX_4V, VEX_LIG;
1450 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1451 XD, VEX_4V, VEX_W, VEX_LIG;
1453 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1454 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1455 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1456 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1458 let Predicates = [HasAVX], AddedComplexity = 1 in {
1459 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1460 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1461 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1462 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1463 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1464 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1466 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1468 def : Pat<(f32 (sint_to_fp GR32:$src)),
1469 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1470 def : Pat<(f32 (sint_to_fp GR64:$src)),
1471 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1472 def : Pat<(f64 (sint_to_fp GR32:$src)),
1473 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1474 def : Pat<(f64 (sint_to_fp GR64:$src)),
1475 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1478 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1479 "cvttss2si\t{$src, $dst|$dst, $src}",
1480 SSE_CVT_SS2SI_32>, XS;
1481 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1482 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1483 SSE_CVT_SS2SI_64>, XS, REX_W;
1484 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1485 "cvttsd2si\t{$src, $dst|$dst, $src}",
1487 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1488 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1489 SSE_CVT_SD2SI>, XD, REX_W;
1490 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1491 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1492 SSE_CVT_Scalar>, XS;
1493 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1494 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1495 SSE_CVT_Scalar>, XS, REX_W;
1496 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1497 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1498 SSE_CVT_Scalar>, XD;
1499 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1500 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1501 SSE_CVT_Scalar>, XD, REX_W;
1503 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1504 // and/or XMM operand(s).
1506 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1507 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1508 string asm, OpndItins itins> {
1509 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1510 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1511 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1512 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1513 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1514 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm>;
1517 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1518 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1519 PatFrag ld_frag, string asm, OpndItins itins,
1521 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1523 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1524 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1525 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1527 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1528 (ins DstRC:$src1, x86memop:$src2),
1530 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1531 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1532 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1536 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1537 f128mem, load, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1538 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1539 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si{q}",
1540 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1542 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1543 f128mem, load, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1544 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1545 f128mem, load, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1548 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1549 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1550 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1551 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1552 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1553 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1555 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1556 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1557 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1558 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1559 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1560 SSE_CVT_Scalar, 0>, XD,
1563 let Constraints = "$src1 = $dst" in {
1564 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1565 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1566 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1567 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1568 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1569 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1570 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1571 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1572 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1573 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1574 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1575 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1580 // Aliases for intrinsics
1581 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1582 f32mem, load, "cvttss2si",
1583 SSE_CVT_SS2SI_32>, XS, VEX;
1584 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1585 int_x86_sse_cvttss2si64, f32mem, load,
1586 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1588 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1589 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1591 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1592 int_x86_sse2_cvttsd2si64, f128mem, load,
1593 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1595 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1596 f32mem, load, "cvttss2si",
1597 SSE_CVT_SS2SI_32>, XS;
1598 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1599 int_x86_sse_cvttss2si64, f32mem, load,
1600 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1602 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1603 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1605 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1606 int_x86_sse2_cvttsd2si64, f128mem, load,
1607 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1610 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1611 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1612 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1613 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1614 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1615 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1616 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1617 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1618 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1619 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1621 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1622 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1623 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1627 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1628 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1629 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1630 SSE_CVT_SS2SI_32>, XS;
1631 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1632 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1633 SSE_CVT_SS2SI_64>, XS, REX_W;
1634 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1635 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1636 SSEPackedSingle, SSE_CVT_PS>, TB,
1637 Requires<[HasSSE2]>;
1640 let Predicates = [HasAVX] in {
1641 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1642 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1643 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1644 (VCVTSS2SIrm addr:$src)>;
1645 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1646 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1647 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1648 (VCVTSS2SI64rm addr:$src)>;
1651 let Predicates = [HasSSE1] in {
1652 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1653 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1654 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1655 (CVTSS2SIrm addr:$src)>;
1656 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1657 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1658 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1659 (CVTSS2SI64rm addr:$src)>;
1664 // Convert scalar double to scalar single
1665 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1666 (ins FR64:$src1, FR64:$src2),
1667 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1668 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1670 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1671 (ins FR64:$src1, f64mem:$src2),
1672 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1673 [], IIC_SSE_CVT_Scalar_RM>,
1674 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1676 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1679 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1680 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1681 [(set FR32:$dst, (fround FR64:$src))],
1682 IIC_SSE_CVT_Scalar_RR>;
1683 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1684 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1685 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1686 IIC_SSE_CVT_Scalar_RM>,
1688 Requires<[HasSSE2, OptForSize]>;
1690 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1691 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1694 let Constraints = "$src1 = $dst" in
1695 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1696 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1697 SSE_CVT_Scalar>, XS;
1699 // Convert scalar single to scalar double
1700 // SSE2 instructions with XS prefix
1701 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1702 (ins FR32:$src1, FR32:$src2),
1703 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1704 [], IIC_SSE_CVT_Scalar_RR>,
1705 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1707 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1708 (ins FR32:$src1, f32mem:$src2),
1709 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1710 [], IIC_SSE_CVT_Scalar_RM>,
1711 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1713 let Predicates = [HasAVX] in {
1714 def : Pat<(f64 (fextend FR32:$src)),
1715 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1716 def : Pat<(fextend (loadf32 addr:$src)),
1717 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1718 def : Pat<(extloadf32 addr:$src),
1719 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1722 def : Pat<(extloadf32 addr:$src),
1723 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1724 Requires<[HasAVX, OptForSpeed]>;
1726 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1727 "cvtss2sd\t{$src, $dst|$dst, $src}",
1728 [(set FR64:$dst, (fextend FR32:$src))],
1729 IIC_SSE_CVT_Scalar_RR>, XS,
1730 Requires<[HasSSE2]>;
1731 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1732 "cvtss2sd\t{$src, $dst|$dst, $src}",
1733 [(set FR64:$dst, (extloadf32 addr:$src))],
1734 IIC_SSE_CVT_Scalar_RM>, XS,
1735 Requires<[HasSSE2, OptForSize]>;
1737 // extload f32 -> f64. This matches load+fextend because we have a hack in
1738 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1740 // Since these loads aren't folded into the fextend, we have to match it
1742 def : Pat<(fextend (loadf32 addr:$src)),
1743 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1744 def : Pat<(extloadf32 addr:$src),
1745 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1747 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1748 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1749 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1750 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1752 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V,
1754 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1755 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1756 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1757 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1758 (load addr:$src2)))],
1759 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V,
1761 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1762 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1763 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1764 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1767 IIC_SSE_CVT_Scalar_RR>, XS,
1768 Requires<[HasSSE2]>;
1769 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1770 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1771 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1772 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1773 (load addr:$src2)))],
1774 IIC_SSE_CVT_Scalar_RM>, XS,
1775 Requires<[HasSSE2]>;
1778 // Convert packed single/double fp to doubleword
1779 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1780 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1781 IIC_SSE_CVT_PS_RR>, VEX;
1782 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1783 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1784 IIC_SSE_CVT_PS_RM>, VEX;
1785 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1786 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1787 IIC_SSE_CVT_PS_RR>, VEX;
1788 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1789 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1790 IIC_SSE_CVT_PS_RM>, VEX;
1791 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1792 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1794 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1798 let Predicates = [HasAVX] in {
1799 def : Pat<(int_x86_sse2_cvtps2dq VR128:$src),
1800 (VCVTPS2DQrr VR128:$src)>;
1801 def : Pat<(int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)),
1802 (VCVTPS2DQrm addr:$src)>;
1805 let Predicates = [HasSSE2] in {
1806 def : Pat<(int_x86_sse2_cvtps2dq VR128:$src),
1807 (CVTPS2DQrr VR128:$src)>;
1808 def : Pat<(int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)),
1809 (CVTPS2DQrm addr:$src)>;
1812 // Convert Packed Double FP to Packed DW Integers
1813 let Predicates = [HasAVX] in {
1814 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1815 // register, but the same isn't true when using memory operands instead.
1816 // Provide other assembly rr and rm forms to address this explicitly.
1817 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1818 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1821 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1822 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1823 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1824 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1827 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1828 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX;
1829 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1830 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1831 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1832 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1835 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1836 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1838 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1839 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1842 let Predicates = [HasAVX] in {
1843 def : Pat<(int_x86_sse2_cvtpd2dq VR128:$src),
1844 (VCVTPD2DQrr VR128:$src)>;
1845 def : Pat<(int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)),
1846 (VCVTPD2DQXrm addr:$src)>;
1849 let Predicates = [HasSSE2] in {
1850 def : Pat<(int_x86_sse2_cvtpd2dq VR128:$src),
1851 (CVTPD2DQrr VR128:$src)>;
1852 def : Pat<(int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)),
1853 (CVTPD2DQrm addr:$src)>;
1856 // Convert with truncation packed single/double fp to doubleword
1857 // SSE2 packed instructions with XS prefix
1858 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1859 "cvttps2dq\t{$src, $dst|$dst, $src}",
1861 (int_x86_sse2_cvttps2dq VR128:$src))],
1862 IIC_SSE_CVT_PS_RR>, VEX;
1863 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1864 "cvttps2dq\t{$src, $dst|$dst, $src}",
1865 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1866 (memopv4f32 addr:$src)))],
1867 IIC_SSE_CVT_PS_RM>, VEX;
1868 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1869 "cvttps2dq\t{$src, $dst|$dst, $src}",
1871 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1872 IIC_SSE_CVT_PS_RR>, VEX;
1873 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1874 "cvttps2dq\t{$src, $dst|$dst, $src}",
1875 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1876 (memopv8f32 addr:$src)))],
1877 IIC_SSE_CVT_PS_RM>, VEX;
1879 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1880 "cvttps2dq\t{$src, $dst|$dst, $src}",
1882 (int_x86_sse2_cvttps2dq VR128:$src))],
1884 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1885 "cvttps2dq\t{$src, $dst|$dst, $src}",
1887 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1890 let Predicates = [HasAVX] in {
1891 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1892 (VCVTDQ2PSrr VR128:$src)>;
1893 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1894 (VCVTDQ2PSrm addr:$src)>;
1896 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1897 (VCVTDQ2PSrr VR128:$src)>;
1898 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1899 (VCVTDQ2PSrm addr:$src)>;
1901 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1902 (VCVTTPS2DQrr VR128:$src)>;
1903 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1904 (VCVTTPS2DQrm addr:$src)>;
1906 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1907 (VCVTDQ2PSYrr VR256:$src)>;
1908 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1909 (VCVTDQ2PSYrm addr:$src)>;
1911 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1912 (VCVTTPS2DQYrr VR256:$src)>;
1913 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1914 (VCVTTPS2DQYrm addr:$src)>;
1917 let Predicates = [HasSSE2] in {
1918 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1919 (CVTDQ2PSrr VR128:$src)>;
1920 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1921 (CVTDQ2PSrm addr:$src)>;
1923 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1924 (CVTDQ2PSrr VR128:$src)>;
1925 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1926 (CVTDQ2PSrm addr:$src)>;
1928 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1929 (CVTTPS2DQrr VR128:$src)>;
1930 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1931 (CVTTPS2DQrm addr:$src)>;
1934 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1935 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1937 (int_x86_sse2_cvttpd2dq VR128:$src))],
1938 IIC_SSE_CVT_PD_RR>, VEX;
1940 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1941 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1942 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1944 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1945 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1946 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1947 (memopv2f64 addr:$src)))],
1950 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1951 // register, but the same isn't true when using memory operands instead.
1952 // Provide other assembly rr and rm forms to address this explicitly.
1955 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1956 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1957 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1958 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1959 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1960 (memopv2f64 addr:$src)))],
1961 IIC_SSE_CVT_PD_RM>, VEX;
1964 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1965 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
1966 IIC_SSE_CVT_PD_RR>, VEX;
1967 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1968 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
1969 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1970 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1971 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1973 let Predicates = [HasAVX] in {
1974 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1975 (VCVTTPD2DQYrr VR256:$src)>;
1976 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1977 (VCVTTPD2DQYrm addr:$src)>;
1978 } // Predicates = [HasAVX]
1980 // Convert packed single to packed double
1981 let Predicates = [HasAVX] in {
1982 // SSE2 instructions without OpSize prefix
1983 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1984 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1985 IIC_SSE_CVT_PD_RR>, TB, VEX;
1986 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1987 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1988 IIC_SSE_CVT_PD_RM>, TB, VEX;
1989 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1990 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1991 IIC_SSE_CVT_PD_RR>, TB, VEX;
1992 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1993 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1994 IIC_SSE_CVT_PD_RM>, TB, VEX;
1997 let Predicates = [HasSSE2] in {
1998 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1999 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2000 IIC_SSE_CVT_PD_RR>, TB;
2001 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2002 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2003 IIC_SSE_CVT_PD_RM>, TB;
2006 let Predicates = [HasAVX] in {
2007 def : Pat<(int_x86_sse2_cvtps2pd VR128:$src),
2008 (VCVTPS2PDrr VR128:$src)>;
2011 let Predicates = [HasSSE2] in {
2012 def : Pat<(int_x86_sse2_cvtps2pd VR128:$src),
2013 (CVTPS2PDrr VR128:$src)>;
2016 // Convert Packed DW Integers to Packed Double FP
2017 let Predicates = [HasAVX] in {
2018 def VCVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2019 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2020 def VCVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2021 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2022 def VCVTDQ2PDYrm : SSDI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2023 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2024 def VCVTDQ2PDYrr : SSDI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2025 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2028 def CVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2029 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2031 def CVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2032 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2035 // 128 bit register conversion intrinsics
2036 let Predicates = [HasAVX] in
2037 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2038 (VCVTDQ2PDrr VR128:$src)>;
2040 let Predicates = [HasSSE2] in
2041 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2042 (CVTDQ2PDrr VR128:$src)>;
2044 // AVX 256-bit register conversion intrinsics
2045 let Predicates = [HasAVX] in {
2046 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
2047 (VCVTDQ2PDYrr VR128:$src)>;
2048 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
2049 (VCVTDQ2PDYrm addr:$src)>;
2051 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
2052 (VCVTPD2DQYrr VR256:$src)>;
2053 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
2054 (VCVTPD2DQYrm addr:$src)>;
2056 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2057 (VCVTDQ2PDYrr VR128:$src)>;
2058 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2059 (VCVTDQ2PDYrm addr:$src)>;
2060 } // Predicates = [HasAVX]
2062 // Convert packed double to packed single
2063 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2064 // register, but the same isn't true when using memory operands instead.
2065 // Provide other assembly rr and rm forms to address this explicitly.
2066 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2067 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2068 IIC_SSE_CVT_PD_RR>, VEX;
2071 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2072 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2073 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2074 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2075 IIC_SSE_CVT_PD_RM>, VEX;
2078 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2079 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2080 IIC_SSE_CVT_PD_RR>, VEX;
2081 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2082 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2083 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2084 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2085 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2087 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2088 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2090 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2091 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2095 let Predicates = [HasAVX] in {
2096 def : Pat<(int_x86_sse2_cvtpd2ps VR128:$src),
2097 (VCVTPD2PSrr VR128:$src)>;
2098 def : Pat<(int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)),
2099 (VCVTPD2PSXrm addr:$src)>;
2102 let Predicates = [HasSSE2] in {
2103 def : Pat<(int_x86_sse2_cvtpd2ps VR128:$src),
2104 (CVTPD2PSrr VR128:$src)>;
2105 def : Pat<(int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)),
2106 (CVTPD2PSrm addr:$src)>;
2109 // AVX 256-bit register conversion intrinsics
2110 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2111 // whenever possible to avoid declaring two versions of each one.
2112 let Predicates = [HasAVX] in {
2113 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2114 (VCVTDQ2PSYrr VR256:$src)>;
2115 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2116 (VCVTDQ2PSYrm addr:$src)>;
2118 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2119 (VCVTPD2PSYrr VR256:$src)>;
2120 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2121 (VCVTPD2PSYrm addr:$src)>;
2123 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2124 (VCVTPS2DQYrr VR256:$src)>;
2125 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2126 (VCVTPS2DQYrm addr:$src)>;
2128 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2129 (VCVTPS2PDYrr VR128:$src)>;
2130 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2131 (VCVTPS2PDYrm addr:$src)>;
2133 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2134 (VCVTTPD2DQYrr VR256:$src)>;
2135 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2136 (VCVTTPD2DQYrm addr:$src)>;
2138 // Match fround and fextend for 128/256-bit conversions
2139 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2140 (VCVTPD2PSYrr VR256:$src)>;
2141 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2142 (VCVTPD2PSYrm addr:$src)>;
2144 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2145 (VCVTPS2PDYrr VR128:$src)>;
2146 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2147 (VCVTPS2PDYrm addr:$src)>;
2150 //===----------------------------------------------------------------------===//
2151 // SSE 1 & 2 - Compare Instructions
2152 //===----------------------------------------------------------------------===//
2154 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2155 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2156 Operand CC, SDNode OpNode, ValueType VT,
2157 PatFrag ld_frag, string asm, string asm_alt,
2159 def rr : SIi8<0xC2, MRMSrcReg,
2160 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2161 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2163 def rm : SIi8<0xC2, MRMSrcMem,
2164 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2165 [(set RC:$dst, (OpNode (VT RC:$src1),
2166 (ld_frag addr:$src2), imm:$cc))],
2169 // Accept explicit immediate argument form instead of comparison code.
2170 let neverHasSideEffects = 1 in {
2171 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2172 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2173 IIC_SSE_ALU_F32S_RR>;
2175 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2176 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2177 IIC_SSE_ALU_F32S_RM>;
2181 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2182 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2183 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2185 XS, VEX_4V, VEX_LIG;
2186 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2187 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2188 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2189 SSE_ALU_F32S>, // same latency as 32 bit compare
2190 XD, VEX_4V, VEX_LIG;
2192 let Constraints = "$src1 = $dst" in {
2193 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2194 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2195 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2197 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2198 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2199 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2200 SSE_ALU_F32S>, // same latency as 32 bit compare
2204 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2205 Intrinsic Int, string asm, OpndItins itins> {
2206 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2207 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2208 [(set VR128:$dst, (Int VR128:$src1,
2209 VR128:$src, imm:$cc))],
2211 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2212 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2213 [(set VR128:$dst, (Int VR128:$src1,
2214 (load addr:$src), imm:$cc))],
2218 // Aliases to match intrinsics which expect XMM operand(s).
2219 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2220 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2223 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2224 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2225 SSE_ALU_F32S>, // same latency as f32
2227 let Constraints = "$src1 = $dst" in {
2228 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2229 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2231 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2232 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2233 SSE_ALU_F32S>, // same latency as f32
2238 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2239 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2240 ValueType vt, X86MemOperand x86memop,
2241 PatFrag ld_frag, string OpcodeStr, Domain d> {
2242 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2243 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2244 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2245 IIC_SSE_COMIS_RR, d>;
2246 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2247 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2248 [(set EFLAGS, (OpNode (vt RC:$src1),
2249 (ld_frag addr:$src2)))],
2250 IIC_SSE_COMIS_RM, d>;
2253 let Defs = [EFLAGS] in {
2254 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2255 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2256 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2257 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2259 let Pattern = []<dag> in {
2260 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2261 "comiss", SSEPackedSingle>, TB, VEX,
2263 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2264 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2268 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2269 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2270 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2271 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2273 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2274 load, "comiss", SSEPackedSingle>, TB, VEX;
2275 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2276 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2277 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2278 "ucomiss", SSEPackedSingle>, TB;
2279 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2280 "ucomisd", SSEPackedDouble>, TB, OpSize;
2282 let Pattern = []<dag> in {
2283 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2284 "comiss", SSEPackedSingle>, TB;
2285 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2286 "comisd", SSEPackedDouble>, TB, OpSize;
2289 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2290 load, "ucomiss", SSEPackedSingle>, TB;
2291 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2292 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2294 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2295 "comiss", SSEPackedSingle>, TB;
2296 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2297 "comisd", SSEPackedDouble>, TB, OpSize;
2298 } // Defs = [EFLAGS]
2300 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2301 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2302 Operand CC, Intrinsic Int, string asm,
2303 string asm_alt, Domain d> {
2304 def rri : PIi8<0xC2, MRMSrcReg,
2305 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2306 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2307 IIC_SSE_CMPP_RR, d>;
2308 def rmi : PIi8<0xC2, MRMSrcMem,
2309 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2310 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2311 IIC_SSE_CMPP_RM, d>;
2313 // Accept explicit immediate argument form instead of comparison code.
2314 let neverHasSideEffects = 1 in {
2315 def rri_alt : PIi8<0xC2, MRMSrcReg,
2316 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2317 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2318 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2319 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2320 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2324 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2325 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2326 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2327 SSEPackedSingle>, TB, VEX_4V;
2328 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2329 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2330 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2331 SSEPackedDouble>, TB, OpSize, VEX_4V;
2332 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2333 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2334 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2335 SSEPackedSingle>, TB, VEX_4V;
2336 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2337 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2338 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2339 SSEPackedDouble>, TB, OpSize, VEX_4V;
2340 let Constraints = "$src1 = $dst" in {
2341 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2342 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2343 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2344 SSEPackedSingle>, TB;
2345 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2346 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2347 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2348 SSEPackedDouble>, TB, OpSize;
2351 let Predicates = [HasAVX] in {
2352 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2353 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2354 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2355 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2356 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2357 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2358 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2359 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2361 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2362 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2363 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2364 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2365 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2366 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2367 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2368 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2371 let Predicates = [HasSSE1] in {
2372 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2373 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2374 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2375 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2378 let Predicates = [HasSSE2] in {
2379 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2380 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2381 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2382 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2385 //===----------------------------------------------------------------------===//
2386 // SSE 1 & 2 - Shuffle Instructions
2387 //===----------------------------------------------------------------------===//
2389 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2390 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2391 ValueType vt, string asm, PatFrag mem_frag,
2392 Domain d, bit IsConvertibleToThreeAddress = 0> {
2393 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2394 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2395 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2396 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2397 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2398 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2399 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2400 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2401 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2404 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2405 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2406 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2407 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2408 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2409 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2410 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2411 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2412 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2413 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2414 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2415 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2417 let Constraints = "$src1 = $dst" in {
2418 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2419 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2420 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2422 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2423 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2424 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2428 let Predicates = [HasAVX] in {
2429 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2430 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2431 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2432 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2433 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2435 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2436 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2437 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2438 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2439 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2442 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2443 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2444 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2445 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2446 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2448 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2449 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2450 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2451 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2452 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2455 let Predicates = [HasSSE1] in {
2456 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2457 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2458 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2459 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2460 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2463 let Predicates = [HasSSE2] in {
2464 // Generic SHUFPD patterns
2465 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2466 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2467 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2468 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2469 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2472 //===----------------------------------------------------------------------===//
2473 // SSE 1 & 2 - Unpack Instructions
2474 //===----------------------------------------------------------------------===//
2476 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2477 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2478 PatFrag mem_frag, RegisterClass RC,
2479 X86MemOperand x86memop, string asm,
2481 def rr : PI<opc, MRMSrcReg,
2482 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2484 (vt (OpNode RC:$src1, RC:$src2)))],
2486 def rm : PI<opc, MRMSrcMem,
2487 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2489 (vt (OpNode RC:$src1,
2490 (mem_frag addr:$src2))))],
2494 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2495 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2496 SSEPackedSingle>, TB, VEX_4V;
2497 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2498 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2499 SSEPackedDouble>, TB, OpSize, VEX_4V;
2500 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2501 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2502 SSEPackedSingle>, TB, VEX_4V;
2503 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2504 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2505 SSEPackedDouble>, TB, OpSize, VEX_4V;
2507 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2508 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2509 SSEPackedSingle>, TB, VEX_4V;
2510 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2511 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2512 SSEPackedDouble>, TB, OpSize, VEX_4V;
2513 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2514 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2515 SSEPackedSingle>, TB, VEX_4V;
2516 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2517 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2518 SSEPackedDouble>, TB, OpSize, VEX_4V;
2520 let Constraints = "$src1 = $dst" in {
2521 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2522 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2523 SSEPackedSingle>, TB;
2524 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2525 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2526 SSEPackedDouble>, TB, OpSize;
2527 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2528 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2529 SSEPackedSingle>, TB;
2530 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2531 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2532 SSEPackedDouble>, TB, OpSize;
2533 } // Constraints = "$src1 = $dst"
2535 let Predicates = [HasAVX], AddedComplexity = 1 in {
2536 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2537 // problem is during lowering, where it's not possible to recognize the load
2538 // fold cause it has two uses through a bitcast. One use disappears at isel
2539 // time and the fold opportunity reappears.
2540 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2541 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2544 let Predicates = [HasSSE2] in {
2545 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2546 // problem is during lowering, where it's not possible to recognize the load
2547 // fold cause it has two uses through a bitcast. One use disappears at isel
2548 // time and the fold opportunity reappears.
2549 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2550 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2553 //===----------------------------------------------------------------------===//
2554 // SSE 1 & 2 - Extract Floating-Point Sign mask
2555 //===----------------------------------------------------------------------===//
2557 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2558 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2560 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2561 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2562 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2563 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2564 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2565 IIC_SSE_MOVMSK, d>, REX_W;
2568 let Predicates = [HasAVX] in {
2569 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2570 "movmskps", SSEPackedSingle>, TB, VEX;
2571 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2572 "movmskpd", SSEPackedDouble>, TB,
2574 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2575 "movmskps", SSEPackedSingle>, TB, VEX;
2576 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2577 "movmskpd", SSEPackedDouble>, TB,
2580 def : Pat<(i32 (X86fgetsign FR32:$src)),
2581 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2583 def : Pat<(i64 (X86fgetsign FR32:$src)),
2584 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2586 def : Pat<(i32 (X86fgetsign FR64:$src)),
2587 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2589 def : Pat<(i64 (X86fgetsign FR64:$src)),
2590 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2594 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2595 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2596 SSEPackedSingle>, TB, VEX;
2597 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2598 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2599 SSEPackedDouble>, TB,
2601 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2602 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2603 SSEPackedSingle>, TB, VEX;
2604 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2605 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2606 SSEPackedDouble>, TB,
2610 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2611 SSEPackedSingle>, TB;
2612 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2613 SSEPackedDouble>, TB, OpSize;
2615 def : Pat<(i32 (X86fgetsign FR32:$src)),
2616 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2617 sub_ss))>, Requires<[HasSSE1]>;
2618 def : Pat<(i64 (X86fgetsign FR32:$src)),
2619 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2620 sub_ss))>, Requires<[HasSSE1]>;
2621 def : Pat<(i32 (X86fgetsign FR64:$src)),
2622 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2623 sub_sd))>, Requires<[HasSSE2]>;
2624 def : Pat<(i64 (X86fgetsign FR64:$src)),
2625 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2626 sub_sd))>, Requires<[HasSSE2]>;
2628 //===---------------------------------------------------------------------===//
2629 // SSE2 - Packed Integer Logical Instructions
2630 //===---------------------------------------------------------------------===//
2632 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2634 /// PDI_binop_rm - Simple SSE2 binary operator.
2635 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2636 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2637 X86MemOperand x86memop,
2639 bit IsCommutable = 0,
2641 let isCommutable = IsCommutable in
2642 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2643 (ins RC:$src1, RC:$src2),
2645 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2646 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2647 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2648 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2649 (ins RC:$src1, x86memop:$src2),
2651 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2652 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2653 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2654 (bitconvert (memop_frag addr:$src2)))))],
2657 } // ExeDomain = SSEPackedInt
2659 // These are ordered here for pattern ordering requirements with the fp versions
2661 let Predicates = [HasAVX] in {
2662 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2663 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2664 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2665 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2666 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2667 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2668 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2669 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2672 let Constraints = "$src1 = $dst" in {
2673 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2674 i128mem, SSE_BIT_ITINS_P, 1>;
2675 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2676 i128mem, SSE_BIT_ITINS_P, 1>;
2677 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2678 i128mem, SSE_BIT_ITINS_P, 1>;
2679 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2680 i128mem, SSE_BIT_ITINS_P, 0>;
2681 } // Constraints = "$src1 = $dst"
2683 let Predicates = [HasAVX2] in {
2684 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2685 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2686 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2687 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2688 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2689 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2690 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2691 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2694 //===----------------------------------------------------------------------===//
2695 // SSE 1 & 2 - Logical Instructions
2696 //===----------------------------------------------------------------------===//
2698 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2700 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2701 SDNode OpNode, OpndItins itins> {
2702 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2703 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2706 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2707 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2710 let Constraints = "$src1 = $dst" in {
2711 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2712 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2715 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2716 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2721 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2722 let mayLoad = 0 in {
2723 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2725 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2727 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2731 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2732 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2735 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2737 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2739 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2740 // are all promoted to v2i64, and the patterns are covered by the int
2741 // version. This is needed in SSE only, because v2i64 isn't supported on
2742 // SSE1, but only on SSE2.
2743 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2744 !strconcat(OpcodeStr, "ps"), f128mem, [],
2745 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2746 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2748 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2749 !strconcat(OpcodeStr, "pd"), f128mem,
2750 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2751 (bc_v2i64 (v2f64 VR128:$src2))))],
2752 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2753 (memopv2i64 addr:$src2)))], 0>,
2755 let Constraints = "$src1 = $dst" in {
2756 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2757 !strconcat(OpcodeStr, "ps"), f128mem,
2758 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2759 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2760 (memopv2i64 addr:$src2)))]>, TB;
2762 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2763 !strconcat(OpcodeStr, "pd"), f128mem,
2764 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2765 (bc_v2i64 (v2f64 VR128:$src2))))],
2766 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2767 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2771 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2773 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2775 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2776 !strconcat(OpcodeStr, "ps"), f256mem,
2777 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2778 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2779 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2781 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2782 !strconcat(OpcodeStr, "pd"), f256mem,
2783 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2784 (bc_v4i64 (v4f64 VR256:$src2))))],
2785 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2786 (memopv4i64 addr:$src2)))], 0>,
2790 // AVX 256-bit packed logical ops forms
2791 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2792 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2793 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2794 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2796 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2797 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2798 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2799 let isCommutable = 0 in
2800 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2802 //===----------------------------------------------------------------------===//
2803 // SSE 1 & 2 - Arithmetic Instructions
2804 //===----------------------------------------------------------------------===//
2806 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2809 /// In addition, we also have a special variant of the scalar form here to
2810 /// represent the associated intrinsic operation. This form is unlike the
2811 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2812 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2814 /// These three forms can each be reg+reg or reg+mem.
2817 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2819 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2822 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2823 OpNode, FR32, f32mem,
2824 itins.s, Is2Addr>, XS;
2825 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2826 OpNode, FR64, f64mem,
2827 itins.d, Is2Addr>, XD;
2830 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2833 let mayLoad = 0 in {
2834 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2835 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2837 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2838 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2843 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2846 let mayLoad = 0 in {
2847 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2848 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2850 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2851 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2856 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2859 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2860 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2861 itins.s, Is2Addr>, XS;
2862 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2863 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2864 itins.d, Is2Addr>, XD;
2867 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2870 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2871 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2872 SSEPackedSingle, itins.s, Is2Addr>,
2875 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2876 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2877 SSEPackedDouble, itins.d, Is2Addr>,
2881 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2883 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2884 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2885 SSEPackedSingle, itins.s, 0>, TB;
2887 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2888 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2889 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2892 // Binary Arithmetic instructions
2893 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2894 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2896 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2897 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2899 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2900 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2902 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2903 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2906 let isCommutable = 0 in {
2907 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2908 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2910 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2911 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2912 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2913 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2915 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2916 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2918 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2919 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2921 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2922 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2923 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2924 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2926 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2927 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2929 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2930 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2931 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2932 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2936 let Constraints = "$src1 = $dst" in {
2937 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2938 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2939 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2940 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2941 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2942 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2944 let isCommutable = 0 in {
2945 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2946 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2947 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2948 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2949 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2950 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2951 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2952 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2953 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2954 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2955 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2956 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2957 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2958 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2963 /// In addition, we also have a special variant of the scalar form here to
2964 /// represent the associated intrinsic operation. This form is unlike the
2965 /// plain scalar form, in that it takes an entire vector (instead of a
2966 /// scalar) and leaves the top elements undefined.
2968 /// And, we have a special variant form for a full-vector intrinsic form.
2970 def SSE_SQRTP : OpndItins<
2971 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2974 def SSE_SQRTS : OpndItins<
2975 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2978 def SSE_RCPP : OpndItins<
2979 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2982 def SSE_RCPS : OpndItins<
2983 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2986 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2987 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2988 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2989 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2990 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2991 [(set FR32:$dst, (OpNode FR32:$src))]>;
2992 // For scalar unary operations, fold a load into the operation
2993 // only in OptForSize mode. It eliminates an instruction, but it also
2994 // eliminates a whole-register clobber (the load), so it introduces a
2995 // partial register update condition.
2996 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2997 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2998 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2999 Requires<[HasSSE1, OptForSize]>;
3000 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3001 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3002 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
3003 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3004 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3005 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
3008 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3009 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3010 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3011 !strconcat(OpcodeStr,
3012 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3014 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3015 !strconcat(OpcodeStr,
3016 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3017 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3018 (ins VR128:$src1, ssmem:$src2),
3019 !strconcat(OpcodeStr,
3020 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3023 /// sse1_fp_unop_p - SSE1 unops in packed form.
3024 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3026 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3027 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3028 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3029 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3030 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3031 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3034 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3035 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3037 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3038 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3039 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3041 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3042 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3043 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3047 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3048 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3049 Intrinsic V4F32Int, OpndItins itins> {
3050 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3051 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3052 [(set VR128:$dst, (V4F32Int VR128:$src))],
3054 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3055 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3056 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3060 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3061 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3062 Intrinsic V4F32Int, OpndItins itins> {
3063 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3064 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3065 [(set VR256:$dst, (V4F32Int VR256:$src))],
3067 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3068 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3069 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3073 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3074 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3075 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3076 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3077 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3078 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3079 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3080 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3081 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3082 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3083 Requires<[HasSSE2, OptForSize]>;
3084 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3085 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3086 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3087 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3088 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3089 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3092 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3093 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3094 let neverHasSideEffects = 1 in {
3095 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3096 !strconcat(OpcodeStr,
3097 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3099 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3100 !strconcat(OpcodeStr,
3101 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3103 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3104 (ins VR128:$src1, sdmem:$src2),
3105 !strconcat(OpcodeStr,
3106 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3109 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3110 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3111 SDNode OpNode, OpndItins itins> {
3112 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3113 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3114 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3115 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3116 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3117 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3120 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3121 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3123 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3124 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3125 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3127 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3128 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3129 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3133 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3134 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3135 Intrinsic V2F64Int, OpndItins itins> {
3136 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3137 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3138 [(set VR128:$dst, (V2F64Int VR128:$src))],
3140 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3141 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3142 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3146 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3147 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3148 Intrinsic V2F64Int, OpndItins itins> {
3149 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3150 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3151 [(set VR256:$dst, (V2F64Int VR256:$src))],
3153 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3154 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3155 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3159 let Predicates = [HasAVX] in {
3161 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3162 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3164 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3165 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3166 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3167 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3168 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3170 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3172 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3174 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3178 // Reciprocal approximations. Note that these typically require refinement
3179 // in order to obtain suitable precision.
3180 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3181 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3182 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3183 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3185 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3188 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3189 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3190 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3191 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3193 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3197 let AddedComplexity = 1 in {
3198 def : Pat<(f32 (fsqrt FR32:$src)),
3199 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3200 def : Pat<(f32 (fsqrt (load addr:$src))),
3201 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3202 Requires<[HasAVX, OptForSize]>;
3203 def : Pat<(f64 (fsqrt FR64:$src)),
3204 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3205 def : Pat<(f64 (fsqrt (load addr:$src))),
3206 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3207 Requires<[HasAVX, OptForSize]>;
3209 def : Pat<(f32 (X86frsqrt FR32:$src)),
3210 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3211 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3212 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3213 Requires<[HasAVX, OptForSize]>;
3215 def : Pat<(f32 (X86frcp FR32:$src)),
3216 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3217 def : Pat<(f32 (X86frcp (load addr:$src))),
3218 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3219 Requires<[HasAVX, OptForSize]>;
3222 let Predicates = [HasAVX], AddedComplexity = 1 in {
3223 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3224 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3225 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3226 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3228 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3229 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3231 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3232 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3233 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3234 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3236 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3237 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3239 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3240 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3241 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3242 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3244 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3245 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3247 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3248 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3249 (VRCPSSr (f32 (IMPLICIT_DEF)),
3250 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3252 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3253 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3257 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3259 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3260 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3261 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3263 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3264 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3266 // Reciprocal approximations. Note that these typically require refinement
3267 // in order to obtain suitable precision.
3268 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3270 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3271 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3273 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3275 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3276 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3278 // There is no f64 version of the reciprocal approximation instructions.
3280 //===----------------------------------------------------------------------===//
3281 // SSE 1 & 2 - Non-temporal stores
3282 //===----------------------------------------------------------------------===//
3284 let AddedComplexity = 400 in { // Prefer non-temporal versions
3285 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3286 (ins f128mem:$dst, VR128:$src),
3287 "movntps\t{$src, $dst|$dst, $src}",
3288 [(alignednontemporalstore (v4f32 VR128:$src),
3290 IIC_SSE_MOVNT>, VEX;
3291 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3292 (ins f128mem:$dst, VR128:$src),
3293 "movntpd\t{$src, $dst|$dst, $src}",
3294 [(alignednontemporalstore (v2f64 VR128:$src),
3296 IIC_SSE_MOVNT>, VEX;
3298 let ExeDomain = SSEPackedInt in
3299 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3300 (ins f128mem:$dst, VR128:$src),
3301 "movntdq\t{$src, $dst|$dst, $src}",
3302 [(alignednontemporalstore (v2i64 VR128:$src),
3304 IIC_SSE_MOVNT>, VEX;
3306 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3307 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3309 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3310 (ins f256mem:$dst, VR256:$src),
3311 "movntps\t{$src, $dst|$dst, $src}",
3312 [(alignednontemporalstore (v8f32 VR256:$src),
3314 IIC_SSE_MOVNT>, VEX;
3315 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3316 (ins f256mem:$dst, VR256:$src),
3317 "movntpd\t{$src, $dst|$dst, $src}",
3318 [(alignednontemporalstore (v4f64 VR256:$src),
3320 IIC_SSE_MOVNT>, VEX;
3321 let ExeDomain = SSEPackedInt in
3322 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3323 (ins f256mem:$dst, VR256:$src),
3324 "movntdq\t{$src, $dst|$dst, $src}",
3325 [(alignednontemporalstore (v4i64 VR256:$src),
3327 IIC_SSE_MOVNT>, VEX;
3330 let AddedComplexity = 400 in { // Prefer non-temporal versions
3331 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3332 "movntps\t{$src, $dst|$dst, $src}",
3333 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3335 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3336 "movntpd\t{$src, $dst|$dst, $src}",
3337 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3340 let ExeDomain = SSEPackedInt in
3341 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3342 "movntdq\t{$src, $dst|$dst, $src}",
3343 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3346 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3347 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3349 // There is no AVX form for instructions below this point
3350 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3351 "movnti{l}\t{$src, $dst|$dst, $src}",
3352 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3354 TB, Requires<[HasSSE2]>;
3355 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3356 "movnti{q}\t{$src, $dst|$dst, $src}",
3357 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3359 TB, Requires<[HasSSE2]>;
3362 //===----------------------------------------------------------------------===//
3363 // SSE 1 & 2 - Prefetch and memory fence
3364 //===----------------------------------------------------------------------===//
3366 // Prefetch intrinsic.
3367 let Predicates = [HasSSE1] in {
3368 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3369 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3370 IIC_SSE_PREFETCH>, TB;
3371 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3372 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3373 IIC_SSE_PREFETCH>, TB;
3374 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3375 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3376 IIC_SSE_PREFETCH>, TB;
3377 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3378 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3379 IIC_SSE_PREFETCH>, TB;
3383 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3384 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3385 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3387 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3388 // was introduced with SSE2, it's backward compatible.
3389 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3391 // Load, store, and memory fence
3392 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3393 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3394 TB, Requires<[HasSSE1]>;
3395 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3396 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3397 TB, Requires<[HasSSE2]>;
3398 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3399 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3400 TB, Requires<[HasSSE2]>;
3402 def : Pat<(X86SFence), (SFENCE)>;
3403 def : Pat<(X86LFence), (LFENCE)>;
3404 def : Pat<(X86MFence), (MFENCE)>;
3406 //===----------------------------------------------------------------------===//
3407 // SSE 1 & 2 - Load/Store XCSR register
3408 //===----------------------------------------------------------------------===//
3410 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3411 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3412 IIC_SSE_LDMXCSR>, VEX;
3413 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3414 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3415 IIC_SSE_STMXCSR>, VEX;
3417 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3418 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3420 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3421 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3424 //===---------------------------------------------------------------------===//
3425 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3426 //===---------------------------------------------------------------------===//
3428 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3430 let neverHasSideEffects = 1 in {
3431 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3432 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3434 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3435 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3438 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3439 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3441 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3442 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3446 let isCodeGenOnly = 1 in {
3447 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3448 "movdqa\t{$src, $dst|$dst, $src}", [],
3451 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3452 "movdqa\t{$src, $dst|$dst, $src}", [],
3455 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3456 "movdqu\t{$src, $dst|$dst, $src}", [],
3459 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3460 "movdqu\t{$src, $dst|$dst, $src}", [],
3465 let canFoldAsLoad = 1, mayLoad = 1 in {
3466 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3467 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3469 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3470 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3472 let Predicates = [HasAVX] in {
3473 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3474 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3476 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3477 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3482 let mayStore = 1 in {
3483 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3484 (ins i128mem:$dst, VR128:$src),
3485 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3487 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3488 (ins i256mem:$dst, VR256:$src),
3489 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3491 let Predicates = [HasAVX] in {
3492 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3493 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3495 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3496 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3501 let neverHasSideEffects = 1 in
3502 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3503 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3505 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3506 "movdqu\t{$src, $dst|$dst, $src}",
3507 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3510 let isCodeGenOnly = 1 in {
3511 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3512 "movdqa\t{$src, $dst|$dst, $src}", [],
3515 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3516 "movdqu\t{$src, $dst|$dst, $src}",
3517 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3520 let canFoldAsLoad = 1, mayLoad = 1 in {
3521 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3522 "movdqa\t{$src, $dst|$dst, $src}",
3523 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3525 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3526 "movdqu\t{$src, $dst|$dst, $src}",
3527 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3529 XS, Requires<[HasSSE2]>;
3532 let mayStore = 1 in {
3533 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3534 "movdqa\t{$src, $dst|$dst, $src}",
3535 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3537 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3538 "movdqu\t{$src, $dst|$dst, $src}",
3539 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3541 XS, Requires<[HasSSE2]>;
3544 // Intrinsic forms of MOVDQU load and store
3545 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3546 "vmovdqu\t{$src, $dst|$dst, $src}",
3547 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3549 XS, VEX, Requires<[HasAVX]>;
3551 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3552 "movdqu\t{$src, $dst|$dst, $src}",
3553 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3555 XS, Requires<[HasSSE2]>;
3557 } // ExeDomain = SSEPackedInt
3559 let Predicates = [HasAVX] in {
3560 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3561 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3564 //===---------------------------------------------------------------------===//
3565 // SSE2 - Packed Integer Arithmetic Instructions
3566 //===---------------------------------------------------------------------===//
3568 def SSE_PMADD : OpndItins<
3569 IIC_SSE_PMADD, IIC_SSE_PMADD
3572 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3574 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3575 RegisterClass RC, PatFrag memop_frag,
3576 X86MemOperand x86memop,
3578 bit IsCommutable = 0,
3580 let isCommutable = IsCommutable in
3581 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3582 (ins RC:$src1, RC:$src2),
3584 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3585 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3586 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3587 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3588 (ins RC:$src1, x86memop:$src2),
3590 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3591 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3592 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3596 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3597 string OpcodeStr, SDNode OpNode,
3598 SDNode OpNode2, RegisterClass RC,
3599 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3600 ShiftOpndItins itins,
3602 // src2 is always 128-bit
3603 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3604 (ins RC:$src1, VR128:$src2),
3606 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3607 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3608 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3610 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3611 (ins RC:$src1, i128mem:$src2),
3613 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3614 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3615 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3616 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3617 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3618 (ins RC:$src1, i32i8imm:$src2),
3620 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3621 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3622 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3625 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3626 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3627 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3628 PatFrag memop_frag, X86MemOperand x86memop,
3630 bit IsCommutable = 0, bit Is2Addr = 1> {
3631 let isCommutable = IsCommutable in
3632 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3633 (ins RC:$src1, RC:$src2),
3635 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3636 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3637 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3638 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3639 (ins RC:$src1, x86memop:$src2),
3641 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3642 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3643 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3644 (bitconvert (memop_frag addr:$src2)))))]>;
3646 } // ExeDomain = SSEPackedInt
3648 // 128-bit Integer Arithmetic
3650 let Predicates = [HasAVX] in {
3651 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3652 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3654 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3655 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3656 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3657 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3658 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3659 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3660 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3661 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3662 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3663 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3664 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3665 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3666 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3667 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3668 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3669 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3670 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3671 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3675 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3676 VR128, memopv2i64, i128mem,
3677 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3678 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3679 VR128, memopv2i64, i128mem,
3680 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3681 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3682 VR128, memopv2i64, i128mem,
3683 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3684 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3685 VR128, memopv2i64, i128mem,
3686 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3687 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3688 VR128, memopv2i64, i128mem,
3689 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3690 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3691 VR128, memopv2i64, i128mem,
3692 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3693 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3694 VR128, memopv2i64, i128mem,
3695 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3696 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3697 VR128, memopv2i64, i128mem,
3698 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3699 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3700 VR128, memopv2i64, i128mem,
3701 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3702 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3703 VR128, memopv2i64, i128mem,
3704 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3705 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3706 VR128, memopv2i64, i128mem,
3707 SSE_PMADD, 1, 0>, VEX_4V;
3708 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3709 VR128, memopv2i64, i128mem,
3710 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3711 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3712 VR128, memopv2i64, i128mem,
3713 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3714 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3715 VR128, memopv2i64, i128mem,
3716 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3717 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3718 VR128, memopv2i64, i128mem,
3719 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3720 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3721 VR128, memopv2i64, i128mem,
3722 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3723 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3724 VR128, memopv2i64, i128mem,
3725 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3726 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3727 VR128, memopv2i64, i128mem,
3728 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3731 let Predicates = [HasAVX2] in {
3732 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3733 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3734 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3735 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3736 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3737 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3738 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3739 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3740 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3741 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3742 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3743 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3744 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3745 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3746 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3747 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3748 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3749 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3750 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3751 VR256, memopv4i64, i256mem,
3752 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3755 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3756 VR256, memopv4i64, i256mem,
3757 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3758 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3759 VR256, memopv4i64, i256mem,
3760 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3761 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3762 VR256, memopv4i64, i256mem,
3763 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3764 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3765 VR256, memopv4i64, i256mem,
3766 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3767 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3768 VR256, memopv4i64, i256mem,
3769 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3770 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3771 VR256, memopv4i64, i256mem,
3772 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3773 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3774 VR256, memopv4i64, i256mem,
3775 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3776 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3777 VR256, memopv4i64, i256mem,
3778 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3779 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3780 VR256, memopv4i64, i256mem,
3781 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3782 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3783 VR256, memopv4i64, i256mem,
3784 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3785 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3786 VR256, memopv4i64, i256mem,
3787 SSE_PMADD, 1, 0>, VEX_4V;
3788 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3789 VR256, memopv4i64, i256mem,
3790 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3791 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3792 VR256, memopv4i64, i256mem,
3793 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3794 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3795 VR256, memopv4i64, i256mem,
3796 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3797 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3798 VR256, memopv4i64, i256mem,
3799 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3800 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3801 VR256, memopv4i64, i256mem,
3802 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3803 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3804 VR256, memopv4i64, i256mem,
3805 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3806 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3807 VR256, memopv4i64, i256mem,
3808 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3811 let Constraints = "$src1 = $dst" in {
3812 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3813 i128mem, SSE_INTALU_ITINS_P, 1>;
3814 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3815 i128mem, SSE_INTALU_ITINS_P, 1>;
3816 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3817 i128mem, SSE_INTALU_ITINS_P, 1>;
3818 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3819 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3820 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3821 i128mem, SSE_INTMUL_ITINS_P, 1>;
3822 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3823 i128mem, SSE_INTALU_ITINS_P>;
3824 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3825 i128mem, SSE_INTALU_ITINS_P>;
3826 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3827 i128mem, SSE_INTALU_ITINS_P>;
3828 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3829 i128mem, SSE_INTALUQ_ITINS_P>;
3830 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3831 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3834 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3835 VR128, memopv2i64, i128mem,
3836 SSE_INTALU_ITINS_P>;
3837 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3838 VR128, memopv2i64, i128mem,
3839 SSE_INTALU_ITINS_P>;
3840 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3841 VR128, memopv2i64, i128mem,
3842 SSE_INTALU_ITINS_P>;
3843 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3844 VR128, memopv2i64, i128mem,
3845 SSE_INTALU_ITINS_P>;
3846 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3847 VR128, memopv2i64, i128mem,
3848 SSE_INTALU_ITINS_P, 1>;
3849 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3850 VR128, memopv2i64, i128mem,
3851 SSE_INTALU_ITINS_P, 1>;
3852 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3853 VR128, memopv2i64, i128mem,
3854 SSE_INTALU_ITINS_P, 1>;
3855 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3856 VR128, memopv2i64, i128mem,
3857 SSE_INTALU_ITINS_P, 1>;
3858 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3859 VR128, memopv2i64, i128mem,
3860 SSE_INTMUL_ITINS_P, 1>;
3861 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3862 VR128, memopv2i64, i128mem,
3863 SSE_INTMUL_ITINS_P, 1>;
3864 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3865 VR128, memopv2i64, i128mem,
3867 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3868 VR128, memopv2i64, i128mem,
3869 SSE_INTALU_ITINS_P, 1>;
3870 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3871 VR128, memopv2i64, i128mem,
3872 SSE_INTALU_ITINS_P, 1>;
3873 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3874 VR128, memopv2i64, i128mem,
3875 SSE_INTALU_ITINS_P, 1>;
3876 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3877 VR128, memopv2i64, i128mem,
3878 SSE_INTALU_ITINS_P, 1>;
3879 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3880 VR128, memopv2i64, i128mem,
3881 SSE_INTALU_ITINS_P, 1>;
3882 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3883 VR128, memopv2i64, i128mem,
3884 SSE_INTALU_ITINS_P, 1>;
3885 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3886 VR128, memopv2i64, i128mem,
3887 SSE_INTALU_ITINS_P, 1>;
3889 } // Constraints = "$src1 = $dst"
3891 //===---------------------------------------------------------------------===//
3892 // SSE2 - Packed Integer Logical Instructions
3893 //===---------------------------------------------------------------------===//
3895 let Predicates = [HasAVX] in {
3896 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3897 VR128, v8i16, v8i16, bc_v8i16,
3898 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3899 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3900 VR128, v4i32, v4i32, bc_v4i32,
3901 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3902 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3903 VR128, v2i64, v2i64, bc_v2i64,
3904 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3906 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3907 VR128, v8i16, v8i16, bc_v8i16,
3908 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3909 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3910 VR128, v4i32, v4i32, bc_v4i32,
3911 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3912 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3913 VR128, v2i64, v2i64, bc_v2i64,
3914 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3916 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3917 VR128, v8i16, v8i16, bc_v8i16,
3918 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3919 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3920 VR128, v4i32, v4i32, bc_v4i32,
3921 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3923 let ExeDomain = SSEPackedInt in {
3924 // 128-bit logical shifts.
3925 def VPSLLDQri : PDIi8<0x73, MRM7r,
3926 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3927 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3929 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3931 def VPSRLDQri : PDIi8<0x73, MRM3r,
3932 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3933 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3935 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3937 // PSRADQri doesn't exist in SSE[1-3].
3939 } // Predicates = [HasAVX]
3941 let Predicates = [HasAVX2] in {
3942 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3943 VR256, v16i16, v8i16, bc_v8i16,
3944 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3945 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3946 VR256, v8i32, v4i32, bc_v4i32,
3947 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3948 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3949 VR256, v4i64, v2i64, bc_v2i64,
3950 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3952 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3953 VR256, v16i16, v8i16, bc_v8i16,
3954 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3955 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3956 VR256, v8i32, v4i32, bc_v4i32,
3957 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3958 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3959 VR256, v4i64, v2i64, bc_v2i64,
3960 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3962 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3963 VR256, v16i16, v8i16, bc_v8i16,
3964 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3965 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3966 VR256, v8i32, v4i32, bc_v4i32,
3967 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3969 let ExeDomain = SSEPackedInt in {
3970 // 256-bit logical shifts.
3971 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3972 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3973 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3975 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3977 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3978 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3979 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3981 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3983 // PSRADQYri doesn't exist in SSE[1-3].
3985 } // Predicates = [HasAVX2]
3987 let Constraints = "$src1 = $dst" in {
3988 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3989 VR128, v8i16, v8i16, bc_v8i16,
3990 SSE_INTSHIFT_ITINS_P>;
3991 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3992 VR128, v4i32, v4i32, bc_v4i32,
3993 SSE_INTSHIFT_ITINS_P>;
3994 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3995 VR128, v2i64, v2i64, bc_v2i64,
3996 SSE_INTSHIFT_ITINS_P>;
3998 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3999 VR128, v8i16, v8i16, bc_v8i16,
4000 SSE_INTSHIFT_ITINS_P>;
4001 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
4002 VR128, v4i32, v4i32, bc_v4i32,
4003 SSE_INTSHIFT_ITINS_P>;
4004 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
4005 VR128, v2i64, v2i64, bc_v2i64,
4006 SSE_INTSHIFT_ITINS_P>;
4008 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
4009 VR128, v8i16, v8i16, bc_v8i16,
4010 SSE_INTSHIFT_ITINS_P>;
4011 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
4012 VR128, v4i32, v4i32, bc_v4i32,
4013 SSE_INTSHIFT_ITINS_P>;
4015 let ExeDomain = SSEPackedInt in {
4016 // 128-bit logical shifts.
4017 def PSLLDQri : PDIi8<0x73, MRM7r,
4018 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4019 "pslldq\t{$src2, $dst|$dst, $src2}",
4021 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
4022 def PSRLDQri : PDIi8<0x73, MRM3r,
4023 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4024 "psrldq\t{$src2, $dst|$dst, $src2}",
4026 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
4027 // PSRADQri doesn't exist in SSE[1-3].
4029 } // Constraints = "$src1 = $dst"
4031 let Predicates = [HasAVX] in {
4032 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4033 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4034 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4035 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4036 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4037 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4039 // Shift up / down and insert zero's.
4040 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4041 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4042 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4043 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4046 let Predicates = [HasAVX2] in {
4047 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4048 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4049 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4050 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4053 let Predicates = [HasSSE2] in {
4054 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4055 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4056 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4057 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4058 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4059 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4061 // Shift up / down and insert zero's.
4062 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4063 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4064 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4065 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4068 //===---------------------------------------------------------------------===//
4069 // SSE2 - Packed Integer Comparison Instructions
4070 //===---------------------------------------------------------------------===//
4072 let Predicates = [HasAVX] in {
4073 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4074 VR128, memopv2i64, i128mem,
4075 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4076 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4077 VR128, memopv2i64, i128mem,
4078 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4079 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4080 VR128, memopv2i64, i128mem,
4081 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4082 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4083 VR128, memopv2i64, i128mem,
4084 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4085 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4086 VR128, memopv2i64, i128mem,
4087 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4088 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4089 VR128, memopv2i64, i128mem,
4090 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4093 let Predicates = [HasAVX2] in {
4094 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4095 VR256, memopv4i64, i256mem,
4096 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4097 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4098 VR256, memopv4i64, i256mem,
4099 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4100 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4101 VR256, memopv4i64, i256mem,
4102 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4103 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4104 VR256, memopv4i64, i256mem,
4105 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4106 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4107 VR256, memopv4i64, i256mem,
4108 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4109 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4110 VR256, memopv4i64, i256mem,
4111 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4114 let Constraints = "$src1 = $dst" in {
4115 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4116 VR128, memopv2i64, i128mem,
4117 SSE_INTALU_ITINS_P, 1>;
4118 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4119 VR128, memopv2i64, i128mem,
4120 SSE_INTALU_ITINS_P, 1>;
4121 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4122 VR128, memopv2i64, i128mem,
4123 SSE_INTALU_ITINS_P, 1>;
4124 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4125 VR128, memopv2i64, i128mem,
4126 SSE_INTALU_ITINS_P>;
4127 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4128 VR128, memopv2i64, i128mem,
4129 SSE_INTALU_ITINS_P>;
4130 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4131 VR128, memopv2i64, i128mem,
4132 SSE_INTALU_ITINS_P>;
4133 } // Constraints = "$src1 = $dst"
4135 //===---------------------------------------------------------------------===//
4136 // SSE2 - Packed Integer Pack Instructions
4137 //===---------------------------------------------------------------------===//
4139 let Predicates = [HasAVX] in {
4140 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4141 VR128, memopv2i64, i128mem,
4142 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4143 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4144 VR128, memopv2i64, i128mem,
4145 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4146 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4147 VR128, memopv2i64, i128mem,
4148 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4151 let Predicates = [HasAVX2] in {
4152 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4153 VR256, memopv4i64, i256mem,
4154 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4155 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4156 VR256, memopv4i64, i256mem,
4157 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4158 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4159 VR256, memopv4i64, i256mem,
4160 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4163 let Constraints = "$src1 = $dst" in {
4164 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4165 VR128, memopv2i64, i128mem,
4166 SSE_INTALU_ITINS_P>;
4167 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4168 VR128, memopv2i64, i128mem,
4169 SSE_INTALU_ITINS_P>;
4170 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4171 VR128, memopv2i64, i128mem,
4172 SSE_INTALU_ITINS_P>;
4173 } // Constraints = "$src1 = $dst"
4175 //===---------------------------------------------------------------------===//
4176 // SSE2 - Packed Integer Shuffle Instructions
4177 //===---------------------------------------------------------------------===//
4179 let ExeDomain = SSEPackedInt in {
4180 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4181 def ri : Ii8<0x70, MRMSrcReg,
4182 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4183 !strconcat(OpcodeStr,
4184 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4185 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4187 def mi : Ii8<0x70, MRMSrcMem,
4188 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4189 !strconcat(OpcodeStr,
4190 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4192 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4197 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4198 def Yri : Ii8<0x70, MRMSrcReg,
4199 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4200 !strconcat(OpcodeStr,
4201 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4202 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4203 def Ymi : Ii8<0x70, MRMSrcMem,
4204 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4205 !strconcat(OpcodeStr,
4206 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4208 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4209 (i8 imm:$src2))))]>;
4211 } // ExeDomain = SSEPackedInt
4213 let Predicates = [HasAVX] in {
4214 let AddedComplexity = 5 in
4215 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4217 // SSE2 with ImmT == Imm8 and XS prefix.
4218 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4220 // SSE2 with ImmT == Imm8 and XD prefix.
4221 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4223 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4224 (VPSHUFDmi addr:$src1, imm:$imm)>;
4225 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4226 (VPSHUFDri VR128:$src1, imm:$imm)>;
4229 let Predicates = [HasAVX2] in {
4230 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4231 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4232 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4235 let Predicates = [HasSSE2] in {
4236 let AddedComplexity = 5 in
4237 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4239 // SSE2 with ImmT == Imm8 and XS prefix.
4240 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4242 // SSE2 with ImmT == Imm8 and XD prefix.
4243 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4245 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4246 (PSHUFDmi addr:$src1, imm:$imm)>;
4247 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4248 (PSHUFDri VR128:$src1, imm:$imm)>;
4251 //===---------------------------------------------------------------------===//
4252 // SSE2 - Packed Integer Unpack Instructions
4253 //===---------------------------------------------------------------------===//
4255 let ExeDomain = SSEPackedInt in {
4256 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4257 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4258 def rr : PDI<opc, MRMSrcReg,
4259 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4261 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4262 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4263 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4265 def rm : PDI<opc, MRMSrcMem,
4266 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4268 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4269 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4270 [(set VR128:$dst, (OpNode VR128:$src1,
4271 (bc_frag (memopv2i64
4276 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4277 SDNode OpNode, PatFrag bc_frag> {
4278 def Yrr : PDI<opc, MRMSrcReg,
4279 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4280 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4281 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4282 def Yrm : PDI<opc, MRMSrcMem,
4283 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4284 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4285 [(set VR256:$dst, (OpNode VR256:$src1,
4286 (bc_frag (memopv4i64 addr:$src2))))]>;
4289 let Predicates = [HasAVX] in {
4290 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4291 bc_v16i8, 0>, VEX_4V;
4292 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4293 bc_v8i16, 0>, VEX_4V;
4294 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4295 bc_v4i32, 0>, VEX_4V;
4296 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4297 bc_v2i64, 0>, VEX_4V;
4299 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4300 bc_v16i8, 0>, VEX_4V;
4301 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4302 bc_v8i16, 0>, VEX_4V;
4303 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4304 bc_v4i32, 0>, VEX_4V;
4305 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4306 bc_v2i64, 0>, VEX_4V;
4309 let Predicates = [HasAVX2] in {
4310 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4312 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4314 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4316 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4319 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4321 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4323 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4325 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4329 let Constraints = "$src1 = $dst" in {
4330 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4332 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4334 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4336 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4339 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4341 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4343 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4345 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4348 } // ExeDomain = SSEPackedInt
4350 // Patterns for using AVX1 instructions with integer vectors
4351 // Here to give AVX2 priority
4352 let Predicates = [HasAVX] in {
4353 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4354 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4355 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4356 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4357 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4358 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4359 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4360 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4362 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4363 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4364 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4365 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4366 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4367 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4368 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4369 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4372 //===---------------------------------------------------------------------===//
4373 // SSE2 - Packed Integer Extract and Insert
4374 //===---------------------------------------------------------------------===//
4376 let ExeDomain = SSEPackedInt in {
4377 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4378 def rri : Ii8<0xC4, MRMSrcReg,
4379 (outs VR128:$dst), (ins VR128:$src1,
4380 GR32:$src2, i32i8imm:$src3),
4382 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4383 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4385 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4386 def rmi : Ii8<0xC4, MRMSrcMem,
4387 (outs VR128:$dst), (ins VR128:$src1,
4388 i16mem:$src2, i32i8imm:$src3),
4390 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4391 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4393 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4394 imm:$src3))], IIC_SSE_PINSRW>;
4398 let Predicates = [HasAVX] in
4399 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4400 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4401 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4402 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4403 imm:$src2))]>, TB, OpSize, VEX;
4404 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4405 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4406 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4407 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4408 imm:$src2))], IIC_SSE_PEXTRW>;
4411 let Predicates = [HasAVX] in {
4412 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4413 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4414 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4415 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4416 []>, TB, OpSize, VEX_4V;
4419 let Constraints = "$src1 = $dst" in
4420 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4422 } // ExeDomain = SSEPackedInt
4424 //===---------------------------------------------------------------------===//
4425 // SSE2 - Packed Mask Creation
4426 //===---------------------------------------------------------------------===//
4428 let ExeDomain = SSEPackedInt in {
4430 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4431 "pmovmskb\t{$src, $dst|$dst, $src}",
4432 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4433 IIC_SSE_MOVMSK>, VEX;
4434 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4435 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4437 let Predicates = [HasAVX2] in {
4438 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4439 "pmovmskb\t{$src, $dst|$dst, $src}",
4440 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4441 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4442 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4445 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4446 "pmovmskb\t{$src, $dst|$dst, $src}",
4447 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4450 } // ExeDomain = SSEPackedInt
4452 //===---------------------------------------------------------------------===//
4453 // SSE2 - Conditional Store
4454 //===---------------------------------------------------------------------===//
4456 let ExeDomain = SSEPackedInt in {
4459 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4460 (ins VR128:$src, VR128:$mask),
4461 "maskmovdqu\t{$mask, $src|$src, $mask}",
4462 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4463 IIC_SSE_MASKMOV>, VEX;
4465 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4466 (ins VR128:$src, VR128:$mask),
4467 "maskmovdqu\t{$mask, $src|$src, $mask}",
4468 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4469 IIC_SSE_MASKMOV>, VEX;
4472 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4473 "maskmovdqu\t{$mask, $src|$src, $mask}",
4474 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4477 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4478 "maskmovdqu\t{$mask, $src|$src, $mask}",
4479 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4482 } // ExeDomain = SSEPackedInt
4484 //===---------------------------------------------------------------------===//
4485 // SSE2 - Move Doubleword
4486 //===---------------------------------------------------------------------===//
4488 //===---------------------------------------------------------------------===//
4489 // Move Int Doubleword to Packed Double Int
4491 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4492 "movd\t{$src, $dst|$dst, $src}",
4494 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4496 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4497 "movd\t{$src, $dst|$dst, $src}",
4499 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4502 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4503 "mov{d|q}\t{$src, $dst|$dst, $src}",
4505 (v2i64 (scalar_to_vector GR64:$src)))],
4506 IIC_SSE_MOVDQ>, VEX;
4507 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4508 "mov{d|q}\t{$src, $dst|$dst, $src}",
4509 [(set FR64:$dst, (bitconvert GR64:$src))],
4510 IIC_SSE_MOVDQ>, VEX;
4512 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4513 "movd\t{$src, $dst|$dst, $src}",
4515 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4516 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4517 "movd\t{$src, $dst|$dst, $src}",
4519 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4521 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4522 "mov{d|q}\t{$src, $dst|$dst, $src}",
4524 (v2i64 (scalar_to_vector GR64:$src)))],
4526 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4527 "mov{d|q}\t{$src, $dst|$dst, $src}",
4528 [(set FR64:$dst, (bitconvert GR64:$src))],
4531 //===---------------------------------------------------------------------===//
4532 // Move Int Doubleword to Single Scalar
4534 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4535 "movd\t{$src, $dst|$dst, $src}",
4536 [(set FR32:$dst, (bitconvert GR32:$src))],
4537 IIC_SSE_MOVDQ>, VEX;
4539 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4540 "movd\t{$src, $dst|$dst, $src}",
4541 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4544 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4545 "movd\t{$src, $dst|$dst, $src}",
4546 [(set FR32:$dst, (bitconvert GR32:$src))],
4549 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4550 "movd\t{$src, $dst|$dst, $src}",
4551 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4554 //===---------------------------------------------------------------------===//
4555 // Move Packed Doubleword Int to Packed Double Int
4557 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4558 "movd\t{$src, $dst|$dst, $src}",
4559 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4560 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4561 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4562 (ins i32mem:$dst, VR128:$src),
4563 "movd\t{$src, $dst|$dst, $src}",
4564 [(store (i32 (vector_extract (v4i32 VR128:$src),
4565 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4567 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4568 "movd\t{$src, $dst|$dst, $src}",
4569 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4570 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4571 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4572 "movd\t{$src, $dst|$dst, $src}",
4573 [(store (i32 (vector_extract (v4i32 VR128:$src),
4574 (iPTR 0))), addr:$dst)],
4577 //===---------------------------------------------------------------------===//
4578 // Move Packed Doubleword Int first element to Doubleword Int
4580 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4581 "mov{d|q}\t{$src, $dst|$dst, $src}",
4582 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4585 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4587 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4588 "mov{d|q}\t{$src, $dst|$dst, $src}",
4589 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4593 //===---------------------------------------------------------------------===//
4594 // Bitcast FR64 <-> GR64
4596 let Predicates = [HasAVX] in
4597 def VMOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4598 "vmovq\t{$src, $dst|$dst, $src}",
4599 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4601 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4602 "mov{d|q}\t{$src, $dst|$dst, $src}",
4603 [(set GR64:$dst, (bitconvert FR64:$src))],
4604 IIC_SSE_MOVDQ>, VEX;
4605 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4606 "movq\t{$src, $dst|$dst, $src}",
4607 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4608 IIC_SSE_MOVDQ>, VEX;
4610 def MOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4611 "movq\t{$src, $dst|$dst, $src}",
4612 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4614 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4615 "mov{d|q}\t{$src, $dst|$dst, $src}",
4616 [(set GR64:$dst, (bitconvert FR64:$src))],
4618 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4619 "movq\t{$src, $dst|$dst, $src}",
4620 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4623 //===---------------------------------------------------------------------===//
4624 // Move Scalar Single to Double Int
4626 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4627 "movd\t{$src, $dst|$dst, $src}",
4628 [(set GR32:$dst, (bitconvert FR32:$src))],
4629 IIC_SSE_MOVD_ToGP>, VEX;
4630 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4631 "movd\t{$src, $dst|$dst, $src}",
4632 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4633 IIC_SSE_MOVDQ>, VEX;
4634 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4635 "movd\t{$src, $dst|$dst, $src}",
4636 [(set GR32:$dst, (bitconvert FR32:$src))],
4638 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4639 "movd\t{$src, $dst|$dst, $src}",
4640 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4643 //===---------------------------------------------------------------------===//
4644 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4646 let AddedComplexity = 15 in {
4647 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4648 "movd\t{$src, $dst|$dst, $src}",
4649 [(set VR128:$dst, (v4i32 (X86vzmovl
4650 (v4i32 (scalar_to_vector GR32:$src)))))],
4651 IIC_SSE_MOVDQ>, VEX;
4652 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4653 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4654 [(set VR128:$dst, (v2i64 (X86vzmovl
4655 (v2i64 (scalar_to_vector GR64:$src)))))],
4659 let AddedComplexity = 15 in {
4660 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4661 "movd\t{$src, $dst|$dst, $src}",
4662 [(set VR128:$dst, (v4i32 (X86vzmovl
4663 (v4i32 (scalar_to_vector GR32:$src)))))],
4665 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4666 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4667 [(set VR128:$dst, (v2i64 (X86vzmovl
4668 (v2i64 (scalar_to_vector GR64:$src)))))],
4672 let AddedComplexity = 20 in {
4673 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4674 "movd\t{$src, $dst|$dst, $src}",
4676 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4677 (loadi32 addr:$src))))))],
4678 IIC_SSE_MOVDQ>, VEX;
4679 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4680 "movd\t{$src, $dst|$dst, $src}",
4682 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4683 (loadi32 addr:$src))))))],
4687 let Predicates = [HasAVX] in {
4688 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4689 let AddedComplexity = 20 in {
4690 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4691 (VMOVZDI2PDIrm addr:$src)>;
4692 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4693 (VMOVZDI2PDIrm addr:$src)>;
4695 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4696 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4697 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4698 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4699 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4700 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4701 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4704 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4705 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4706 (MOVZDI2PDIrm addr:$src)>;
4707 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4708 (MOVZDI2PDIrm addr:$src)>;
4711 // These are the correct encodings of the instructions so that we know how to
4712 // read correct assembly, even though we continue to emit the wrong ones for
4713 // compatibility with Darwin's buggy assembler.
4714 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4715 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4716 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4717 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4718 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4719 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4720 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4721 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4722 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4723 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4724 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4725 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4727 //===---------------------------------------------------------------------===//
4728 // SSE2 - Move Quadword
4729 //===---------------------------------------------------------------------===//
4731 //===---------------------------------------------------------------------===//
4732 // Move Quadword Int to Packed Quadword Int
4734 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4735 "vmovq\t{$src, $dst|$dst, $src}",
4737 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4738 VEX, Requires<[HasAVX]>;
4739 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4740 "movq\t{$src, $dst|$dst, $src}",
4742 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4744 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4746 //===---------------------------------------------------------------------===//
4747 // Move Packed Quadword Int to Quadword Int
4749 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4750 "movq\t{$src, $dst|$dst, $src}",
4751 [(store (i64 (vector_extract (v2i64 VR128:$src),
4752 (iPTR 0))), addr:$dst)],
4753 IIC_SSE_MOVDQ>, VEX;
4754 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4755 "movq\t{$src, $dst|$dst, $src}",
4756 [(store (i64 (vector_extract (v2i64 VR128:$src),
4757 (iPTR 0))), addr:$dst)],
4760 //===---------------------------------------------------------------------===//
4761 // Store / copy lower 64-bits of a XMM register.
4763 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4764 "movq\t{$src, $dst|$dst, $src}",
4765 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4766 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4767 "movq\t{$src, $dst|$dst, $src}",
4768 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4771 let AddedComplexity = 20 in
4772 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4773 "vmovq\t{$src, $dst|$dst, $src}",
4775 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4776 (loadi64 addr:$src))))))],
4778 XS, VEX, Requires<[HasAVX]>;
4780 let AddedComplexity = 20 in
4781 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4782 "movq\t{$src, $dst|$dst, $src}",
4784 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4785 (loadi64 addr:$src))))))],
4787 XS, Requires<[HasSSE2]>;
4789 let Predicates = [HasAVX], AddedComplexity = 20 in {
4790 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4791 (VMOVZQI2PQIrm addr:$src)>;
4792 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4793 (VMOVZQI2PQIrm addr:$src)>;
4794 def : Pat<(v2i64 (X86vzload addr:$src)),
4795 (VMOVZQI2PQIrm addr:$src)>;
4798 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4799 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4800 (MOVZQI2PQIrm addr:$src)>;
4801 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4802 (MOVZQI2PQIrm addr:$src)>;
4803 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4806 let Predicates = [HasAVX] in {
4807 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4808 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4809 def : Pat<(v4i64 (X86vzload addr:$src)),
4810 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4813 //===---------------------------------------------------------------------===//
4814 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4815 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4817 let AddedComplexity = 15 in
4818 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4819 "vmovq\t{$src, $dst|$dst, $src}",
4820 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4822 XS, VEX, Requires<[HasAVX]>;
4823 let AddedComplexity = 15 in
4824 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4825 "movq\t{$src, $dst|$dst, $src}",
4826 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4828 XS, Requires<[HasSSE2]>;
4830 let AddedComplexity = 20 in
4831 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4832 "vmovq\t{$src, $dst|$dst, $src}",
4833 [(set VR128:$dst, (v2i64 (X86vzmovl
4834 (loadv2i64 addr:$src))))],
4836 XS, VEX, Requires<[HasAVX]>;
4837 let AddedComplexity = 20 in {
4838 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4839 "movq\t{$src, $dst|$dst, $src}",
4840 [(set VR128:$dst, (v2i64 (X86vzmovl
4841 (loadv2i64 addr:$src))))],
4843 XS, Requires<[HasSSE2]>;
4846 let AddedComplexity = 20 in {
4847 let Predicates = [HasAVX] in {
4848 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4849 (VMOVZPQILo2PQIrm addr:$src)>;
4850 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4851 (VMOVZPQILo2PQIrr VR128:$src)>;
4853 let Predicates = [HasSSE2] in {
4854 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4855 (MOVZPQILo2PQIrm addr:$src)>;
4856 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4857 (MOVZPQILo2PQIrr VR128:$src)>;
4861 // Instructions to match in the assembler
4862 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4863 "movq\t{$src, $dst|$dst, $src}", [],
4864 IIC_SSE_MOVDQ>, VEX, VEX_W;
4865 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4866 "movq\t{$src, $dst|$dst, $src}", [],
4867 IIC_SSE_MOVDQ>, VEX, VEX_W;
4868 // Recognize "movd" with GR64 destination, but encode as a "movq"
4869 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4870 "movd\t{$src, $dst|$dst, $src}", [],
4871 IIC_SSE_MOVDQ>, VEX, VEX_W;
4873 // Instructions for the disassembler
4874 // xr = XMM register
4877 let Predicates = [HasAVX] in
4878 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4879 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4880 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4881 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4883 //===---------------------------------------------------------------------===//
4884 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4885 //===---------------------------------------------------------------------===//
4886 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4887 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4888 X86MemOperand x86memop> {
4889 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4890 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4891 [(set RC:$dst, (vt (OpNode RC:$src)))],
4893 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4894 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4895 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4899 let Predicates = [HasAVX] in {
4900 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4901 v4f32, VR128, memopv4f32, f128mem>, VEX;
4902 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4903 v4f32, VR128, memopv4f32, f128mem>, VEX;
4904 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4905 v8f32, VR256, memopv8f32, f256mem>, VEX;
4906 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4907 v8f32, VR256, memopv8f32, f256mem>, VEX;
4909 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4910 memopv4f32, f128mem>;
4911 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4912 memopv4f32, f128mem>;
4914 let Predicates = [HasAVX] in {
4915 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4916 (VMOVSHDUPrr VR128:$src)>;
4917 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4918 (VMOVSHDUPrm addr:$src)>;
4919 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4920 (VMOVSLDUPrr VR128:$src)>;
4921 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4922 (VMOVSLDUPrm addr:$src)>;
4923 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4924 (VMOVSHDUPYrr VR256:$src)>;
4925 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4926 (VMOVSHDUPYrm addr:$src)>;
4927 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4928 (VMOVSLDUPYrr VR256:$src)>;
4929 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4930 (VMOVSLDUPYrm addr:$src)>;
4933 let Predicates = [HasSSE3] in {
4934 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4935 (MOVSHDUPrr VR128:$src)>;
4936 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4937 (MOVSHDUPrm addr:$src)>;
4938 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4939 (MOVSLDUPrr VR128:$src)>;
4940 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4941 (MOVSLDUPrm addr:$src)>;
4944 //===---------------------------------------------------------------------===//
4945 // SSE3 - Replicate Double FP - MOVDDUP
4946 //===---------------------------------------------------------------------===//
4948 multiclass sse3_replicate_dfp<string OpcodeStr> {
4949 let neverHasSideEffects = 1 in
4950 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4951 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4952 [], IIC_SSE_MOV_LH>;
4953 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4954 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4957 (scalar_to_vector (loadf64 addr:$src)))))],
4961 // FIXME: Merge with above classe when there're patterns for the ymm version
4962 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4963 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4964 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4965 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4966 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4967 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4970 (scalar_to_vector (loadf64 addr:$src)))))]>;
4973 let Predicates = [HasAVX] in {
4974 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4975 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4978 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4980 let Predicates = [HasAVX] in {
4981 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4982 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4983 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4984 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4985 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4986 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4987 def : Pat<(X86Movddup (bc_v2f64
4988 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4989 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4992 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4993 (VMOVDDUPYrm addr:$src)>;
4994 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4995 (VMOVDDUPYrm addr:$src)>;
4996 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4997 (VMOVDDUPYrm addr:$src)>;
4998 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4999 (VMOVDDUPYrr VR256:$src)>;
5002 let Predicates = [HasSSE3] in {
5003 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5004 (MOVDDUPrm addr:$src)>;
5005 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5006 (MOVDDUPrm addr:$src)>;
5007 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5008 (MOVDDUPrm addr:$src)>;
5009 def : Pat<(X86Movddup (bc_v2f64
5010 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5011 (MOVDDUPrm addr:$src)>;
5014 //===---------------------------------------------------------------------===//
5015 // SSE3 - Move Unaligned Integer
5016 //===---------------------------------------------------------------------===//
5018 let Predicates = [HasAVX] in {
5019 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5020 "vlddqu\t{$src, $dst|$dst, $src}",
5021 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5022 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5023 "vlddqu\t{$src, $dst|$dst, $src}",
5024 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5026 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5027 "lddqu\t{$src, $dst|$dst, $src}",
5028 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5031 //===---------------------------------------------------------------------===//
5032 // SSE3 - Arithmetic
5033 //===---------------------------------------------------------------------===//
5035 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5036 X86MemOperand x86memop, OpndItins itins,
5038 def rr : I<0xD0, MRMSrcReg,
5039 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5041 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5042 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5043 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
5044 def rm : I<0xD0, MRMSrcMem,
5045 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5047 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5048 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5049 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5052 let Predicates = [HasAVX] in {
5053 let ExeDomain = SSEPackedSingle in {
5054 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5055 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5056 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5057 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5059 let ExeDomain = SSEPackedDouble in {
5060 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5061 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5062 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5063 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5066 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5067 let ExeDomain = SSEPackedSingle in
5068 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5069 f128mem, SSE_ALU_F32P>, TB, XD;
5070 let ExeDomain = SSEPackedDouble in
5071 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5072 f128mem, SSE_ALU_F64P>, TB, OpSize;
5075 //===---------------------------------------------------------------------===//
5076 // SSE3 Instructions
5077 //===---------------------------------------------------------------------===//
5080 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5081 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5082 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5084 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5086 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5088 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5090 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5091 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5092 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5093 IIC_SSE_HADDSUB_RM>;
5095 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5096 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5097 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5099 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5100 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5101 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5103 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5105 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5106 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5107 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5108 IIC_SSE_HADDSUB_RM>;
5111 let Predicates = [HasAVX] in {
5112 let ExeDomain = SSEPackedSingle in {
5113 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5114 X86fhadd, 0>, VEX_4V;
5115 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5116 X86fhsub, 0>, VEX_4V;
5117 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5118 X86fhadd, 0>, VEX_4V;
5119 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5120 X86fhsub, 0>, VEX_4V;
5122 let ExeDomain = SSEPackedDouble in {
5123 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5124 X86fhadd, 0>, VEX_4V;
5125 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5126 X86fhsub, 0>, VEX_4V;
5127 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5128 X86fhadd, 0>, VEX_4V;
5129 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5130 X86fhsub, 0>, VEX_4V;
5134 let Constraints = "$src1 = $dst" in {
5135 let ExeDomain = SSEPackedSingle in {
5136 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5137 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5139 let ExeDomain = SSEPackedDouble in {
5140 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5141 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5145 //===---------------------------------------------------------------------===//
5146 // SSSE3 - Packed Absolute Instructions
5147 //===---------------------------------------------------------------------===//
5150 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5151 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5152 Intrinsic IntId128> {
5153 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5155 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5156 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5159 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5161 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5164 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5168 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5169 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5170 Intrinsic IntId256> {
5171 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5173 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5174 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5177 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5179 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5182 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5185 let Predicates = [HasAVX] in {
5186 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5187 int_x86_ssse3_pabs_b_128>, VEX;
5188 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5189 int_x86_ssse3_pabs_w_128>, VEX;
5190 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5191 int_x86_ssse3_pabs_d_128>, VEX;
5194 let Predicates = [HasAVX2] in {
5195 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5196 int_x86_avx2_pabs_b>, VEX;
5197 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5198 int_x86_avx2_pabs_w>, VEX;
5199 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5200 int_x86_avx2_pabs_d>, VEX;
5203 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5204 int_x86_ssse3_pabs_b_128>;
5205 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5206 int_x86_ssse3_pabs_w_128>;
5207 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5208 int_x86_ssse3_pabs_d_128>;
5210 //===---------------------------------------------------------------------===//
5211 // SSSE3 - Packed Binary Operator Instructions
5212 //===---------------------------------------------------------------------===//
5214 def SSE_PHADDSUBD : OpndItins<
5215 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5217 def SSE_PHADDSUBSW : OpndItins<
5218 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5220 def SSE_PHADDSUBW : OpndItins<
5221 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5223 def SSE_PSHUFB : OpndItins<
5224 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5226 def SSE_PSIGN : OpndItins<
5227 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5229 def SSE_PMULHRSW : OpndItins<
5230 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5233 /// SS3I_binop_rm - Simple SSSE3 bin op
5234 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5235 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5236 X86MemOperand x86memop, OpndItins itins,
5238 let isCommutable = 1 in
5239 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5240 (ins RC:$src1, RC:$src2),
5242 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5243 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5244 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5246 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5247 (ins RC:$src1, x86memop:$src2),
5249 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5250 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5252 (OpVT (OpNode RC:$src1,
5253 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5256 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5257 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5258 Intrinsic IntId128, OpndItins itins,
5260 let isCommutable = 1 in
5261 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5262 (ins VR128:$src1, VR128:$src2),
5264 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5265 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5266 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5268 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5269 (ins VR128:$src1, i128mem:$src2),
5271 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5272 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5274 (IntId128 VR128:$src1,
5275 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5278 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5279 Intrinsic IntId256> {
5280 let isCommutable = 1 in
5281 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5282 (ins VR256:$src1, VR256:$src2),
5283 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5284 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5286 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5287 (ins VR256:$src1, i256mem:$src2),
5288 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5290 (IntId256 VR256:$src1,
5291 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5294 let ImmT = NoImm, Predicates = [HasAVX] in {
5295 let isCommutable = 0 in {
5296 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5297 memopv2i64, i128mem,
5298 SSE_PHADDSUBW, 0>, VEX_4V;
5299 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5300 memopv2i64, i128mem,
5301 SSE_PHADDSUBD, 0>, VEX_4V;
5302 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5303 memopv2i64, i128mem,
5304 SSE_PHADDSUBW, 0>, VEX_4V;
5305 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5306 memopv2i64, i128mem,
5307 SSE_PHADDSUBD, 0>, VEX_4V;
5308 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5309 memopv2i64, i128mem,
5310 SSE_PSIGN, 0>, VEX_4V;
5311 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5312 memopv2i64, i128mem,
5313 SSE_PSIGN, 0>, VEX_4V;
5314 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5315 memopv2i64, i128mem,
5316 SSE_PSIGN, 0>, VEX_4V;
5317 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5318 memopv2i64, i128mem,
5319 SSE_PSHUFB, 0>, VEX_4V;
5320 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5321 int_x86_ssse3_phadd_sw_128,
5322 SSE_PHADDSUBSW, 0>, VEX_4V;
5323 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5324 int_x86_ssse3_phsub_sw_128,
5325 SSE_PHADDSUBSW, 0>, VEX_4V;
5326 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5327 int_x86_ssse3_pmadd_ub_sw_128,
5328 SSE_PMADD, 0>, VEX_4V;
5330 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5331 int_x86_ssse3_pmul_hr_sw_128,
5332 SSE_PMULHRSW, 0>, VEX_4V;
5335 let ImmT = NoImm, Predicates = [HasAVX2] in {
5336 let isCommutable = 0 in {
5337 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5338 memopv4i64, i256mem,
5339 SSE_PHADDSUBW, 0>, VEX_4V;
5340 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5341 memopv4i64, i256mem,
5342 SSE_PHADDSUBW, 0>, VEX_4V;
5343 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5344 memopv4i64, i256mem,
5345 SSE_PHADDSUBW, 0>, VEX_4V;
5346 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5347 memopv4i64, i256mem,
5348 SSE_PHADDSUBW, 0>, VEX_4V;
5349 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5350 memopv4i64, i256mem,
5351 SSE_PHADDSUBW, 0>, VEX_4V;
5352 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5353 memopv4i64, i256mem,
5354 SSE_PHADDSUBW, 0>, VEX_4V;
5355 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5356 memopv4i64, i256mem,
5357 SSE_PHADDSUBW, 0>, VEX_4V;
5358 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5359 memopv4i64, i256mem,
5360 SSE_PHADDSUBW, 0>, VEX_4V;
5361 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5362 int_x86_avx2_phadd_sw>, VEX_4V;
5363 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5364 int_x86_avx2_phsub_sw>, VEX_4V;
5365 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5366 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5368 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5369 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5372 // None of these have i8 immediate fields.
5373 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5374 let isCommutable = 0 in {
5375 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5376 memopv2i64, i128mem, SSE_PHADDSUBW>;
5377 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5378 memopv2i64, i128mem, SSE_PHADDSUBD>;
5379 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5380 memopv2i64, i128mem, SSE_PHADDSUBW>;
5381 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5382 memopv2i64, i128mem, SSE_PHADDSUBD>;
5383 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5384 memopv2i64, i128mem, SSE_PSIGN>;
5385 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5386 memopv2i64, i128mem, SSE_PSIGN>;
5387 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5388 memopv2i64, i128mem, SSE_PSIGN>;
5389 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5390 memopv2i64, i128mem, SSE_PSHUFB>;
5391 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5392 int_x86_ssse3_phadd_sw_128,
5394 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5395 int_x86_ssse3_phsub_sw_128,
5397 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5398 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5400 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5401 int_x86_ssse3_pmul_hr_sw_128,
5405 //===---------------------------------------------------------------------===//
5406 // SSSE3 - Packed Align Instruction Patterns
5407 //===---------------------------------------------------------------------===//
5409 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5410 let neverHasSideEffects = 1 in {
5411 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5412 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5414 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5416 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5417 [], IIC_SSE_PALIGNR>, OpSize;
5419 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5420 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5422 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5424 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5425 [], IIC_SSE_PALIGNR>, OpSize;
5429 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5430 let neverHasSideEffects = 1 in {
5431 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5432 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5434 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5437 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5438 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5440 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5445 let Predicates = [HasAVX] in
5446 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5447 let Predicates = [HasAVX2] in
5448 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5449 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5450 defm PALIGN : ssse3_palign<"palignr">;
5452 let Predicates = [HasAVX2] in {
5453 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5454 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5455 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5456 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5457 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5458 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5459 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5460 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5463 let Predicates = [HasAVX] in {
5464 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5465 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5466 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5467 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5468 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5469 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5470 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5471 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5474 let Predicates = [HasSSSE3] in {
5475 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5476 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5477 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5478 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5479 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5480 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5481 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5482 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5485 //===---------------------------------------------------------------------===//
5486 // SSSE3 - Thread synchronization
5487 //===---------------------------------------------------------------------===//
5489 let usesCustomInserter = 1 in {
5490 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5491 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5492 Requires<[HasSSE3]>;
5493 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5494 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5495 Requires<[HasSSE3]>;
5498 let Uses = [EAX, ECX, EDX] in
5499 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5500 TB, Requires<[HasSSE3]>;
5501 let Uses = [ECX, EAX] in
5502 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5503 TB, Requires<[HasSSE3]>;
5505 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5506 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5508 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5509 Requires<[In32BitMode]>;
5510 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5511 Requires<[In64BitMode]>;
5513 //===----------------------------------------------------------------------===//
5514 // SSE4.1 - Packed Move with Sign/Zero Extend
5515 //===----------------------------------------------------------------------===//
5517 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5518 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5519 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5520 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5522 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5523 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5525 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5529 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5531 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5532 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5533 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5535 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5536 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5537 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5540 let Predicates = [HasAVX] in {
5541 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5543 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5545 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5547 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5549 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5551 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5555 let Predicates = [HasAVX2] in {
5556 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5557 int_x86_avx2_pmovsxbw>, VEX;
5558 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5559 int_x86_avx2_pmovsxwd>, VEX;
5560 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5561 int_x86_avx2_pmovsxdq>, VEX;
5562 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5563 int_x86_avx2_pmovzxbw>, VEX;
5564 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5565 int_x86_avx2_pmovzxwd>, VEX;
5566 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5567 int_x86_avx2_pmovzxdq>, VEX;
5570 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5571 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5572 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5573 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5574 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5575 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5577 let Predicates = [HasAVX] in {
5578 // Common patterns involving scalar load.
5579 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5580 (VPMOVSXBWrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5582 (VPMOVSXBWrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5585 (VPMOVSXWDrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5587 (VPMOVSXWDrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5590 (VPMOVSXDQrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5592 (VPMOVSXDQrm addr:$src)>;
5594 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5595 (VPMOVZXBWrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5597 (VPMOVZXBWrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5600 (VPMOVZXWDrm addr:$src)>;
5601 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5602 (VPMOVZXWDrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5605 (VPMOVZXDQrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5607 (VPMOVZXDQrm addr:$src)>;
5610 let Predicates = [HasSSE41] in {
5611 // Common patterns involving scalar load.
5612 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5613 (PMOVSXBWrm addr:$src)>;
5614 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5615 (PMOVSXBWrm addr:$src)>;
5617 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5618 (PMOVSXWDrm addr:$src)>;
5619 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5620 (PMOVSXWDrm addr:$src)>;
5622 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5623 (PMOVSXDQrm addr:$src)>;
5624 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5625 (PMOVSXDQrm addr:$src)>;
5627 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5628 (PMOVZXBWrm addr:$src)>;
5629 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5630 (PMOVZXBWrm addr:$src)>;
5632 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5633 (PMOVZXWDrm addr:$src)>;
5634 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5635 (PMOVZXWDrm addr:$src)>;
5637 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5638 (PMOVZXDQrm addr:$src)>;
5639 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5640 (PMOVZXDQrm addr:$src)>;
5643 let Predicates = [HasAVX2] in {
5644 let AddedComplexity = 15 in {
5645 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5646 (VPMOVZXDQYrr VR128:$src)>;
5647 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5648 (VPMOVZXWDYrr VR128:$src)>;
5651 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5652 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5655 let Predicates = [HasAVX] in {
5656 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5657 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5660 let Predicates = [HasSSE41] in {
5661 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5662 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5666 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5667 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5668 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5669 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5671 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5672 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5674 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5678 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5680 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5681 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5682 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5684 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5685 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5687 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5691 let Predicates = [HasAVX] in {
5692 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5694 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5696 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5698 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5702 let Predicates = [HasAVX2] in {
5703 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5704 int_x86_avx2_pmovsxbd>, VEX;
5705 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5706 int_x86_avx2_pmovsxwq>, VEX;
5707 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5708 int_x86_avx2_pmovzxbd>, VEX;
5709 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5710 int_x86_avx2_pmovzxwq>, VEX;
5713 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5714 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5715 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5716 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5718 let Predicates = [HasAVX] in {
5719 // Common patterns involving scalar load
5720 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5721 (VPMOVSXBDrm addr:$src)>;
5722 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5723 (VPMOVSXWQrm addr:$src)>;
5725 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5726 (VPMOVZXBDrm addr:$src)>;
5727 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5728 (VPMOVZXWQrm addr:$src)>;
5731 let Predicates = [HasSSE41] in {
5732 // Common patterns involving scalar load
5733 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5734 (PMOVSXBDrm addr:$src)>;
5735 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5736 (PMOVSXWQrm addr:$src)>;
5738 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5739 (PMOVZXBDrm addr:$src)>;
5740 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5741 (PMOVZXWQrm addr:$src)>;
5744 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5745 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5746 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5747 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5749 // Expecting a i16 load any extended to i32 value.
5750 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5751 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5752 [(set VR128:$dst, (IntId (bitconvert
5753 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5757 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5759 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5760 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5761 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5763 // Expecting a i16 load any extended to i32 value.
5764 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5765 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5766 [(set VR256:$dst, (IntId (bitconvert
5767 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5771 let Predicates = [HasAVX] in {
5772 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5774 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5777 let Predicates = [HasAVX2] in {
5778 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5779 int_x86_avx2_pmovsxbq>, VEX;
5780 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5781 int_x86_avx2_pmovzxbq>, VEX;
5783 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5784 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5786 let Predicates = [HasAVX] in {
5787 // Common patterns involving scalar load
5788 def : Pat<(int_x86_sse41_pmovsxbq
5789 (bitconvert (v4i32 (X86vzmovl
5790 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5791 (VPMOVSXBQrm addr:$src)>;
5793 def : Pat<(int_x86_sse41_pmovzxbq
5794 (bitconvert (v4i32 (X86vzmovl
5795 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5796 (VPMOVZXBQrm addr:$src)>;
5799 let Predicates = [HasSSE41] in {
5800 // Common patterns involving scalar load
5801 def : Pat<(int_x86_sse41_pmovsxbq
5802 (bitconvert (v4i32 (X86vzmovl
5803 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5804 (PMOVSXBQrm addr:$src)>;
5806 def : Pat<(int_x86_sse41_pmovzxbq
5807 (bitconvert (v4i32 (X86vzmovl
5808 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5809 (PMOVZXBQrm addr:$src)>;
5812 //===----------------------------------------------------------------------===//
5813 // SSE4.1 - Extract Instructions
5814 //===----------------------------------------------------------------------===//
5816 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5817 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5818 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5819 (ins VR128:$src1, i32i8imm:$src2),
5820 !strconcat(OpcodeStr,
5821 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5822 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5824 let neverHasSideEffects = 1, mayStore = 1 in
5825 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5826 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5827 !strconcat(OpcodeStr,
5828 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5831 // There's an AssertZext in the way of writing the store pattern
5832 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5835 let Predicates = [HasAVX] in {
5836 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5837 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5838 (ins VR128:$src1, i32i8imm:$src2),
5839 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5842 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5845 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5846 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5847 let neverHasSideEffects = 1, mayStore = 1 in
5848 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5849 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5850 !strconcat(OpcodeStr,
5851 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5854 // There's an AssertZext in the way of writing the store pattern
5855 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5858 let Predicates = [HasAVX] in
5859 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5861 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5864 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5865 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5866 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5867 (ins VR128:$src1, i32i8imm:$src2),
5868 !strconcat(OpcodeStr,
5869 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5871 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5872 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5873 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5874 !strconcat(OpcodeStr,
5875 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5876 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5877 addr:$dst)]>, OpSize;
5880 let Predicates = [HasAVX] in
5881 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5883 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5885 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5886 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5887 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5888 (ins VR128:$src1, i32i8imm:$src2),
5889 !strconcat(OpcodeStr,
5890 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5892 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5893 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5894 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5895 !strconcat(OpcodeStr,
5896 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5897 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5898 addr:$dst)]>, OpSize, REX_W;
5901 let Predicates = [HasAVX] in
5902 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5904 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5906 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5908 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5909 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5910 (ins VR128:$src1, i32i8imm:$src2),
5911 !strconcat(OpcodeStr,
5912 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5914 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5916 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5917 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5918 !strconcat(OpcodeStr,
5919 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5920 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5921 addr:$dst)]>, OpSize;
5924 let ExeDomain = SSEPackedSingle in {
5925 let Predicates = [HasAVX] in {
5926 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5927 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5928 (ins VR128:$src1, i32i8imm:$src2),
5929 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5932 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5935 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5936 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5939 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5941 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5944 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5945 Requires<[HasSSE41]>;
5947 //===----------------------------------------------------------------------===//
5948 // SSE4.1 - Insert Instructions
5949 //===----------------------------------------------------------------------===//
5951 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5952 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5953 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5955 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5957 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5959 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5960 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5961 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5963 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5965 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5967 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5968 imm:$src3))]>, OpSize;
5971 let Predicates = [HasAVX] in
5972 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5973 let Constraints = "$src1 = $dst" in
5974 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5976 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5977 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5978 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5980 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5982 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5984 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5986 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5987 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5989 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5991 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5993 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5994 imm:$src3)))]>, OpSize;
5997 let Predicates = [HasAVX] in
5998 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5999 let Constraints = "$src1 = $dst" in
6000 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6002 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6003 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6004 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6006 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6008 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6010 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6012 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6013 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6015 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6017 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6019 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6020 imm:$src3)))]>, OpSize;
6023 let Predicates = [HasAVX] in
6024 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6025 let Constraints = "$src1 = $dst" in
6026 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6028 // insertps has a few different modes, there's the first two here below which
6029 // are optimized inserts that won't zero arbitrary elements in the destination
6030 // vector. The next one matches the intrinsic and could zero arbitrary elements
6031 // in the target vector.
6032 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6033 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6034 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6036 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6038 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6040 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6042 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6043 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6045 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6047 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6049 (X86insrtps VR128:$src1,
6050 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6051 imm:$src3))]>, OpSize;
6054 let ExeDomain = SSEPackedSingle in {
6055 let Predicates = [HasAVX] in
6056 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6057 let Constraints = "$src1 = $dst" in
6058 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6061 //===----------------------------------------------------------------------===//
6062 // SSE4.1 - Round Instructions
6063 //===----------------------------------------------------------------------===//
6065 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6066 X86MemOperand x86memop, RegisterClass RC,
6067 PatFrag mem_frag32, PatFrag mem_frag64,
6068 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6069 let ExeDomain = SSEPackedSingle in {
6070 // Intrinsic operation, reg.
6071 // Vector intrinsic operation, reg
6072 def PSr : SS4AIi8<opcps, MRMSrcReg,
6073 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6074 !strconcat(OpcodeStr,
6075 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6076 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6079 // Vector intrinsic operation, mem
6080 def PSm : SS4AIi8<opcps, MRMSrcMem,
6081 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6082 !strconcat(OpcodeStr,
6083 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6085 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6087 } // ExeDomain = SSEPackedSingle
6089 let ExeDomain = SSEPackedDouble in {
6090 // Vector intrinsic operation, reg
6091 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6092 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6093 !strconcat(OpcodeStr,
6094 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6095 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6098 // Vector intrinsic operation, mem
6099 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6100 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6101 !strconcat(OpcodeStr,
6102 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6104 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6106 } // ExeDomain = SSEPackedDouble
6109 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6112 Intrinsic F64Int, bit Is2Addr = 1> {
6113 let ExeDomain = GenericDomain in {
6115 def SSr : SS4AIi8<opcss, MRMSrcReg,
6116 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6118 !strconcat(OpcodeStr,
6119 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6120 !strconcat(OpcodeStr,
6121 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6124 // Intrinsic operation, reg.
6125 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6126 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6128 !strconcat(OpcodeStr,
6129 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6130 !strconcat(OpcodeStr,
6131 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6132 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6135 // Intrinsic operation, mem.
6136 def SSm : SS4AIi8<opcss, MRMSrcMem,
6137 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6139 !strconcat(OpcodeStr,
6140 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6141 !strconcat(OpcodeStr,
6142 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6144 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6148 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6149 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6151 !strconcat(OpcodeStr,
6152 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6153 !strconcat(OpcodeStr,
6154 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6157 // Intrinsic operation, reg.
6158 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6159 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6161 !strconcat(OpcodeStr,
6162 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6163 !strconcat(OpcodeStr,
6164 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6165 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6168 // Intrinsic operation, mem.
6169 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6170 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6172 !strconcat(OpcodeStr,
6173 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6174 !strconcat(OpcodeStr,
6175 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6177 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6179 } // ExeDomain = GenericDomain
6182 // FP round - roundss, roundps, roundsd, roundpd
6183 let Predicates = [HasAVX] in {
6185 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6186 memopv4f32, memopv2f64,
6187 int_x86_sse41_round_ps,
6188 int_x86_sse41_round_pd>, VEX;
6189 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6190 memopv8f32, memopv4f64,
6191 int_x86_avx_round_ps_256,
6192 int_x86_avx_round_pd_256>, VEX;
6193 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6194 int_x86_sse41_round_ss,
6195 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6197 def : Pat<(ffloor FR32:$src),
6198 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6199 def : Pat<(f64 (ffloor FR64:$src)),
6200 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6201 def : Pat<(f32 (fnearbyint FR32:$src)),
6202 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6203 def : Pat<(f64 (fnearbyint FR64:$src)),
6204 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6205 def : Pat<(f32 (fceil FR32:$src)),
6206 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6207 def : Pat<(f64 (fceil FR64:$src)),
6208 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6209 def : Pat<(f32 (frint FR32:$src)),
6210 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6211 def : Pat<(f64 (frint FR64:$src)),
6212 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6213 def : Pat<(f32 (ftrunc FR32:$src)),
6214 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6215 def : Pat<(f64 (ftrunc FR64:$src)),
6216 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6219 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6220 memopv4f32, memopv2f64,
6221 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6222 let Constraints = "$src1 = $dst" in
6223 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6224 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6226 def : Pat<(ffloor FR32:$src),
6227 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6228 def : Pat<(f64 (ffloor FR64:$src)),
6229 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6230 def : Pat<(f32 (fnearbyint FR32:$src)),
6231 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6232 def : Pat<(f64 (fnearbyint FR64:$src)),
6233 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6234 def : Pat<(f32 (fceil FR32:$src)),
6235 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6236 def : Pat<(f64 (fceil FR64:$src)),
6237 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6238 def : Pat<(f32 (frint FR32:$src)),
6239 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6240 def : Pat<(f64 (frint FR64:$src)),
6241 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6242 def : Pat<(f32 (ftrunc FR32:$src)),
6243 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6244 def : Pat<(f64 (ftrunc FR64:$src)),
6245 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6247 //===----------------------------------------------------------------------===//
6248 // SSE4.1 - Packed Bit Test
6249 //===----------------------------------------------------------------------===//
6251 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6252 // the intel intrinsic that corresponds to this.
6253 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6254 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6255 "vptest\t{$src2, $src1|$src1, $src2}",
6256 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6258 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6259 "vptest\t{$src2, $src1|$src1, $src2}",
6260 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6263 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6264 "vptest\t{$src2, $src1|$src1, $src2}",
6265 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6267 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6268 "vptest\t{$src2, $src1|$src1, $src2}",
6269 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6273 let Defs = [EFLAGS] in {
6274 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6275 "ptest\t{$src2, $src1|$src1, $src2}",
6276 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6278 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6279 "ptest\t{$src2, $src1|$src1, $src2}",
6280 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6284 // The bit test instructions below are AVX only
6285 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6286 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6287 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6288 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6289 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6290 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6291 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6292 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6296 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6297 let ExeDomain = SSEPackedSingle in {
6298 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6299 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6301 let ExeDomain = SSEPackedDouble in {
6302 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6303 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6307 //===----------------------------------------------------------------------===//
6308 // SSE4.1 - Misc Instructions
6309 //===----------------------------------------------------------------------===//
6311 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6312 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6313 "popcnt{w}\t{$src, $dst|$dst, $src}",
6314 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6316 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6317 "popcnt{w}\t{$src, $dst|$dst, $src}",
6318 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6319 (implicit EFLAGS)]>, OpSize, XS;
6321 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6322 "popcnt{l}\t{$src, $dst|$dst, $src}",
6323 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6325 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6326 "popcnt{l}\t{$src, $dst|$dst, $src}",
6327 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6328 (implicit EFLAGS)]>, XS;
6330 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6331 "popcnt{q}\t{$src, $dst|$dst, $src}",
6332 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6334 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6335 "popcnt{q}\t{$src, $dst|$dst, $src}",
6336 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6337 (implicit EFLAGS)]>, XS;
6342 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6343 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6344 Intrinsic IntId128> {
6345 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6347 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6348 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6349 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6351 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6354 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6357 let Predicates = [HasAVX] in
6358 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6359 int_x86_sse41_phminposuw>, VEX;
6360 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6361 int_x86_sse41_phminposuw>;
6363 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6364 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6365 Intrinsic IntId128, bit Is2Addr = 1> {
6366 let isCommutable = 1 in
6367 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6368 (ins VR128:$src1, VR128:$src2),
6370 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6372 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6373 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6374 (ins VR128:$src1, i128mem:$src2),
6376 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6377 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6379 (IntId128 VR128:$src1,
6380 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6383 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6384 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6385 Intrinsic IntId256> {
6386 let isCommutable = 1 in
6387 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6388 (ins VR256:$src1, VR256:$src2),
6389 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6390 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6391 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6392 (ins VR256:$src1, i256mem:$src2),
6393 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6395 (IntId256 VR256:$src1,
6396 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6399 let Predicates = [HasAVX] in {
6400 let isCommutable = 0 in
6401 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6403 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6405 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6407 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6409 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6411 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6413 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6415 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6417 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6419 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6423 let Predicates = [HasAVX2] in {
6424 let isCommutable = 0 in
6425 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6426 int_x86_avx2_packusdw>, VEX_4V;
6427 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6428 int_x86_avx2_pmins_b>, VEX_4V;
6429 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6430 int_x86_avx2_pmins_d>, VEX_4V;
6431 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6432 int_x86_avx2_pminu_d>, VEX_4V;
6433 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6434 int_x86_avx2_pminu_w>, VEX_4V;
6435 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6436 int_x86_avx2_pmaxs_b>, VEX_4V;
6437 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6438 int_x86_avx2_pmaxs_d>, VEX_4V;
6439 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6440 int_x86_avx2_pmaxu_d>, VEX_4V;
6441 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6442 int_x86_avx2_pmaxu_w>, VEX_4V;
6443 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6444 int_x86_avx2_pmul_dq>, VEX_4V;
6447 let Constraints = "$src1 = $dst" in {
6448 let isCommutable = 0 in
6449 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6450 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6451 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6452 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6453 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6454 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6455 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6456 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6457 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6458 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6461 /// SS48I_binop_rm - Simple SSE41 binary operator.
6462 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6463 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6464 X86MemOperand x86memop, bit Is2Addr = 1> {
6465 let isCommutable = 1 in
6466 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6467 (ins RC:$src1, RC:$src2),
6469 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6470 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6471 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6472 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6473 (ins RC:$src1, x86memop:$src2),
6475 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6476 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6478 (OpVT (OpNode RC:$src1,
6479 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6482 let Predicates = [HasAVX] in {
6483 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6484 memopv2i64, i128mem, 0>, VEX_4V;
6485 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6486 memopv2i64, i128mem, 0>, VEX_4V;
6488 let Predicates = [HasAVX2] in {
6489 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6490 memopv4i64, i256mem, 0>, VEX_4V;
6491 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6492 memopv4i64, i256mem, 0>, VEX_4V;
6495 let Constraints = "$src1 = $dst" in {
6496 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6497 memopv2i64, i128mem>;
6498 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6499 memopv2i64, i128mem>;
6502 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6503 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6504 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6505 X86MemOperand x86memop, bit Is2Addr = 1> {
6506 let isCommutable = 1 in
6507 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6508 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6510 !strconcat(OpcodeStr,
6511 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6512 !strconcat(OpcodeStr,
6513 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6514 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6516 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6517 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6519 !strconcat(OpcodeStr,
6520 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6521 !strconcat(OpcodeStr,
6522 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6525 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6529 let Predicates = [HasAVX] in {
6530 let isCommutable = 0 in {
6531 let ExeDomain = SSEPackedSingle in {
6532 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6533 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6534 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6535 int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
6537 let ExeDomain = SSEPackedDouble in {
6538 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6539 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6540 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6541 int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
6543 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6544 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6545 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6546 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6548 let ExeDomain = SSEPackedSingle in
6549 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6550 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6551 let ExeDomain = SSEPackedDouble in
6552 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6553 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6554 let ExeDomain = SSEPackedSingle in
6555 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6556 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6559 let Predicates = [HasAVX2] in {
6560 let isCommutable = 0 in {
6561 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6562 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6563 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6564 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6568 let Constraints = "$src1 = $dst" in {
6569 let isCommutable = 0 in {
6570 let ExeDomain = SSEPackedSingle in
6571 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6572 VR128, memopv4f32, f128mem>;
6573 let ExeDomain = SSEPackedDouble in
6574 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6575 VR128, memopv2f64, f128mem>;
6576 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6577 VR128, memopv2i64, i128mem>;
6578 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6579 VR128, memopv2i64, i128mem>;
6581 let ExeDomain = SSEPackedSingle in
6582 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6583 VR128, memopv4f32, f128mem>;
6584 let ExeDomain = SSEPackedDouble in
6585 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6586 VR128, memopv2f64, f128mem>;
6589 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6590 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6591 RegisterClass RC, X86MemOperand x86memop,
6592 PatFrag mem_frag, Intrinsic IntId> {
6593 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6594 (ins RC:$src1, RC:$src2, RC:$src3),
6595 !strconcat(OpcodeStr,
6596 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6597 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6598 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6600 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6601 (ins RC:$src1, x86memop:$src2, RC:$src3),
6602 !strconcat(OpcodeStr,
6603 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6605 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6607 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6610 let Predicates = [HasAVX] in {
6611 let ExeDomain = SSEPackedDouble in {
6612 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6613 memopv2f64, int_x86_sse41_blendvpd>;
6614 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6615 memopv4f64, int_x86_avx_blendv_pd_256>;
6616 } // ExeDomain = SSEPackedDouble
6617 let ExeDomain = SSEPackedSingle in {
6618 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6619 memopv4f32, int_x86_sse41_blendvps>;
6620 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6621 memopv8f32, int_x86_avx_blendv_ps_256>;
6622 } // ExeDomain = SSEPackedSingle
6623 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6624 memopv2i64, int_x86_sse41_pblendvb>;
6627 let Predicates = [HasAVX2] in {
6628 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6629 memopv4i64, int_x86_avx2_pblendvb>;
6632 let Predicates = [HasAVX] in {
6633 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6634 (v16i8 VR128:$src2))),
6635 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6636 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6637 (v4i32 VR128:$src2))),
6638 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6639 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6640 (v4f32 VR128:$src2))),
6641 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6642 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6643 (v2i64 VR128:$src2))),
6644 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6645 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6646 (v2f64 VR128:$src2))),
6647 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6648 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6649 (v8i32 VR256:$src2))),
6650 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6651 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6652 (v8f32 VR256:$src2))),
6653 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6654 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6655 (v4i64 VR256:$src2))),
6656 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6657 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6658 (v4f64 VR256:$src2))),
6659 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6661 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6663 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6664 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6666 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6668 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6670 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6671 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6673 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6674 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6676 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6679 let Predicates = [HasAVX2] in {
6680 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6681 (v32i8 VR256:$src2))),
6682 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6683 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6685 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6688 /// SS41I_ternary_int - SSE 4.1 ternary operator
6689 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6690 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6691 X86MemOperand x86memop, Intrinsic IntId> {
6692 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6693 (ins VR128:$src1, VR128:$src2),
6694 !strconcat(OpcodeStr,
6695 "\t{$src2, $dst|$dst, $src2}"),
6696 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6699 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6700 (ins VR128:$src1, x86memop:$src2),
6701 !strconcat(OpcodeStr,
6702 "\t{$src2, $dst|$dst, $src2}"),
6705 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6709 let ExeDomain = SSEPackedDouble in
6710 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6711 int_x86_sse41_blendvpd>;
6712 let ExeDomain = SSEPackedSingle in
6713 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6714 int_x86_sse41_blendvps>;
6715 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6716 int_x86_sse41_pblendvb>;
6718 // Aliases with the implicit xmm0 argument
6719 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6720 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6721 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6722 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6723 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6724 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6725 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6726 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6727 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6728 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6729 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6730 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6732 let Predicates = [HasSSE41] in {
6733 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6734 (v16i8 VR128:$src2))),
6735 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6736 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6737 (v4i32 VR128:$src2))),
6738 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6739 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6740 (v4f32 VR128:$src2))),
6741 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6742 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6743 (v2i64 VR128:$src2))),
6744 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6745 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6746 (v2f64 VR128:$src2))),
6747 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6749 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6751 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6752 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6754 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6755 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6757 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6761 let Predicates = [HasAVX] in
6762 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6763 "vmovntdqa\t{$src, $dst|$dst, $src}",
6764 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6766 let Predicates = [HasAVX2] in
6767 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6768 "vmovntdqa\t{$src, $dst|$dst, $src}",
6769 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6771 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6772 "movntdqa\t{$src, $dst|$dst, $src}",
6773 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6776 //===----------------------------------------------------------------------===//
6777 // SSE4.2 - Compare Instructions
6778 //===----------------------------------------------------------------------===//
6780 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6781 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6782 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6783 X86MemOperand x86memop, bit Is2Addr = 1> {
6784 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6785 (ins RC:$src1, RC:$src2),
6787 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6788 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6789 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6791 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6792 (ins RC:$src1, x86memop:$src2),
6794 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6795 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6797 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6800 let Predicates = [HasAVX] in
6801 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6802 memopv2i64, i128mem, 0>, VEX_4V;
6804 let Predicates = [HasAVX2] in
6805 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6806 memopv4i64, i256mem, 0>, VEX_4V;
6808 let Constraints = "$src1 = $dst" in
6809 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6810 memopv2i64, i128mem>;
6812 //===----------------------------------------------------------------------===//
6813 // SSE4.2 - String/text Processing Instructions
6814 //===----------------------------------------------------------------------===//
6816 // Packed Compare Implicit Length Strings, Return Mask
6817 multiclass pseudo_pcmpistrm<string asm> {
6818 def REG : PseudoI<(outs VR128:$dst),
6819 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6820 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6822 def MEM : PseudoI<(outs VR128:$dst),
6823 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6824 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6825 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6828 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6829 let AddedComplexity = 1 in
6830 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6831 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6834 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6835 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6836 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6837 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6839 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6840 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6841 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6844 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6845 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6846 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6847 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6849 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6850 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6851 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6854 // Packed Compare Explicit Length Strings, Return Mask
6855 multiclass pseudo_pcmpestrm<string asm> {
6856 def REG : PseudoI<(outs VR128:$dst),
6857 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6858 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6859 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6860 def MEM : PseudoI<(outs VR128:$dst),
6861 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6862 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6863 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6866 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6867 let AddedComplexity = 1 in
6868 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6869 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6872 let Predicates = [HasAVX],
6873 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6874 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6875 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6876 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6878 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6879 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6880 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6883 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6884 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6885 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6886 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6888 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6889 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6890 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6893 // Packed Compare Implicit Length Strings, Return Index
6894 let Defs = [ECX, EFLAGS] in {
6895 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6896 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6897 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6898 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6899 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6900 (implicit EFLAGS)]>, OpSize;
6901 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6902 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6903 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6904 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6905 (implicit EFLAGS)]>, OpSize;
6909 let Predicates = [HasAVX] in {
6910 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6912 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6914 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6916 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6918 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6920 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6924 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6925 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6926 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6927 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6928 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6929 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6931 // Packed Compare Explicit Length Strings, Return Index
6932 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6933 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6934 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6935 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6936 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6937 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6938 (implicit EFLAGS)]>, OpSize;
6939 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6940 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6941 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6943 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6944 (implicit EFLAGS)]>, OpSize;
6948 let Predicates = [HasAVX] in {
6949 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6951 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6953 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6955 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6957 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6959 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6963 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6964 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6965 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6966 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6967 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6968 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6970 //===----------------------------------------------------------------------===//
6971 // SSE4.2 - CRC Instructions
6972 //===----------------------------------------------------------------------===//
6974 // No CRC instructions have AVX equivalents
6976 // crc intrinsic instruction
6977 // This set of instructions are only rm, the only difference is the size
6979 let Constraints = "$src1 = $dst" in {
6980 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6981 (ins GR32:$src1, i8mem:$src2),
6982 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_32_8 GR32:$src1,
6985 (load addr:$src2)))]>;
6986 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6987 (ins GR32:$src1, GR8:$src2),
6988 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6991 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6992 (ins GR32:$src1, i16mem:$src2),
6993 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6995 (int_x86_sse42_crc32_32_16 GR32:$src1,
6996 (load addr:$src2)))]>,
6998 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6999 (ins GR32:$src1, GR16:$src2),
7000 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7002 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7004 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7005 (ins GR32:$src1, i32mem:$src2),
7006 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7008 (int_x86_sse42_crc32_32_32 GR32:$src1,
7009 (load addr:$src2)))]>;
7010 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7011 (ins GR32:$src1, GR32:$src2),
7012 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7014 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7015 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7016 (ins GR64:$src1, i8mem:$src2),
7017 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7019 (int_x86_sse42_crc32_64_8 GR64:$src1,
7020 (load addr:$src2)))]>,
7022 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7023 (ins GR64:$src1, GR8:$src2),
7024 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7026 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7028 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7029 (ins GR64:$src1, i64mem:$src2),
7030 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7032 (int_x86_sse42_crc32_64_64 GR64:$src1,
7033 (load addr:$src2)))]>,
7035 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7036 (ins GR64:$src1, GR64:$src2),
7037 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7039 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7043 //===----------------------------------------------------------------------===//
7044 // AES-NI Instructions
7045 //===----------------------------------------------------------------------===//
7047 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7048 Intrinsic IntId128, bit Is2Addr = 1> {
7049 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7050 (ins VR128:$src1, VR128:$src2),
7052 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7053 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7054 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7056 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7057 (ins VR128:$src1, i128mem:$src2),
7059 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7060 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7062 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7065 // Perform One Round of an AES Encryption/Decryption Flow
7066 let Predicates = [HasAVX, HasAES] in {
7067 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7068 int_x86_aesni_aesenc, 0>, VEX_4V;
7069 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7070 int_x86_aesni_aesenclast, 0>, VEX_4V;
7071 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7072 int_x86_aesni_aesdec, 0>, VEX_4V;
7073 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7074 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7077 let Constraints = "$src1 = $dst" in {
7078 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7079 int_x86_aesni_aesenc>;
7080 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7081 int_x86_aesni_aesenclast>;
7082 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7083 int_x86_aesni_aesdec>;
7084 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7085 int_x86_aesni_aesdeclast>;
7088 // Perform the AES InvMixColumn Transformation
7089 let Predicates = [HasAVX, HasAES] in {
7090 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7092 "vaesimc\t{$src1, $dst|$dst, $src1}",
7094 (int_x86_aesni_aesimc VR128:$src1))]>,
7096 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7097 (ins i128mem:$src1),
7098 "vaesimc\t{$src1, $dst|$dst, $src1}",
7099 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7102 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7104 "aesimc\t{$src1, $dst|$dst, $src1}",
7106 (int_x86_aesni_aesimc VR128:$src1))]>,
7108 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7109 (ins i128mem:$src1),
7110 "aesimc\t{$src1, $dst|$dst, $src1}",
7111 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7114 // AES Round Key Generation Assist
7115 let Predicates = [HasAVX, HasAES] in {
7116 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7117 (ins VR128:$src1, i8imm:$src2),
7118 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7120 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7122 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7123 (ins i128mem:$src1, i8imm:$src2),
7124 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7126 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7129 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7130 (ins VR128:$src1, i8imm:$src2),
7131 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7133 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7135 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7136 (ins i128mem:$src1, i8imm:$src2),
7137 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7139 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7142 //===----------------------------------------------------------------------===//
7143 // PCLMUL Instructions
7144 //===----------------------------------------------------------------------===//
7146 // AVX carry-less Multiplication instructions
7147 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7148 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7149 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7151 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7153 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7154 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7155 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7156 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7157 (memopv2i64 addr:$src2), imm:$src3))]>;
7159 // Carry-less Multiplication instructions
7160 let Constraints = "$src1 = $dst" in {
7161 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7162 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7163 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7165 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7167 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7168 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7169 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7170 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7171 (memopv2i64 addr:$src2), imm:$src3))]>;
7172 } // Constraints = "$src1 = $dst"
7175 multiclass pclmul_alias<string asm, int immop> {
7176 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7177 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7179 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7180 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7182 def : InstAlias<!strconcat("vpclmul", asm,
7183 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7184 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7186 def : InstAlias<!strconcat("vpclmul", asm,
7187 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7188 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7190 defm : pclmul_alias<"hqhq", 0x11>;
7191 defm : pclmul_alias<"hqlq", 0x01>;
7192 defm : pclmul_alias<"lqhq", 0x10>;
7193 defm : pclmul_alias<"lqlq", 0x00>;
7195 //===----------------------------------------------------------------------===//
7196 // SSE4A Instructions
7197 //===----------------------------------------------------------------------===//
7199 let Predicates = [HasSSE4A] in {
7201 let Constraints = "$src = $dst" in {
7202 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7203 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7204 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7205 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7206 imm:$idx))]>, TB, OpSize;
7207 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7208 (ins VR128:$src, VR128:$mask),
7209 "extrq\t{$mask, $src|$src, $mask}",
7210 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7211 VR128:$mask))]>, TB, OpSize;
7213 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7214 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7215 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7216 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7217 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7218 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7219 (ins VR128:$src, VR128:$mask),
7220 "insertq\t{$mask, $src|$src, $mask}",
7221 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7222 VR128:$mask))]>, XD;
7225 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7226 "movntss\t{$src, $dst|$dst, $src}",
7227 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7229 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7230 "movntsd\t{$src, $dst|$dst, $src}",
7231 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7234 //===----------------------------------------------------------------------===//
7236 //===----------------------------------------------------------------------===//
7238 //===----------------------------------------------------------------------===//
7239 // VBROADCAST - Load from memory and broadcast to all elements of the
7240 // destination operand
7242 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7243 X86MemOperand x86memop, Intrinsic Int> :
7244 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7245 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7246 [(set RC:$dst, (Int addr:$src))]>, VEX;
7248 // AVX2 adds register forms
7249 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7251 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7252 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7253 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7255 let ExeDomain = SSEPackedSingle in {
7256 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7257 int_x86_avx_vbroadcast_ss>;
7258 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7259 int_x86_avx_vbroadcast_ss_256>;
7261 let ExeDomain = SSEPackedDouble in
7262 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7263 int_x86_avx_vbroadcast_sd_256>;
7264 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7265 int_x86_avx_vbroadcastf128_pd_256>;
7267 let ExeDomain = SSEPackedSingle in {
7268 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7269 int_x86_avx2_vbroadcast_ss_ps>;
7270 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7271 int_x86_avx2_vbroadcast_ss_ps_256>;
7273 let ExeDomain = SSEPackedDouble in
7274 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7275 int_x86_avx2_vbroadcast_sd_pd_256>;
7277 let Predicates = [HasAVX2] in
7278 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7279 int_x86_avx2_vbroadcasti128>;
7281 let Predicates = [HasAVX] in
7282 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7283 (VBROADCASTF128 addr:$src)>;
7286 //===----------------------------------------------------------------------===//
7287 // VINSERTF128 - Insert packed floating-point values
7289 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7290 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7291 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7292 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7295 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7296 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7297 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7301 let Predicates = [HasAVX] in {
7302 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7304 (VINSERTF128rr VR256:$src1, VR128:$src2,
7305 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7306 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7308 (VINSERTF128rr VR256:$src1, VR128:$src2,
7309 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7310 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7312 (VINSERTF128rr VR256:$src1, VR128:$src2,
7313 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7314 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7316 (VINSERTF128rr VR256:$src1, VR128:$src2,
7317 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7318 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7320 (VINSERTF128rr VR256:$src1, VR128:$src2,
7321 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7322 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7324 (VINSERTF128rr VR256:$src1, VR128:$src2,
7325 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7327 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7329 (VINSERTF128rm VR256:$src1, addr:$src2,
7330 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7331 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7333 (VINSERTF128rm VR256:$src1, addr:$src2,
7334 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7335 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7337 (VINSERTF128rm VR256:$src1, addr:$src2,
7338 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7341 //===----------------------------------------------------------------------===//
7342 // VEXTRACTF128 - Extract packed floating-point values
7344 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7345 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7346 (ins VR256:$src1, i8imm:$src2),
7347 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7350 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7351 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7352 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7356 // Extract and store.
7357 let Predicates = [HasAVX] in {
7358 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7359 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7360 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7361 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7362 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7363 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7365 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7366 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7367 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7368 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7369 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7370 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7374 let Predicates = [HasAVX] in {
7375 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7376 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7377 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7378 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7379 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7380 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7382 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7383 (v4f32 (VEXTRACTF128rr
7384 (v8f32 VR256:$src1),
7385 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7386 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7387 (v2f64 (VEXTRACTF128rr
7388 (v4f64 VR256:$src1),
7389 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7390 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7391 (v2i64 (VEXTRACTF128rr
7392 (v4i64 VR256:$src1),
7393 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7394 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7395 (v4i32 (VEXTRACTF128rr
7396 (v8i32 VR256:$src1),
7397 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7398 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7399 (v8i16 (VEXTRACTF128rr
7400 (v16i16 VR256:$src1),
7401 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7402 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7403 (v16i8 (VEXTRACTF128rr
7404 (v32i8 VR256:$src1),
7405 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7408 //===----------------------------------------------------------------------===//
7409 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7411 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7412 Intrinsic IntLd, Intrinsic IntLd256,
7413 Intrinsic IntSt, Intrinsic IntSt256> {
7414 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7415 (ins VR128:$src1, f128mem:$src2),
7416 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7417 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7419 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7420 (ins VR256:$src1, f256mem:$src2),
7421 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7422 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7424 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7425 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7426 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7427 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7428 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7429 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7430 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7431 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7434 let ExeDomain = SSEPackedSingle in
7435 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7436 int_x86_avx_maskload_ps,
7437 int_x86_avx_maskload_ps_256,
7438 int_x86_avx_maskstore_ps,
7439 int_x86_avx_maskstore_ps_256>;
7440 let ExeDomain = SSEPackedDouble in
7441 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7442 int_x86_avx_maskload_pd,
7443 int_x86_avx_maskload_pd_256,
7444 int_x86_avx_maskstore_pd,
7445 int_x86_avx_maskstore_pd_256>;
7447 //===----------------------------------------------------------------------===//
7448 // VPERMIL - Permute Single and Double Floating-Point Values
7450 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7451 RegisterClass RC, X86MemOperand x86memop_f,
7452 X86MemOperand x86memop_i, PatFrag i_frag,
7453 Intrinsic IntVar, ValueType vt> {
7454 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7455 (ins RC:$src1, RC:$src2),
7456 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7457 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7458 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7459 (ins RC:$src1, x86memop_i:$src2),
7460 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7461 [(set RC:$dst, (IntVar RC:$src1,
7462 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7464 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7465 (ins RC:$src1, i8imm:$src2),
7466 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7467 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7468 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7469 (ins x86memop_f:$src1, i8imm:$src2),
7470 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7472 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7475 let ExeDomain = SSEPackedSingle in {
7476 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7477 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7478 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7479 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7481 let ExeDomain = SSEPackedDouble in {
7482 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7483 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7484 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7485 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7488 let Predicates = [HasAVX] in {
7489 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7490 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7491 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7492 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7493 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7495 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7496 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7497 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7499 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7500 (VPERMILPDri VR128:$src1, imm:$imm)>;
7501 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7502 (VPERMILPDmi addr:$src1, imm:$imm)>;
7505 //===----------------------------------------------------------------------===//
7506 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7508 let ExeDomain = SSEPackedSingle in {
7509 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7510 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7511 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7512 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7513 (i8 imm:$src3))))]>, VEX_4V;
7514 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7515 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7516 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7517 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7518 (i8 imm:$src3)))]>, VEX_4V;
7521 let Predicates = [HasAVX] in {
7522 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7523 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7524 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7525 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7526 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7527 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7528 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7529 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7530 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7531 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7533 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7534 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7535 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7536 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7537 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7538 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7539 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7540 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7541 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7542 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7543 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7544 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7545 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7546 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7547 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7548 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7549 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7550 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7553 //===----------------------------------------------------------------------===//
7554 // VZERO - Zero YMM registers
7556 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7557 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7558 // Zero All YMM registers
7559 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7560 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7562 // Zero Upper bits of YMM registers
7563 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7564 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7567 //===----------------------------------------------------------------------===//
7568 // Half precision conversion instructions
7569 //===----------------------------------------------------------------------===//
7570 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7571 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7572 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7573 [(set RC:$dst, (Int VR128:$src))]>,
7575 let neverHasSideEffects = 1, mayLoad = 1 in
7576 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7577 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7580 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7581 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7582 (ins RC:$src1, i32i8imm:$src2),
7583 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7584 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7586 let neverHasSideEffects = 1, mayStore = 1 in
7587 def mr : Ii8<0x1D, MRMDestMem, (outs),
7588 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7589 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7593 let Predicates = [HasAVX, HasF16C] in {
7594 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7595 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7596 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7597 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7600 //===----------------------------------------------------------------------===//
7601 // AVX2 Instructions
7602 //===----------------------------------------------------------------------===//
7604 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7605 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7606 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7607 X86MemOperand x86memop> {
7608 let isCommutable = 1 in
7609 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7610 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7611 !strconcat(OpcodeStr,
7612 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7613 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7615 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7616 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7617 !strconcat(OpcodeStr,
7618 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7621 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7625 let isCommutable = 0 in {
7626 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7627 VR128, memopv2i64, i128mem>;
7628 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7629 VR256, memopv4i64, i256mem>;
7632 //===----------------------------------------------------------------------===//
7633 // VPBROADCAST - Load from memory and broadcast to all elements of the
7634 // destination operand
7636 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7637 X86MemOperand x86memop, PatFrag ld_frag,
7638 Intrinsic Int128, Intrinsic Int256> {
7639 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7640 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7641 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7642 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7643 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7645 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7646 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7647 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7648 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7649 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7650 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7652 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7655 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7656 int_x86_avx2_pbroadcastb_128,
7657 int_x86_avx2_pbroadcastb_256>;
7658 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7659 int_x86_avx2_pbroadcastw_128,
7660 int_x86_avx2_pbroadcastw_256>;
7661 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7662 int_x86_avx2_pbroadcastd_128,
7663 int_x86_avx2_pbroadcastd_256>;
7664 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7665 int_x86_avx2_pbroadcastq_128,
7666 int_x86_avx2_pbroadcastq_256>;
7668 let Predicates = [HasAVX2] in {
7669 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7670 (VPBROADCASTBrm addr:$src)>;
7671 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7672 (VPBROADCASTBYrm addr:$src)>;
7673 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7674 (VPBROADCASTWrm addr:$src)>;
7675 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7676 (VPBROADCASTWYrm addr:$src)>;
7677 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7678 (VPBROADCASTDrm addr:$src)>;
7679 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7680 (VPBROADCASTDYrm addr:$src)>;
7681 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7682 (VPBROADCASTQrm addr:$src)>;
7683 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7684 (VPBROADCASTQYrm addr:$src)>;
7686 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7687 (VPBROADCASTBrr VR128:$src)>;
7688 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7689 (VPBROADCASTBYrr VR128:$src)>;
7690 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7691 (VPBROADCASTWrr VR128:$src)>;
7692 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7693 (VPBROADCASTWYrr VR128:$src)>;
7694 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7695 (VPBROADCASTDrr VR128:$src)>;
7696 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7697 (VPBROADCASTDYrr VR128:$src)>;
7698 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7699 (VPBROADCASTQrr VR128:$src)>;
7700 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7701 (VPBROADCASTQYrr VR128:$src)>;
7702 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7703 (VBROADCASTSSrr VR128:$src)>;
7704 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7705 (VBROADCASTSSYrr VR128:$src)>;
7706 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7707 (VPBROADCASTQrr VR128:$src)>;
7708 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7709 (VBROADCASTSDYrr VR128:$src)>;
7711 // Provide fallback in case the load node that is used in the patterns above
7712 // is used by additional users, which prevents the pattern selection.
7713 let AddedComplexity = 20 in {
7714 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7716 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7717 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7719 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7720 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7722 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd))>;
7724 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7726 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7727 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7729 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7730 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7732 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd))>;
7736 // AVX1 broadcast patterns
7737 let Predicates = [HasAVX] in {
7738 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7739 (VBROADCASTSSYrm addr:$src)>;
7740 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7741 (VBROADCASTSDYrm addr:$src)>;
7742 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7743 (VBROADCASTSSYrm addr:$src)>;
7744 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7745 (VBROADCASTSDYrm addr:$src)>;
7746 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7747 (VBROADCASTSSrm addr:$src)>;
7748 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7749 (VBROADCASTSSrm addr:$src)>;
7751 // Provide fallback in case the load node that is used in the patterns above
7752 // is used by additional users, which prevents the pattern selection.
7753 let AddedComplexity = 20 in {
7754 // 128bit broadcasts:
7755 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7757 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0)>;
7758 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7759 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7761 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0),
7764 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss),
7766 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7767 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7769 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
7773 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
7776 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7778 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0)>;
7779 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7780 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7782 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0),
7785 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss),
7787 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7788 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7790 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
7794 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
7799 //===----------------------------------------------------------------------===//
7800 // VPERM - Permute instructions
7803 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7805 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7806 (ins VR256:$src1, VR256:$src2),
7807 !strconcat(OpcodeStr,
7808 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7810 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7811 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7812 (ins VR256:$src1, i256mem:$src2),
7813 !strconcat(OpcodeStr,
7814 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7816 (OpVT (X86VPermv VR256:$src1,
7817 (bitconvert (mem_frag addr:$src2)))))]>,
7821 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7822 let ExeDomain = SSEPackedSingle in
7823 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7825 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7827 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7828 (ins VR256:$src1, i8imm:$src2),
7829 !strconcat(OpcodeStr,
7830 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7832 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7833 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7834 (ins i256mem:$src1, i8imm:$src2),
7835 !strconcat(OpcodeStr,
7836 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7838 (OpVT (X86VPermi (mem_frag addr:$src1),
7839 (i8 imm:$src2))))]>, VEX;
7842 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7843 let ExeDomain = SSEPackedDouble in
7844 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7846 //===----------------------------------------------------------------------===//
7847 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7849 let AddedComplexity = 1 in {
7850 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7851 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7852 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7853 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7854 (i8 imm:$src3))))]>, VEX_4V;
7855 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7856 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7857 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7858 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7859 (i8 imm:$src3)))]>, VEX_4V;
7862 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7863 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7864 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7865 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7866 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7867 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7868 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7870 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7872 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7873 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7874 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7875 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7876 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7878 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7882 //===----------------------------------------------------------------------===//
7883 // VINSERTI128 - Insert packed integer values
7885 let neverHasSideEffects = 1 in {
7886 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7887 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7888 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7891 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7892 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7893 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7897 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7898 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7900 (VINSERTI128rr VR256:$src1, VR128:$src2,
7901 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7902 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7904 (VINSERTI128rr VR256:$src1, VR128:$src2,
7905 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7906 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7908 (VINSERTI128rr VR256:$src1, VR128:$src2,
7909 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7910 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7912 (VINSERTI128rr VR256:$src1, VR128:$src2,
7913 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7916 //===----------------------------------------------------------------------===//
7917 // VEXTRACTI128 - Extract packed integer values
7919 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7920 (ins VR256:$src1, i8imm:$src2),
7921 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7923 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7925 let neverHasSideEffects = 1, mayStore = 1 in
7926 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7927 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7928 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7930 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7931 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7932 (v2i64 (VEXTRACTI128rr
7933 (v4i64 VR256:$src1),
7934 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7935 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7936 (v4i32 (VEXTRACTI128rr
7937 (v8i32 VR256:$src1),
7938 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7939 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7940 (v8i16 (VEXTRACTI128rr
7941 (v16i16 VR256:$src1),
7942 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7943 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7944 (v16i8 (VEXTRACTI128rr
7945 (v32i8 VR256:$src1),
7946 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7949 //===----------------------------------------------------------------------===//
7950 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7952 multiclass avx2_pmovmask<string OpcodeStr,
7953 Intrinsic IntLd128, Intrinsic IntLd256,
7954 Intrinsic IntSt128, Intrinsic IntSt256> {
7955 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7956 (ins VR128:$src1, i128mem:$src2),
7957 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7958 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7959 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7960 (ins VR256:$src1, i256mem:$src2),
7961 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7962 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7963 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7964 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7965 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7966 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7967 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7968 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7969 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7970 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7973 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7974 int_x86_avx2_maskload_d,
7975 int_x86_avx2_maskload_d_256,
7976 int_x86_avx2_maskstore_d,
7977 int_x86_avx2_maskstore_d_256>;
7978 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7979 int_x86_avx2_maskload_q,
7980 int_x86_avx2_maskload_q_256,
7981 int_x86_avx2_maskstore_q,
7982 int_x86_avx2_maskstore_q_256>, VEX_W;
7985 //===----------------------------------------------------------------------===//
7986 // Variable Bit Shifts
7988 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7989 ValueType vt128, ValueType vt256> {
7990 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7991 (ins VR128:$src1, VR128:$src2),
7992 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7994 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7996 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7997 (ins VR128:$src1, i128mem:$src2),
7998 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8000 (vt128 (OpNode VR128:$src1,
8001 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8003 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8004 (ins VR256:$src1, VR256:$src2),
8005 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8007 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8009 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8010 (ins VR256:$src1, i256mem:$src2),
8011 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8013 (vt256 (OpNode VR256:$src1,
8014 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8018 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8019 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8020 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8021 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8022 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8024 //===----------------------------------------------------------------------===//
8025 // VGATHER - GATHER Operations
8026 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8027 X86MemOperand memop128, X86MemOperand memop256> {
8028 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8029 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8030 !strconcat(OpcodeStr,
8031 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8033 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8034 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8035 !strconcat(OpcodeStr,
8036 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8037 []>, VEX_4VOp3, VEX_L;
8040 let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8041 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8042 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8043 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8044 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8045 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8046 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8047 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8048 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;