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,
386 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
387 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
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 AVX1 ISA doesn't 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 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
402 isPseudo = 1, Predicates = [HasAVX] in {
403 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
404 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
407 let Predicates = [HasAVX] in
408 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
410 let Predicates = [HasAVX2] in {
411 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
412 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
413 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
414 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
417 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
418 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
419 let Predicates = [HasAVX1Only] in {
420 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
421 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
422 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
424 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
425 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
426 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
428 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
429 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
430 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
432 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
433 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
434 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
437 // We set canFoldAsLoad because this can be converted to a constant-pool
438 // load of an all-ones value if folding it would be beneficial.
439 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
441 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
442 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
443 let Predicates = [HasAVX2] in
444 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
445 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
449 //===----------------------------------------------------------------------===//
450 // SSE 1 & 2 - Move FP Scalar Instructions
452 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
453 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
454 // is used instead. Register-to-register movss/movsd is not modeled as an
455 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
456 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
457 //===----------------------------------------------------------------------===//
459 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
460 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
461 [(set VR128:$dst, (vt (OpNode VR128:$src1,
462 (scalar_to_vector RC:$src2))))],
465 // Loading from memory automatically zeroing upper bits.
466 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
467 PatFrag mem_pat, string OpcodeStr> :
468 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
469 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
470 [(set RC:$dst, (mem_pat addr:$src))],
474 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
475 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
477 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
478 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
481 // For the disassembler
482 let isCodeGenOnly = 1 in {
483 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
484 (ins VR128:$src1, FR32:$src2),
485 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
488 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
489 (ins VR128:$src1, FR64:$src2),
490 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
495 let canFoldAsLoad = 1, isReMaterializable = 1 in {
496 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
498 let AddedComplexity = 20 in
499 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
503 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
504 "movss\t{$src, $dst|$dst, $src}",
505 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
507 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
508 "movsd\t{$src, $dst|$dst, $src}",
509 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
513 let Constraints = "$src1 = $dst" in {
514 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
515 "movss\t{$src2, $dst|$dst, $src2}">, XS;
516 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
517 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
519 // For the disassembler
520 let isCodeGenOnly = 1 in {
521 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
522 (ins VR128:$src1, FR32:$src2),
523 "movss\t{$src2, $dst|$dst, $src2}", [],
524 IIC_SSE_MOV_S_RR>, XS;
525 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
526 (ins VR128:$src1, FR64:$src2),
527 "movsd\t{$src2, $dst|$dst, $src2}", [],
528 IIC_SSE_MOV_S_RR>, XD;
532 let canFoldAsLoad = 1, isReMaterializable = 1 in {
533 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
535 let AddedComplexity = 20 in
536 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
539 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
540 "movss\t{$src, $dst|$dst, $src}",
541 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
542 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
543 "movsd\t{$src, $dst|$dst, $src}",
544 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
547 let Predicates = [HasAVX] in {
548 let AddedComplexity = 15 in {
549 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
550 // MOVS{S,D} to the lower bits.
551 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
552 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
553 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
554 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
555 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
556 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
557 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
558 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
560 // Move low f32 and clear high bits.
561 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
562 (SUBREG_TO_REG (i32 0),
563 (VMOVSSrr (v4f32 (V_SET0)),
564 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
565 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
566 (SUBREG_TO_REG (i32 0),
567 (VMOVSSrr (v4i32 (V_SET0)),
568 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
571 let AddedComplexity = 20 in {
572 // MOVSSrm zeros the high parts of the register; represent this
573 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
574 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
575 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
576 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
577 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
578 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
579 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
581 // MOVSDrm zeros the high parts of the register; represent this
582 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
583 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
584 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
585 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
586 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
587 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
588 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
589 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
590 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
591 def : Pat<(v2f64 (X86vzload addr:$src)),
592 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
594 // Represent the same patterns above but in the form they appear for
596 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
597 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
598 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
599 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
600 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
601 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
602 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
603 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
604 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
606 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
607 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
608 (SUBREG_TO_REG (i32 0),
609 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
611 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
612 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
613 (SUBREG_TO_REG (i64 0),
614 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
616 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
617 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
618 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
620 // Move low f64 and clear high bits.
621 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
622 (SUBREG_TO_REG (i32 0),
623 (VMOVSDrr (v2f64 (V_SET0)),
624 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
626 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
627 (SUBREG_TO_REG (i32 0),
628 (VMOVSDrr (v2i64 (V_SET0)),
629 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
631 // Extract and store.
632 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
634 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
635 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
637 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
639 // Shuffle with VMOVSS
640 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
641 (VMOVSSrr (v4i32 VR128:$src1),
642 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
643 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
644 (VMOVSSrr (v4f32 VR128:$src1),
645 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
648 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
649 (SUBREG_TO_REG (i32 0),
650 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
651 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
653 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
654 (SUBREG_TO_REG (i32 0),
655 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
656 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
659 // Shuffle with VMOVSD
660 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
661 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
662 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
663 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
664 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
665 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
666 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
667 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
670 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
671 (SUBREG_TO_REG (i32 0),
672 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
673 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
675 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
676 (SUBREG_TO_REG (i32 0),
677 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
678 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
682 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
683 // is during lowering, where it's not possible to recognize the fold cause
684 // it has two uses through a bitcast. One use disappears at isel time and the
685 // fold opportunity reappears.
686 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
687 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
688 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
689 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
690 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
691 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
692 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
693 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
696 let Predicates = [HasSSE1] in {
697 let AddedComplexity = 15 in {
698 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
699 // MOVSS to the lower bits.
700 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
701 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
702 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
703 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
704 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
705 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
708 let AddedComplexity = 20 in {
709 // MOVSSrm already zeros the high parts of the register.
710 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
711 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
712 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
713 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
714 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
715 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
718 // Extract and store.
719 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
721 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
723 // Shuffle with MOVSS
724 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
725 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
726 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
727 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
730 let Predicates = [HasSSE2] in {
731 let AddedComplexity = 15 in {
732 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
733 // MOVSD to the lower bits.
734 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
735 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
738 let AddedComplexity = 20 in {
739 // MOVSDrm already zeros the high parts of the register.
740 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
741 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
742 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
743 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
744 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
745 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
746 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
747 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
748 def : Pat<(v2f64 (X86vzload addr:$src)),
749 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
752 // Extract and store.
753 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
755 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
757 // Shuffle with MOVSD
758 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
759 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
760 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
761 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
762 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
763 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
764 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
765 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
767 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
768 // is during lowering, where it's not possible to recognize the fold cause
769 // it has two uses through a bitcast. One use disappears at isel time and the
770 // fold opportunity reappears.
771 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
772 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
773 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
774 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
775 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
776 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
777 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
778 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
781 //===----------------------------------------------------------------------===//
782 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
783 //===----------------------------------------------------------------------===//
785 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
786 X86MemOperand x86memop, PatFrag ld_frag,
787 string asm, Domain d,
789 bit IsReMaterializable = 1> {
790 let neverHasSideEffects = 1 in
791 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
792 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
793 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
794 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
795 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
796 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
799 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
800 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
802 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
803 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
805 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
806 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
808 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
809 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
812 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
813 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
815 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
816 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
818 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
819 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
821 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
822 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
824 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
825 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
827 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
828 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
830 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
831 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
833 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
834 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
837 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
838 "movaps\t{$src, $dst|$dst, $src}",
839 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
840 IIC_SSE_MOVA_P_MR>, VEX;
841 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
842 "movapd\t{$src, $dst|$dst, $src}",
843 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
844 IIC_SSE_MOVA_P_MR>, VEX;
845 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
846 "movups\t{$src, $dst|$dst, $src}",
847 [(store (v4f32 VR128:$src), addr:$dst)],
848 IIC_SSE_MOVU_P_MR>, VEX;
849 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
850 "movupd\t{$src, $dst|$dst, $src}",
851 [(store (v2f64 VR128:$src), addr:$dst)],
852 IIC_SSE_MOVU_P_MR>, VEX;
853 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
854 "movaps\t{$src, $dst|$dst, $src}",
855 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
856 IIC_SSE_MOVA_P_MR>, VEX;
857 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
858 "movapd\t{$src, $dst|$dst, $src}",
859 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
860 IIC_SSE_MOVA_P_MR>, VEX;
861 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
862 "movups\t{$src, $dst|$dst, $src}",
863 [(store (v8f32 VR256:$src), addr:$dst)],
864 IIC_SSE_MOVU_P_MR>, VEX;
865 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
866 "movupd\t{$src, $dst|$dst, $src}",
867 [(store (v4f64 VR256:$src), addr:$dst)],
868 IIC_SSE_MOVU_P_MR>, VEX;
871 let isCodeGenOnly = 1 in {
872 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
874 "movaps\t{$src, $dst|$dst, $src}", [],
875 IIC_SSE_MOVA_P_RR>, VEX;
876 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
878 "movapd\t{$src, $dst|$dst, $src}", [],
879 IIC_SSE_MOVA_P_RR>, VEX;
880 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
882 "movups\t{$src, $dst|$dst, $src}", [],
883 IIC_SSE_MOVU_P_RR>, VEX;
884 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
886 "movupd\t{$src, $dst|$dst, $src}", [],
887 IIC_SSE_MOVU_P_RR>, VEX;
888 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
890 "movaps\t{$src, $dst|$dst, $src}", [],
891 IIC_SSE_MOVA_P_RR>, VEX;
892 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
894 "movapd\t{$src, $dst|$dst, $src}", [],
895 IIC_SSE_MOVA_P_RR>, VEX;
896 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
898 "movups\t{$src, $dst|$dst, $src}", [],
899 IIC_SSE_MOVU_P_RR>, VEX;
900 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
902 "movupd\t{$src, $dst|$dst, $src}", [],
903 IIC_SSE_MOVU_P_RR>, VEX;
906 let Predicates = [HasAVX] in {
907 def : Pat<(v8i32 (X86vzmovl
908 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
909 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
910 def : Pat<(v4i64 (X86vzmovl
911 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
912 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
913 def : Pat<(v8f32 (X86vzmovl
914 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
915 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
916 def : Pat<(v4f64 (X86vzmovl
917 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
918 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
922 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
923 (VMOVUPSYmr addr:$dst, VR256:$src)>;
924 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
925 (VMOVUPDYmr addr:$dst, VR256:$src)>;
927 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
928 "movaps\t{$src, $dst|$dst, $src}",
929 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
931 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
932 "movapd\t{$src, $dst|$dst, $src}",
933 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
935 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
936 "movups\t{$src, $dst|$dst, $src}",
937 [(store (v4f32 VR128:$src), addr:$dst)],
939 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
940 "movupd\t{$src, $dst|$dst, $src}",
941 [(store (v2f64 VR128:$src), addr:$dst)],
945 let isCodeGenOnly = 1 in {
946 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
947 "movaps\t{$src, $dst|$dst, $src}", [],
949 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
950 "movapd\t{$src, $dst|$dst, $src}", [],
952 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
953 "movups\t{$src, $dst|$dst, $src}", [],
955 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
956 "movupd\t{$src, $dst|$dst, $src}", [],
960 let Predicates = [HasAVX] in {
961 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
962 (VMOVUPSmr addr:$dst, VR128:$src)>;
963 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
964 (VMOVUPDmr addr:$dst, VR128:$src)>;
967 let Predicates = [HasSSE1] in
968 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
969 (MOVUPSmr addr:$dst, VR128:$src)>;
970 let Predicates = [HasSSE2] in
971 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
972 (MOVUPDmr addr:$dst, VR128:$src)>;
974 // Use vmovaps/vmovups for AVX integer load/store.
975 let Predicates = [HasAVX] in {
976 // 128-bit load/store
977 def : Pat<(alignedloadv2i64 addr:$src),
978 (VMOVAPSrm addr:$src)>;
979 def : Pat<(loadv2i64 addr:$src),
980 (VMOVUPSrm addr:$src)>;
982 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
983 (VMOVAPSmr addr:$dst, VR128:$src)>;
984 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
985 (VMOVAPSmr addr:$dst, VR128:$src)>;
986 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
987 (VMOVAPSmr addr:$dst, VR128:$src)>;
988 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
989 (VMOVAPSmr addr:$dst, VR128:$src)>;
990 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
991 (VMOVUPSmr addr:$dst, VR128:$src)>;
992 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
993 (VMOVUPSmr addr:$dst, VR128:$src)>;
994 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
995 (VMOVUPSmr addr:$dst, VR128:$src)>;
996 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
997 (VMOVUPSmr addr:$dst, VR128:$src)>;
999 // 256-bit load/store
1000 def : Pat<(alignedloadv4i64 addr:$src),
1001 (VMOVAPSYrm addr:$src)>;
1002 def : Pat<(loadv4i64 addr:$src),
1003 (VMOVUPSYrm addr:$src)>;
1004 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1005 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1006 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1007 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1008 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1009 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1010 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1011 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1012 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1013 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1014 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1015 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1016 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1017 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1018 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1019 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1022 // Use movaps / movups for SSE integer load / store (one byte shorter).
1023 // The instructions selected below are then converted to MOVDQA/MOVDQU
1024 // during the SSE domain pass.
1025 let Predicates = [HasSSE1] in {
1026 def : Pat<(alignedloadv2i64 addr:$src),
1027 (MOVAPSrm addr:$src)>;
1028 def : Pat<(loadv2i64 addr:$src),
1029 (MOVUPSrm addr:$src)>;
1031 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1032 (MOVAPSmr addr:$dst, VR128:$src)>;
1033 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1034 (MOVAPSmr addr:$dst, VR128:$src)>;
1035 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1036 (MOVAPSmr addr:$dst, VR128:$src)>;
1037 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1038 (MOVAPSmr addr:$dst, VR128:$src)>;
1039 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1040 (MOVUPSmr addr:$dst, VR128:$src)>;
1041 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1042 (MOVUPSmr addr:$dst, VR128:$src)>;
1043 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1044 (MOVUPSmr addr:$dst, VR128:$src)>;
1045 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1046 (MOVUPSmr addr:$dst, VR128:$src)>;
1049 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1050 // bits are disregarded. FIXME: Set encoding to pseudo!
1051 let neverHasSideEffects = 1 in {
1052 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1053 "movaps\t{$src, $dst|$dst, $src}", [],
1054 IIC_SSE_MOVA_P_RR>, VEX;
1055 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1056 "movapd\t{$src, $dst|$dst, $src}", [],
1057 IIC_SSE_MOVA_P_RR>, VEX;
1058 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1059 "movaps\t{$src, $dst|$dst, $src}", [],
1061 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1062 "movapd\t{$src, $dst|$dst, $src}", [],
1066 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1067 // bits are disregarded. FIXME: Set encoding to pseudo!
1068 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1069 let isCodeGenOnly = 1 in {
1070 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1071 "movaps\t{$src, $dst|$dst, $src}",
1072 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1073 IIC_SSE_MOVA_P_RM>, VEX;
1074 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1075 "movapd\t{$src, $dst|$dst, $src}",
1076 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1077 IIC_SSE_MOVA_P_RM>, VEX;
1079 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1080 "movaps\t{$src, $dst|$dst, $src}",
1081 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1083 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1084 "movapd\t{$src, $dst|$dst, $src}",
1085 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1089 //===----------------------------------------------------------------------===//
1090 // SSE 1 & 2 - Move Low packed FP Instructions
1091 //===----------------------------------------------------------------------===//
1093 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1094 SDNode psnode, SDNode pdnode, string base_opc,
1095 string asm_opr, InstrItinClass itin> {
1096 def PSrm : PI<opc, MRMSrcMem,
1097 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1098 !strconcat(base_opc, "s", asm_opr),
1101 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1102 itin, SSEPackedSingle>, TB;
1104 def PDrm : PI<opc, MRMSrcMem,
1105 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1106 !strconcat(base_opc, "d", asm_opr),
1107 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1108 (scalar_to_vector (loadf64 addr:$src2)))))],
1109 itin, SSEPackedDouble>, TB, OpSize;
1112 let AddedComplexity = 20 in {
1113 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1114 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1115 IIC_SSE_MOV_LH>, VEX_4V;
1117 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1118 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1119 "\t{$src2, $dst|$dst, $src2}",
1123 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1124 "movlps\t{$src, $dst|$dst, $src}",
1125 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1126 (iPTR 0))), addr:$dst)],
1127 IIC_SSE_MOV_LH>, VEX;
1128 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1129 "movlpd\t{$src, $dst|$dst, $src}",
1130 [(store (f64 (vector_extract (v2f64 VR128:$src),
1131 (iPTR 0))), addr:$dst)],
1132 IIC_SSE_MOV_LH>, VEX;
1133 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1134 "movlps\t{$src, $dst|$dst, $src}",
1135 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1136 (iPTR 0))), addr:$dst)],
1138 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1139 "movlpd\t{$src, $dst|$dst, $src}",
1140 [(store (f64 (vector_extract (v2f64 VR128:$src),
1141 (iPTR 0))), addr:$dst)],
1144 let Predicates = [HasAVX] in {
1145 // Shuffle with VMOVLPS
1146 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1147 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1148 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1149 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1151 // Shuffle with VMOVLPD
1152 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1153 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1154 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1155 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1158 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1160 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1161 def : Pat<(store (v4i32 (X86Movlps
1162 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1163 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1164 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1166 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1167 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1169 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1172 let Predicates = [HasSSE1] in {
1173 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1174 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1175 (iPTR 0))), addr:$src1),
1176 (MOVLPSmr addr:$src1, VR128:$src2)>;
1178 // Shuffle with MOVLPS
1179 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1180 (MOVLPSrm VR128:$src1, addr:$src2)>;
1181 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1182 (MOVLPSrm VR128:$src1, addr:$src2)>;
1183 def : Pat<(X86Movlps VR128:$src1,
1184 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1185 (MOVLPSrm VR128:$src1, addr:$src2)>;
1188 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1190 (MOVLPSmr addr:$src1, VR128:$src2)>;
1191 def : Pat<(store (v4i32 (X86Movlps
1192 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1194 (MOVLPSmr addr:$src1, VR128:$src2)>;
1197 let Predicates = [HasSSE2] in {
1198 // Shuffle with MOVLPD
1199 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1200 (MOVLPDrm VR128:$src1, addr:$src2)>;
1201 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1202 (MOVLPDrm VR128:$src1, addr:$src2)>;
1205 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1207 (MOVLPDmr addr:$src1, VR128:$src2)>;
1208 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1210 (MOVLPDmr addr:$src1, VR128:$src2)>;
1213 //===----------------------------------------------------------------------===//
1214 // SSE 1 & 2 - Move Hi packed FP Instructions
1215 //===----------------------------------------------------------------------===//
1217 let AddedComplexity = 20 in {
1218 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1219 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1220 IIC_SSE_MOV_LH>, VEX_4V;
1222 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1223 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1224 "\t{$src2, $dst|$dst, $src2}",
1228 // v2f64 extract element 1 is always custom lowered to unpack high to low
1229 // and extract element 0 so the non-store version isn't too horrible.
1230 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1231 "movhps\t{$src, $dst|$dst, $src}",
1232 [(store (f64 (vector_extract
1233 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1234 (bc_v2f64 (v4f32 VR128:$src))),
1235 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1236 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1237 "movhpd\t{$src, $dst|$dst, $src}",
1238 [(store (f64 (vector_extract
1239 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1240 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1241 def MOVHPSmr : PSI<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>;
1247 def MOVHPDmr : PDI<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>;
1253 let Predicates = [HasAVX] in {
1255 def : Pat<(X86Movlhps VR128:$src1,
1256 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1257 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1258 def : Pat<(X86Movlhps VR128:$src1,
1259 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1260 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1262 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1263 // is during lowering, where it's not possible to recognize the load fold
1264 // cause it has two uses through a bitcast. One use disappears at isel time
1265 // and the fold opportunity reappears.
1266 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1267 (scalar_to_vector (loadf64 addr:$src2)))),
1268 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1271 let Predicates = [HasSSE1] in {
1273 def : Pat<(X86Movlhps VR128:$src1,
1274 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1275 (MOVHPSrm VR128:$src1, addr:$src2)>;
1276 def : Pat<(X86Movlhps VR128:$src1,
1277 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1278 (MOVHPSrm VR128:$src1, addr:$src2)>;
1281 let Predicates = [HasSSE2] in {
1282 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1283 // is during lowering, where it's not possible to recognize the load fold
1284 // cause it has two uses through a bitcast. One use disappears at isel time
1285 // and the fold opportunity reappears.
1286 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1287 (scalar_to_vector (loadf64 addr:$src2)))),
1288 (MOVHPDrm VR128:$src1, addr:$src2)>;
1291 //===----------------------------------------------------------------------===//
1292 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1293 //===----------------------------------------------------------------------===//
1295 let AddedComplexity = 20 in {
1296 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1297 (ins VR128:$src1, VR128:$src2),
1298 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1300 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1303 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1304 (ins VR128:$src1, VR128:$src2),
1305 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1307 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1311 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1312 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1313 (ins VR128:$src1, VR128:$src2),
1314 "movlhps\t{$src2, $dst|$dst, $src2}",
1316 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1318 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1319 (ins VR128:$src1, VR128:$src2),
1320 "movhlps\t{$src2, $dst|$dst, $src2}",
1322 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1326 let Predicates = [HasAVX] in {
1328 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1329 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1330 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1331 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1334 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1335 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1338 let Predicates = [HasSSE1] in {
1340 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1341 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1342 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1343 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1346 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1347 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1350 //===----------------------------------------------------------------------===//
1351 // SSE 1 & 2 - Conversion Instructions
1352 //===----------------------------------------------------------------------===//
1354 def SSE_CVT_PD : OpndItins<
1355 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1358 def SSE_CVT_PS : OpndItins<
1359 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1362 def SSE_CVT_Scalar : OpndItins<
1363 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1366 def SSE_CVT_SS2SI_32 : OpndItins<
1367 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1370 def SSE_CVT_SS2SI_64 : OpndItins<
1371 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1374 def SSE_CVT_SD2SI : OpndItins<
1375 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1378 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1379 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1380 string asm, OpndItins itins> {
1381 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1382 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1384 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1385 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1389 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1390 X86MemOperand x86memop, string asm, Domain d,
1392 let neverHasSideEffects = 1 in {
1393 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1396 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1401 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1402 X86MemOperand x86memop, string asm> {
1403 let neverHasSideEffects = 1 in {
1404 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1405 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1407 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1408 (ins DstRC:$src1, x86memop:$src),
1409 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1410 } // neverHasSideEffects = 1
1413 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1414 "cvttss2si\t{$src, $dst|$dst, $src}",
1417 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1418 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1420 XS, VEX, VEX_W, VEX_LIG;
1421 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1422 "cvttsd2si\t{$src, $dst|$dst, $src}",
1425 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1426 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1428 XD, VEX, VEX_W, VEX_LIG;
1430 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1431 // register, but the same isn't true when only using memory operands,
1432 // provide other assembly "l" and "q" forms to address this explicitly
1433 // where appropriate to do so.
1434 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1435 XS, VEX_4V, VEX_LIG;
1436 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1437 XS, VEX_4V, VEX_W, VEX_LIG;
1438 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1439 XD, VEX_4V, VEX_LIG;
1440 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1441 XD, VEX_4V, VEX_W, VEX_LIG;
1443 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1444 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1445 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1446 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1448 let Predicates = [HasAVX], AddedComplexity = 1 in {
1449 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1450 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1451 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1452 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1453 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1454 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1455 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1456 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1458 def : Pat<(f32 (sint_to_fp GR32:$src)),
1459 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1460 def : Pat<(f32 (sint_to_fp GR64:$src)),
1461 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1462 def : Pat<(f64 (sint_to_fp GR32:$src)),
1463 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1464 def : Pat<(f64 (sint_to_fp GR64:$src)),
1465 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1468 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1469 "cvttss2si\t{$src, $dst|$dst, $src}",
1470 SSE_CVT_SS2SI_32>, XS;
1471 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1472 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1473 SSE_CVT_SS2SI_64>, XS, REX_W;
1474 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1475 "cvttsd2si\t{$src, $dst|$dst, $src}",
1477 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1478 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1479 SSE_CVT_SD2SI>, XD, REX_W;
1480 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1481 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1482 SSE_CVT_Scalar>, XS;
1483 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1484 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1485 SSE_CVT_Scalar>, XS, REX_W;
1486 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1487 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1488 SSE_CVT_Scalar>, XD;
1489 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1490 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1491 SSE_CVT_Scalar>, XD, REX_W;
1493 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1494 // and/or XMM operand(s).
1496 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1497 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1498 string asm, OpndItins itins> {
1499 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1500 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1501 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1502 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1503 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1504 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1507 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1508 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1509 PatFrag ld_frag, string asm, OpndItins itins,
1511 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1513 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1514 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1515 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1517 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1518 (ins DstRC:$src1, x86memop:$src2),
1520 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1521 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1522 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1526 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1527 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
1528 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1529 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1530 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
1531 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1533 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1534 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1535 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1536 sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1539 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1540 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1541 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1542 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1543 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1544 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1546 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1547 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1548 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1549 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1550 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1551 SSE_CVT_Scalar, 0>, XD,
1554 let Constraints = "$src1 = $dst" in {
1555 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1556 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1557 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1558 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1559 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1560 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1561 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1562 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1563 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1564 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1565 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1566 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1571 // Aliases for intrinsics
1572 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1573 ssmem, sse_load_f32, "cvttss2si",
1574 SSE_CVT_SS2SI_32>, XS, VEX;
1575 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1576 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1577 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1579 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1580 sdmem, sse_load_f64, "cvttsd2si",
1581 SSE_CVT_SD2SI>, XD, VEX;
1582 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1583 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1584 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1586 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1587 ssmem, sse_load_f32, "cvttss2si",
1588 SSE_CVT_SS2SI_32>, XS;
1589 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1590 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1591 "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
1592 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1593 sdmem, sse_load_f64, "cvttsd2si",
1595 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1596 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1597 "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1599 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1600 ssmem, sse_load_f32, "cvtss2si{l}",
1601 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1602 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1603 ssmem, sse_load_f32, "cvtss2si{q}",
1604 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1606 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1607 ssmem, sse_load_f32, "cvtss2si{l}",
1608 SSE_CVT_SS2SI_32>, XS;
1609 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1610 ssmem, sse_load_f32, "cvtss2si{q}",
1611 SSE_CVT_SS2SI_64>, XS, REX_W;
1613 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1614 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1615 SSEPackedSingle, SSE_CVT_PS>,
1616 TB, VEX, Requires<[HasAVX]>;
1617 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1618 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1619 SSEPackedSingle, SSE_CVT_PS>,
1620 TB, VEX, Requires<[HasAVX]>;
1622 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1623 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1624 SSEPackedSingle, SSE_CVT_PS>,
1625 TB, Requires<[HasSSE2]>;
1629 // Convert scalar double to scalar single
1630 let neverHasSideEffects = 1 in {
1631 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1632 (ins FR64:$src1, FR64:$src2),
1633 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1634 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1636 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1637 (ins FR64:$src1, f64mem:$src2),
1638 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1639 [], IIC_SSE_CVT_Scalar_RM>,
1640 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1643 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1646 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1647 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1648 [(set FR32:$dst, (fround FR64:$src))],
1649 IIC_SSE_CVT_Scalar_RR>;
1650 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1651 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1652 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1653 IIC_SSE_CVT_Scalar_RM>,
1655 Requires<[HasSSE2, OptForSize]>;
1657 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1658 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1659 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1661 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1662 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
1663 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1664 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1665 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1666 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1667 VR128:$src1, sse_load_f64:$src2))],
1668 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
1670 let Constraints = "$src1 = $dst" in {
1671 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1672 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1673 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1675 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1676 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[HasSSE2]>;
1677 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1678 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1679 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1680 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1681 VR128:$src1, sse_load_f64:$src2))],
1682 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[HasSSE2]>;
1685 // Convert scalar single to scalar double
1686 // SSE2 instructions with XS prefix
1687 let neverHasSideEffects = 1 in {
1688 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1689 (ins FR32:$src1, FR32:$src2),
1690 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1691 [], IIC_SSE_CVT_Scalar_RR>,
1692 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1694 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1695 (ins FR32:$src1, f32mem:$src2),
1696 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1697 [], IIC_SSE_CVT_Scalar_RM>,
1698 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1701 let AddedComplexity = 1 in { // give AVX priority
1702 def : Pat<(f64 (fextend FR32:$src)),
1703 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1704 def : Pat<(fextend (loadf32 addr:$src)),
1705 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1707 def : Pat<(extloadf32 addr:$src),
1708 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1709 Requires<[HasAVX, OptForSize]>;
1710 def : Pat<(extloadf32 addr:$src),
1711 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1712 Requires<[HasAVX, OptForSpeed]>;
1713 } // AddedComplexity = 1
1715 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1716 "cvtss2sd\t{$src, $dst|$dst, $src}",
1717 [(set FR64:$dst, (fextend FR32:$src))],
1718 IIC_SSE_CVT_Scalar_RR>, XS,
1719 Requires<[HasSSE2]>;
1720 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1721 "cvtss2sd\t{$src, $dst|$dst, $src}",
1722 [(set FR64:$dst, (extloadf32 addr:$src))],
1723 IIC_SSE_CVT_Scalar_RM>, XS,
1724 Requires<[HasSSE2, OptForSize]>;
1726 // extload f32 -> f64. This matches load+fextend because we have a hack in
1727 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1729 // Since these loads aren't folded into the fextend, we have to match it
1731 def : Pat<(fextend (loadf32 addr:$src)),
1732 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1733 def : Pat<(extloadf32 addr:$src),
1734 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1736 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1737 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1738 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1740 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1741 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
1742 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1743 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1744 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1746 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1747 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
1748 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1749 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1750 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1751 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1753 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1754 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[HasSSE2]>;
1755 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1756 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1757 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1759 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1760 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[HasSSE2]>;
1763 // Convert packed single/double fp to doubleword
1764 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1765 "cvtps2dq\t{$src, $dst|$dst, $src}",
1766 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1767 IIC_SSE_CVT_PS_RR>, VEX;
1768 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1769 "cvtps2dq\t{$src, $dst|$dst, $src}",
1771 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1772 IIC_SSE_CVT_PS_RM>, VEX;
1773 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1774 "cvtps2dq\t{$src, $dst|$dst, $src}",
1776 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1777 IIC_SSE_CVT_PS_RR>, VEX;
1778 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1779 "cvtps2dq\t{$src, $dst|$dst, $src}",
1781 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1782 IIC_SSE_CVT_PS_RM>, VEX;
1783 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1784 "cvtps2dq\t{$src, $dst|$dst, $src}",
1785 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1787 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1788 "cvtps2dq\t{$src, $dst|$dst, $src}",
1790 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1794 // Convert Packed Double FP to Packed DW Integers
1795 let Predicates = [HasAVX] in {
1796 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1797 // register, but the same isn't true when using memory operands instead.
1798 // Provide other assembly rr and rm forms to address this explicitly.
1799 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1800 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1801 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1805 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1806 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1807 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1808 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1810 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1813 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1814 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1816 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX;
1817 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1818 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1820 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1822 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1823 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1826 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1827 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1829 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1831 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1832 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1833 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1836 // Convert with truncation packed single/double fp to doubleword
1837 // SSE2 packed instructions with XS prefix
1838 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1839 "cvttps2dq\t{$src, $dst|$dst, $src}",
1841 (int_x86_sse2_cvttps2dq VR128:$src))],
1842 IIC_SSE_CVT_PS_RR>, VEX;
1843 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1844 "cvttps2dq\t{$src, $dst|$dst, $src}",
1845 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1846 (memopv4f32 addr:$src)))],
1847 IIC_SSE_CVT_PS_RM>, VEX;
1848 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1849 "cvttps2dq\t{$src, $dst|$dst, $src}",
1851 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1852 IIC_SSE_CVT_PS_RR>, VEX;
1853 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1854 "cvttps2dq\t{$src, $dst|$dst, $src}",
1855 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1856 (memopv8f32 addr:$src)))],
1857 IIC_SSE_CVT_PS_RM>, VEX;
1859 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1860 "cvttps2dq\t{$src, $dst|$dst, $src}",
1861 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1863 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1864 "cvttps2dq\t{$src, $dst|$dst, $src}",
1866 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1869 let Predicates = [HasAVX] in {
1870 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1871 (VCVTDQ2PSrr VR128:$src)>;
1872 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1873 (VCVTDQ2PSrm addr:$src)>;
1875 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1876 (VCVTDQ2PSrr VR128:$src)>;
1877 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1878 (VCVTDQ2PSrm addr:$src)>;
1880 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1881 (VCVTTPS2DQrr VR128:$src)>;
1882 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1883 (VCVTTPS2DQrm addr:$src)>;
1885 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1886 (VCVTDQ2PSYrr VR256:$src)>;
1887 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1888 (VCVTDQ2PSYrm addr:$src)>;
1890 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1891 (VCVTTPS2DQYrr VR256:$src)>;
1892 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1893 (VCVTTPS2DQYrm addr:$src)>;
1896 let Predicates = [HasSSE2] in {
1897 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1898 (CVTDQ2PSrr VR128:$src)>;
1899 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1900 (CVTDQ2PSrm addr:$src)>;
1902 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1903 (CVTDQ2PSrr VR128:$src)>;
1904 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1905 (CVTDQ2PSrm addr:$src)>;
1907 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1908 (CVTTPS2DQrr VR128:$src)>;
1909 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1910 (CVTTPS2DQrm addr:$src)>;
1913 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1914 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1916 (int_x86_sse2_cvttpd2dq VR128:$src))],
1917 IIC_SSE_CVT_PD_RR>, VEX;
1919 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1920 // register, but the same isn't true when using memory operands instead.
1921 // Provide other assembly rr and rm forms to address this explicitly.
1924 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1925 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1926 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1927 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1928 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1929 (memopv2f64 addr:$src)))],
1930 IIC_SSE_CVT_PD_RM>, VEX;
1933 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1934 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1936 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
1937 IIC_SSE_CVT_PD_RR>, VEX;
1938 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1939 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1941 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
1942 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1943 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1944 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1946 let Predicates = [HasAVX] in {
1947 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1948 (VCVTTPD2DQYrr VR256:$src)>;
1949 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1950 (VCVTTPD2DQYrm addr:$src)>;
1951 } // Predicates = [HasAVX]
1953 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1954 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1955 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1957 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1958 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1959 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1960 (memopv2f64 addr:$src)))],
1963 // Convert packed single to packed double
1964 let Predicates = [HasAVX] in {
1965 // SSE2 instructions without OpSize prefix
1966 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1967 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1968 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1969 IIC_SSE_CVT_PD_RR>, TB, VEX;
1970 let neverHasSideEffects = 1, mayLoad = 1 in
1971 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1972 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1973 IIC_SSE_CVT_PD_RM>, TB, VEX;
1974 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1975 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1977 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
1978 IIC_SSE_CVT_PD_RR>, TB, VEX;
1979 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1980 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1982 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
1983 IIC_SSE_CVT_PD_RM>, TB, VEX;
1986 let Predicates = [HasSSE2] in {
1987 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1988 "cvtps2pd\t{$src, $dst|$dst, $src}",
1989 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1990 IIC_SSE_CVT_PD_RR>, TB;
1991 let neverHasSideEffects = 1, mayLoad = 1 in
1992 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1993 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
1994 IIC_SSE_CVT_PD_RM>, TB;
1997 // Convert Packed DW Integers to Packed Double FP
1998 let Predicates = [HasAVX] in {
1999 let neverHasSideEffects = 1, mayLoad = 1 in
2000 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2001 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2003 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2004 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2006 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2007 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2008 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2010 (int_x86_avx_cvtdq2_pd_256
2011 (bitconvert (memopv2i64 addr:$src))))]>, VEX;
2012 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2013 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2015 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX;
2018 let neverHasSideEffects = 1, mayLoad = 1 in
2019 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2020 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2022 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2023 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2024 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2027 // AVX 256-bit register conversion intrinsics
2028 let Predicates = [HasAVX] in {
2029 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2030 (VCVTDQ2PDYrr VR128:$src)>;
2031 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2032 (VCVTDQ2PDYrm addr:$src)>;
2033 } // Predicates = [HasAVX]
2035 // Convert packed double to packed single
2036 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2037 // register, but the same isn't true when using memory operands instead.
2038 // Provide other assembly rr and rm forms to address this explicitly.
2039 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2040 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2041 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2042 IIC_SSE_CVT_PD_RR>, VEX;
2045 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2046 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2047 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2048 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2050 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2051 IIC_SSE_CVT_PD_RM>, VEX;
2054 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2055 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2057 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2058 IIC_SSE_CVT_PD_RR>, VEX;
2059 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2060 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2062 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2063 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2064 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2065 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2067 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2068 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2069 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2071 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2072 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2074 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2078 // AVX 256-bit register conversion intrinsics
2079 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2080 // whenever possible to avoid declaring two versions of each one.
2081 let Predicates = [HasAVX] in {
2082 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2083 (VCVTDQ2PSYrr VR256:$src)>;
2084 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2085 (VCVTDQ2PSYrm addr:$src)>;
2087 // Match fround and fextend for 128/256-bit conversions
2088 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2089 (VCVTPD2PSYrr VR256:$src)>;
2090 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2091 (VCVTPD2PSYrm addr:$src)>;
2093 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2094 (VCVTPS2PDrr VR128:$src)>;
2095 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2096 (VCVTPS2PDYrr VR128:$src)>;
2097 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2098 (VCVTPS2PDYrm addr:$src)>;
2101 let Predicates = [HasSSE2] in {
2102 // Match fextend for 128 conversions
2103 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2104 (CVTPS2PDrr VR128:$src)>;
2107 //===----------------------------------------------------------------------===//
2108 // SSE 1 & 2 - Compare Instructions
2109 //===----------------------------------------------------------------------===//
2111 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2112 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2113 Operand CC, SDNode OpNode, ValueType VT,
2114 PatFrag ld_frag, string asm, string asm_alt,
2116 def rr : SIi8<0xC2, MRMSrcReg,
2117 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2118 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2120 def rm : SIi8<0xC2, MRMSrcMem,
2121 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2122 [(set RC:$dst, (OpNode (VT RC:$src1),
2123 (ld_frag addr:$src2), imm:$cc))],
2126 // Accept explicit immediate argument form instead of comparison code.
2127 let neverHasSideEffects = 1 in {
2128 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2129 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2130 IIC_SSE_ALU_F32S_RR>;
2132 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2133 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2134 IIC_SSE_ALU_F32S_RM>;
2138 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2139 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2140 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2142 XS, VEX_4V, VEX_LIG;
2143 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2144 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2145 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2146 SSE_ALU_F32S>, // same latency as 32 bit compare
2147 XD, VEX_4V, VEX_LIG;
2149 let Constraints = "$src1 = $dst" in {
2150 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2151 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2152 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2154 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2155 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2156 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2157 SSE_ALU_F32S>, // same latency as 32 bit compare
2161 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2162 Intrinsic Int, string asm, OpndItins itins> {
2163 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2164 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2165 [(set VR128:$dst, (Int VR128:$src1,
2166 VR128:$src, imm:$cc))],
2168 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2169 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2170 [(set VR128:$dst, (Int VR128:$src1,
2171 (load addr:$src), imm:$cc))],
2175 // Aliases to match intrinsics which expect XMM operand(s).
2176 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2177 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2180 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2181 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2182 SSE_ALU_F32S>, // same latency as f32
2184 let Constraints = "$src1 = $dst" in {
2185 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2186 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2188 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2189 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2190 SSE_ALU_F32S>, // same latency as f32
2195 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2196 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2197 ValueType vt, X86MemOperand x86memop,
2198 PatFrag ld_frag, string OpcodeStr, Domain d> {
2199 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2200 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2201 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2202 IIC_SSE_COMIS_RR, d>;
2203 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2204 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2205 [(set EFLAGS, (OpNode (vt RC:$src1),
2206 (ld_frag addr:$src2)))],
2207 IIC_SSE_COMIS_RM, d>;
2210 let Defs = [EFLAGS] in {
2211 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2212 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2213 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2214 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2216 let Pattern = []<dag> in {
2217 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2218 "comiss", SSEPackedSingle>, TB, VEX,
2220 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2221 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2225 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2226 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2227 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2228 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2230 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2231 load, "comiss", SSEPackedSingle>, TB, VEX;
2232 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2233 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2234 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2235 "ucomiss", SSEPackedSingle>, TB;
2236 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2237 "ucomisd", SSEPackedDouble>, TB, OpSize;
2239 let Pattern = []<dag> in {
2240 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2241 "comiss", SSEPackedSingle>, TB;
2242 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2243 "comisd", SSEPackedDouble>, TB, OpSize;
2246 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2247 load, "ucomiss", SSEPackedSingle>, TB;
2248 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2249 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2251 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2252 "comiss", SSEPackedSingle>, TB;
2253 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2254 "comisd", SSEPackedDouble>, TB, OpSize;
2255 } // Defs = [EFLAGS]
2257 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2258 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2259 Operand CC, Intrinsic Int, string asm,
2260 string asm_alt, Domain d> {
2261 def rri : PIi8<0xC2, MRMSrcReg,
2262 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2263 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2264 IIC_SSE_CMPP_RR, d>;
2265 def rmi : PIi8<0xC2, MRMSrcMem,
2266 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2267 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2268 IIC_SSE_CMPP_RM, d>;
2270 // Accept explicit immediate argument form instead of comparison code.
2271 let neverHasSideEffects = 1 in {
2272 def rri_alt : PIi8<0xC2, MRMSrcReg,
2273 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2274 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2275 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2276 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2277 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2281 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2282 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2283 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2284 SSEPackedSingle>, TB, VEX_4V;
2285 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2286 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2287 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2288 SSEPackedDouble>, TB, OpSize, VEX_4V;
2289 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2290 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2291 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2292 SSEPackedSingle>, TB, VEX_4V;
2293 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2294 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2295 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2296 SSEPackedDouble>, TB, OpSize, VEX_4V;
2297 let Constraints = "$src1 = $dst" in {
2298 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2299 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2300 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2301 SSEPackedSingle>, TB;
2302 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2303 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2304 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2305 SSEPackedDouble>, TB, OpSize;
2308 let Predicates = [HasAVX] in {
2309 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2310 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2311 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2312 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2313 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2314 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2315 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2316 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2318 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2319 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2320 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2321 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2322 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2323 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2324 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2325 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2328 let Predicates = [HasSSE1] in {
2329 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2330 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2331 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2332 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2335 let Predicates = [HasSSE2] in {
2336 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2337 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2338 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2339 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2342 //===----------------------------------------------------------------------===//
2343 // SSE 1 & 2 - Shuffle Instructions
2344 //===----------------------------------------------------------------------===//
2346 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2347 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2348 ValueType vt, string asm, PatFrag mem_frag,
2349 Domain d, bit IsConvertibleToThreeAddress = 0> {
2350 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2351 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2352 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2353 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2354 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2355 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2356 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2357 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2358 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2361 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2362 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2363 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2364 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2365 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2366 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2367 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2368 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2369 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2370 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2371 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2372 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2374 let Constraints = "$src1 = $dst" in {
2375 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2376 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2377 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2379 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2380 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2381 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2385 let Predicates = [HasAVX] in {
2386 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2387 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2388 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2389 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2390 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2392 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2393 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2394 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2395 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2396 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2399 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2400 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2401 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2402 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2403 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2405 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2406 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2407 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2408 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2409 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2412 let Predicates = [HasSSE1] in {
2413 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2414 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2415 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2416 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2417 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2420 let Predicates = [HasSSE2] in {
2421 // Generic SHUFPD patterns
2422 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2423 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2424 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2425 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2426 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2429 //===----------------------------------------------------------------------===//
2430 // SSE 1 & 2 - Unpack Instructions
2431 //===----------------------------------------------------------------------===//
2433 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2434 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2435 PatFrag mem_frag, RegisterClass RC,
2436 X86MemOperand x86memop, string asm,
2438 def rr : PI<opc, MRMSrcReg,
2439 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2441 (vt (OpNode RC:$src1, RC:$src2)))],
2443 def rm : PI<opc, MRMSrcMem,
2444 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2446 (vt (OpNode RC:$src1,
2447 (mem_frag addr:$src2))))],
2451 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2452 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2453 SSEPackedSingle>, TB, VEX_4V;
2454 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2455 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2456 SSEPackedDouble>, TB, OpSize, VEX_4V;
2457 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2458 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2459 SSEPackedSingle>, TB, VEX_4V;
2460 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2461 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2462 SSEPackedDouble>, TB, OpSize, VEX_4V;
2464 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2465 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2466 SSEPackedSingle>, TB, VEX_4V;
2467 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2468 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 SSEPackedDouble>, TB, OpSize, VEX_4V;
2470 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2471 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 SSEPackedSingle>, TB, VEX_4V;
2473 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2474 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2475 SSEPackedDouble>, TB, OpSize, VEX_4V;
2477 let Constraints = "$src1 = $dst" in {
2478 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2479 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2480 SSEPackedSingle>, TB;
2481 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2482 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2483 SSEPackedDouble>, TB, OpSize;
2484 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2485 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2486 SSEPackedSingle>, TB;
2487 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2488 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2489 SSEPackedDouble>, TB, OpSize;
2490 } // Constraints = "$src1 = $dst"
2492 let Predicates = [HasAVX1Only] in {
2493 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2494 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2495 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2496 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2497 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2498 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2499 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2500 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2502 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2503 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2504 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2505 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2506 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2507 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2508 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2509 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2512 let Predicates = [HasAVX], AddedComplexity = 1 in {
2513 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2514 // problem is during lowering, where it's not possible to recognize the load
2515 // fold cause it has two uses through a bitcast. One use disappears at isel
2516 // time and the fold opportunity reappears.
2517 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2518 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2521 let Predicates = [HasSSE2] in {
2522 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2523 // problem is during lowering, where it's not possible to recognize the load
2524 // fold cause it has two uses through a bitcast. One use disappears at isel
2525 // time and the fold opportunity reappears.
2526 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2527 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2530 //===----------------------------------------------------------------------===//
2531 // SSE 1 & 2 - Extract Floating-Point Sign mask
2532 //===----------------------------------------------------------------------===//
2534 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2535 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2537 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2538 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2539 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2540 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2541 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2542 IIC_SSE_MOVMSK, d>, REX_W;
2545 let Predicates = [HasAVX] in {
2546 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2547 "movmskps", SSEPackedSingle>, TB, VEX;
2548 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2549 "movmskpd", SSEPackedDouble>, TB,
2551 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2552 "movmskps", SSEPackedSingle>, TB, VEX;
2553 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2554 "movmskpd", SSEPackedDouble>, TB,
2557 def : Pat<(i32 (X86fgetsign FR32:$src)),
2558 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2559 def : Pat<(i64 (X86fgetsign FR32:$src)),
2560 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2561 def : Pat<(i32 (X86fgetsign FR64:$src)),
2562 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2563 def : Pat<(i64 (X86fgetsign FR64:$src)),
2564 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2567 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2568 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2569 SSEPackedSingle>, TB, VEX;
2570 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2571 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2572 SSEPackedDouble>, TB,
2574 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2575 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2576 SSEPackedSingle>, TB, VEX;
2577 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2578 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2579 SSEPackedDouble>, TB,
2583 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2584 SSEPackedSingle>, TB;
2585 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2586 SSEPackedDouble>, TB, OpSize;
2588 def : Pat<(i32 (X86fgetsign FR32:$src)),
2589 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2590 Requires<[HasSSE1]>;
2591 def : Pat<(i64 (X86fgetsign FR32:$src)),
2592 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2593 Requires<[HasSSE1]>;
2594 def : Pat<(i32 (X86fgetsign FR64:$src)),
2595 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2596 Requires<[HasSSE2]>;
2597 def : Pat<(i64 (X86fgetsign FR64:$src)),
2598 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2599 Requires<[HasSSE2]>;
2601 //===---------------------------------------------------------------------===//
2602 // SSE2 - Packed Integer Logical Instructions
2603 //===---------------------------------------------------------------------===//
2605 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2607 /// PDI_binop_rm - Simple SSE2 binary operator.
2608 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2609 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2610 X86MemOperand x86memop,
2612 bit IsCommutable = 0,
2614 let isCommutable = IsCommutable in
2615 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2616 (ins RC:$src1, RC:$src2),
2618 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2619 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2620 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2621 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2622 (ins RC:$src1, x86memop:$src2),
2624 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2625 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2626 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2627 (bitconvert (memop_frag addr:$src2)))))],
2630 } // ExeDomain = SSEPackedInt
2632 // These are ordered here for pattern ordering requirements with the fp versions
2634 let Predicates = [HasAVX] in {
2635 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2636 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2637 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2638 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2639 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2640 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2641 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2642 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2645 let Constraints = "$src1 = $dst" in {
2646 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2647 i128mem, SSE_BIT_ITINS_P, 1>;
2648 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2649 i128mem, SSE_BIT_ITINS_P, 1>;
2650 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2651 i128mem, SSE_BIT_ITINS_P, 1>;
2652 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2653 i128mem, SSE_BIT_ITINS_P, 0>;
2654 } // Constraints = "$src1 = $dst"
2656 let Predicates = [HasAVX2] in {
2657 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2658 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2659 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2660 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2661 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2662 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2663 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2664 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2667 //===----------------------------------------------------------------------===//
2668 // SSE 1 & 2 - Logical Instructions
2669 //===----------------------------------------------------------------------===//
2671 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2673 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2674 SDNode OpNode, OpndItins itins> {
2675 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2676 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2679 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2680 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2683 let Constraints = "$src1 = $dst" in {
2684 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2685 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2688 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2689 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2694 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2695 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2697 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2699 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2702 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2703 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2706 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2708 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2710 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2711 // are all promoted to v2i64, and the patterns are covered by the int
2712 // version. This is needed in SSE only, because v2i64 isn't supported on
2713 // SSE1, but only on SSE2.
2714 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2715 !strconcat(OpcodeStr, "ps"), f128mem, [],
2716 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2717 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2719 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2720 !strconcat(OpcodeStr, "pd"), f128mem,
2721 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2722 (bc_v2i64 (v2f64 VR128:$src2))))],
2723 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2724 (memopv2i64 addr:$src2)))], 0>,
2726 let Constraints = "$src1 = $dst" in {
2727 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2728 !strconcat(OpcodeStr, "ps"), f128mem,
2729 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2730 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2731 (memopv2i64 addr:$src2)))]>, TB;
2733 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2734 !strconcat(OpcodeStr, "pd"), f128mem,
2735 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2736 (bc_v2i64 (v2f64 VR128:$src2))))],
2737 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2738 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2742 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2744 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2746 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2747 !strconcat(OpcodeStr, "ps"), f256mem,
2748 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2749 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2750 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2752 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2753 !strconcat(OpcodeStr, "pd"), f256mem,
2754 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2755 (bc_v4i64 (v4f64 VR256:$src2))))],
2756 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2757 (memopv4i64 addr:$src2)))], 0>,
2761 // AVX 256-bit packed logical ops forms
2762 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2763 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2764 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2765 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2767 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2768 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2769 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2770 let isCommutable = 0 in
2771 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2773 //===----------------------------------------------------------------------===//
2774 // SSE 1 & 2 - Arithmetic Instructions
2775 //===----------------------------------------------------------------------===//
2777 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2780 /// In addition, we also have a special variant of the scalar form here to
2781 /// represent the associated intrinsic operation. This form is unlike the
2782 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2783 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2785 /// These three forms can each be reg+reg or reg+mem.
2788 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2790 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2793 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2794 OpNode, FR32, f32mem,
2795 itins.s, Is2Addr>, XS;
2796 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2797 OpNode, FR64, f64mem,
2798 itins.d, Is2Addr>, XD;
2801 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2804 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2805 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2807 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2808 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2812 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2815 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2816 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2818 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2819 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2823 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2826 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2827 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2828 itins.s, Is2Addr>, XS;
2829 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2830 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2831 itins.d, Is2Addr>, XD;
2834 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2837 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2838 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2839 SSEPackedSingle, itins.s, Is2Addr>,
2842 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2843 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2844 SSEPackedDouble, itins.d, Is2Addr>,
2848 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2850 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2851 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2852 SSEPackedSingle, itins.s, 0>, TB;
2854 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2855 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2856 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2859 // Binary Arithmetic instructions
2860 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2861 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2863 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2864 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2866 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2867 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2869 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2870 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2873 let isCommutable = 0 in {
2874 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2875 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2877 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2878 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2880 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2881 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2883 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2884 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2886 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2887 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2889 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2890 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2891 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2892 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2894 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2895 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2897 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2898 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2899 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2900 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2904 let Constraints = "$src1 = $dst" in {
2905 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2906 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2907 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2908 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2909 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2910 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2912 let isCommutable = 0 in {
2913 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2914 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2915 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2916 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2917 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2918 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2919 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2920 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2921 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2922 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2923 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2924 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2925 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2926 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2930 let isCommutable = 1, isCodeGenOnly = 1 in {
2931 defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
2933 defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>,
2934 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>, VEX_4V;
2935 defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
2937 defm VMINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P, 0>,
2938 basic_sse12_fp_binop_p_y<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>, VEX_4V;
2939 let Constraints = "$src1 = $dst" in {
2940 defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>,
2941 basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
2942 defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>,
2943 basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
2948 /// In addition, we also have a special variant of the scalar form here to
2949 /// represent the associated intrinsic operation. This form is unlike the
2950 /// plain scalar form, in that it takes an entire vector (instead of a
2951 /// scalar) and leaves the top elements undefined.
2953 /// And, we have a special variant form for a full-vector intrinsic form.
2955 def SSE_SQRTP : OpndItins<
2956 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2959 def SSE_SQRTS : OpndItins<
2960 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2963 def SSE_RCPP : OpndItins<
2964 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2967 def SSE_RCPS : OpndItins<
2968 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2971 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2972 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2973 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2974 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2975 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2976 [(set FR32:$dst, (OpNode FR32:$src))]>;
2977 // For scalar unary operations, fold a load into the operation
2978 // only in OptForSize mode. It eliminates an instruction, but it also
2979 // eliminates a whole-register clobber (the load), so it introduces a
2980 // partial register update condition.
2981 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2982 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2983 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2984 Requires<[HasSSE1, OptForSize]>;
2985 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2986 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2987 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2988 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2989 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2990 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2993 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2994 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2995 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2996 !strconcat(OpcodeStr,
2997 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2998 let mayLoad = 1 in {
2999 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3000 !strconcat(OpcodeStr,
3001 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3002 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3003 (ins VR128:$src1, ssmem:$src2),
3004 !strconcat(OpcodeStr,
3005 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3009 /// sse1_fp_unop_p - SSE1 unops in packed form.
3010 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3012 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3013 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3014 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3015 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3016 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3017 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3020 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3021 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3023 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3024 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3025 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3027 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3028 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3029 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3033 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3034 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3035 Intrinsic V4F32Int, OpndItins itins> {
3036 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3037 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3038 [(set VR128:$dst, (V4F32Int VR128:$src))],
3040 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3041 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3042 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3046 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3047 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3048 Intrinsic V4F32Int, OpndItins itins> {
3049 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3050 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3051 [(set VR256:$dst, (V4F32Int VR256:$src))],
3053 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3054 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3055 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3059 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3060 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3061 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3062 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3063 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3064 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3065 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3066 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3067 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3068 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3069 Requires<[HasSSE2, OptForSize]>;
3070 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3071 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3072 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3073 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3074 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3075 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3078 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3079 let hasSideEffects = 0 in
3080 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3081 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3082 !strconcat(OpcodeStr,
3083 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3084 let mayLoad = 1 in {
3085 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3086 !strconcat(OpcodeStr,
3087 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3088 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3089 (ins VR128:$src1, sdmem:$src2),
3090 !strconcat(OpcodeStr,
3091 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3095 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3096 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3097 SDNode OpNode, OpndItins itins> {
3098 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3099 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3100 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3101 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3102 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3103 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3106 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3107 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3109 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3110 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3111 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3113 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3114 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3115 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3119 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3120 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3121 Intrinsic V2F64Int, OpndItins itins> {
3122 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3123 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3124 [(set VR128:$dst, (V2F64Int VR128:$src))],
3126 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3127 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3128 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3132 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3133 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3134 Intrinsic V2F64Int, OpndItins itins> {
3135 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3136 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3137 [(set VR256:$dst, (V2F64Int VR256:$src))],
3139 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3140 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3141 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3145 let Predicates = [HasAVX] in {
3147 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3148 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3150 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3151 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3152 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3153 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3154 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3156 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3158 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3160 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3164 // Reciprocal approximations. Note that these typically require refinement
3165 // in order to obtain suitable precision.
3166 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3167 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3168 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3169 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3171 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3174 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3175 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3176 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3177 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3179 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3183 let AddedComplexity = 1 in {
3184 def : Pat<(f32 (fsqrt FR32:$src)),
3185 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3186 def : Pat<(f32 (fsqrt (load addr:$src))),
3187 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3188 Requires<[HasAVX, OptForSize]>;
3189 def : Pat<(f64 (fsqrt FR64:$src)),
3190 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3191 def : Pat<(f64 (fsqrt (load addr:$src))),
3192 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3193 Requires<[HasAVX, OptForSize]>;
3195 def : Pat<(f32 (X86frsqrt FR32:$src)),
3196 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3197 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3198 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3199 Requires<[HasAVX, OptForSize]>;
3201 def : Pat<(f32 (X86frcp FR32:$src)),
3202 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3203 def : Pat<(f32 (X86frcp (load addr:$src))),
3204 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3205 Requires<[HasAVX, OptForSize]>;
3208 let Predicates = [HasAVX], AddedComplexity = 1 in {
3209 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3210 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3211 (COPY_TO_REGCLASS VR128:$src, FR32)),
3213 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3214 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3216 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3217 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3218 (COPY_TO_REGCLASS VR128:$src, FR64)),
3220 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3221 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3223 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3224 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3225 (COPY_TO_REGCLASS VR128:$src, FR32)),
3227 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3228 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3230 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3231 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3232 (COPY_TO_REGCLASS VR128:$src, FR32)),
3234 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3235 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3239 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3241 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3242 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3243 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3245 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3246 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3248 // Reciprocal approximations. Note that these typically require refinement
3249 // in order to obtain suitable precision.
3250 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3252 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3253 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3255 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3257 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3258 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3260 // There is no f64 version of the reciprocal approximation instructions.
3262 //===----------------------------------------------------------------------===//
3263 // SSE 1 & 2 - Non-temporal stores
3264 //===----------------------------------------------------------------------===//
3266 let AddedComplexity = 400 in { // Prefer non-temporal versions
3267 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3268 (ins f128mem:$dst, VR128:$src),
3269 "movntps\t{$src, $dst|$dst, $src}",
3270 [(alignednontemporalstore (v4f32 VR128:$src),
3272 IIC_SSE_MOVNT>, VEX;
3273 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3274 (ins f128mem:$dst, VR128:$src),
3275 "movntpd\t{$src, $dst|$dst, $src}",
3276 [(alignednontemporalstore (v2f64 VR128:$src),
3278 IIC_SSE_MOVNT>, VEX;
3280 let ExeDomain = SSEPackedInt in
3281 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3282 (ins f128mem:$dst, VR128:$src),
3283 "movntdq\t{$src, $dst|$dst, $src}",
3284 [(alignednontemporalstore (v2i64 VR128:$src),
3286 IIC_SSE_MOVNT>, VEX;
3288 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3289 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3291 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3292 (ins f256mem:$dst, VR256:$src),
3293 "movntps\t{$src, $dst|$dst, $src}",
3294 [(alignednontemporalstore (v8f32 VR256:$src),
3296 IIC_SSE_MOVNT>, VEX;
3297 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3298 (ins f256mem:$dst, VR256:$src),
3299 "movntpd\t{$src, $dst|$dst, $src}",
3300 [(alignednontemporalstore (v4f64 VR256:$src),
3302 IIC_SSE_MOVNT>, VEX;
3303 let ExeDomain = SSEPackedInt in
3304 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3305 (ins f256mem:$dst, VR256:$src),
3306 "movntdq\t{$src, $dst|$dst, $src}",
3307 [(alignednontemporalstore (v4i64 VR256:$src),
3309 IIC_SSE_MOVNT>, VEX;
3312 let AddedComplexity = 400 in { // Prefer non-temporal versions
3313 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3314 "movntps\t{$src, $dst|$dst, $src}",
3315 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3317 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3318 "movntpd\t{$src, $dst|$dst, $src}",
3319 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3322 let ExeDomain = SSEPackedInt in
3323 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3324 "movntdq\t{$src, $dst|$dst, $src}",
3325 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3328 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3329 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3331 // There is no AVX form for instructions below this point
3332 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3333 "movnti{l}\t{$src, $dst|$dst, $src}",
3334 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3336 TB, Requires<[HasSSE2]>;
3337 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3338 "movnti{q}\t{$src, $dst|$dst, $src}",
3339 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3341 TB, Requires<[HasSSE2]>;
3344 //===----------------------------------------------------------------------===//
3345 // SSE 1 & 2 - Prefetch and memory fence
3346 //===----------------------------------------------------------------------===//
3348 // Prefetch intrinsic.
3349 let Predicates = [HasSSE1] in {
3350 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3351 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3352 IIC_SSE_PREFETCH>, TB;
3353 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3354 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3355 IIC_SSE_PREFETCH>, TB;
3356 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3357 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3358 IIC_SSE_PREFETCH>, TB;
3359 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3360 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3361 IIC_SSE_PREFETCH>, TB;
3365 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3366 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3367 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3369 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3370 // was introduced with SSE2, it's backward compatible.
3371 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3373 // Load, store, and memory fence
3374 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3375 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3376 TB, Requires<[HasSSE1]>;
3377 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3378 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3379 TB, Requires<[HasSSE2]>;
3380 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3381 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3382 TB, Requires<[HasSSE2]>;
3384 def : Pat<(X86SFence), (SFENCE)>;
3385 def : Pat<(X86LFence), (LFENCE)>;
3386 def : Pat<(X86MFence), (MFENCE)>;
3388 //===----------------------------------------------------------------------===//
3389 // SSE 1 & 2 - Load/Store XCSR register
3390 //===----------------------------------------------------------------------===//
3392 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3393 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3394 IIC_SSE_LDMXCSR>, VEX;
3395 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3396 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3397 IIC_SSE_STMXCSR>, VEX;
3399 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3400 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3402 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3403 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3406 //===---------------------------------------------------------------------===//
3407 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3408 //===---------------------------------------------------------------------===//
3410 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3412 let neverHasSideEffects = 1 in {
3413 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3414 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3416 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3417 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3420 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3421 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3423 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3424 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3428 let isCodeGenOnly = 1 in {
3429 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3430 "movdqa\t{$src, $dst|$dst, $src}", [],
3433 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3434 "movdqa\t{$src, $dst|$dst, $src}", [],
3437 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3438 "movdqu\t{$src, $dst|$dst, $src}", [],
3441 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3442 "movdqu\t{$src, $dst|$dst, $src}", [],
3447 let canFoldAsLoad = 1, mayLoad = 1 in {
3448 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3449 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3451 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3452 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3454 let Predicates = [HasAVX] in {
3455 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3456 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3458 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3459 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3464 let mayStore = 1 in {
3465 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3466 (ins i128mem:$dst, VR128:$src),
3467 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3469 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3470 (ins i256mem:$dst, VR256:$src),
3471 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3473 let Predicates = [HasAVX] in {
3474 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3475 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3477 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3478 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3483 let neverHasSideEffects = 1 in
3484 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3485 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3487 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3488 "movdqu\t{$src, $dst|$dst, $src}",
3489 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3492 let isCodeGenOnly = 1 in {
3493 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3494 "movdqa\t{$src, $dst|$dst, $src}", [],
3497 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3498 "movdqu\t{$src, $dst|$dst, $src}",
3499 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3502 let canFoldAsLoad = 1, mayLoad = 1 in {
3503 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3504 "movdqa\t{$src, $dst|$dst, $src}",
3505 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3507 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3508 "movdqu\t{$src, $dst|$dst, $src}",
3509 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3511 XS, Requires<[HasSSE2]>;
3514 let mayStore = 1 in {
3515 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3516 "movdqa\t{$src, $dst|$dst, $src}",
3517 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3519 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3520 "movdqu\t{$src, $dst|$dst, $src}",
3521 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3523 XS, Requires<[HasSSE2]>;
3526 // Intrinsic forms of MOVDQU load and store
3527 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3528 "vmovdqu\t{$src, $dst|$dst, $src}",
3529 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3531 XS, VEX, Requires<[HasAVX]>;
3533 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3534 "movdqu\t{$src, $dst|$dst, $src}",
3535 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3537 XS, Requires<[HasSSE2]>;
3539 } // ExeDomain = SSEPackedInt
3541 let Predicates = [HasAVX] in {
3542 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3543 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3546 //===---------------------------------------------------------------------===//
3547 // SSE2 - Packed Integer Arithmetic Instructions
3548 //===---------------------------------------------------------------------===//
3550 def SSE_PMADD : OpndItins<
3551 IIC_SSE_PMADD, IIC_SSE_PMADD
3554 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3556 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3557 RegisterClass RC, PatFrag memop_frag,
3558 X86MemOperand x86memop,
3560 bit IsCommutable = 0,
3562 let isCommutable = IsCommutable in
3563 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3564 (ins RC:$src1, RC:$src2),
3566 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3567 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3568 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3569 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3570 (ins RC:$src1, x86memop:$src2),
3572 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3573 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3574 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3578 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3579 string OpcodeStr, SDNode OpNode,
3580 SDNode OpNode2, RegisterClass RC,
3581 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3582 ShiftOpndItins itins,
3584 // src2 is always 128-bit
3585 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3586 (ins RC:$src1, VR128:$src2),
3588 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3589 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3590 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3592 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3593 (ins RC:$src1, i128mem:$src2),
3595 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3596 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3597 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3598 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3599 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3600 (ins RC:$src1, i32i8imm:$src2),
3602 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3603 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3604 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3607 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3608 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3609 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3610 PatFrag memop_frag, X86MemOperand x86memop,
3612 bit IsCommutable = 0, bit Is2Addr = 1> {
3613 let isCommutable = IsCommutable in
3614 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3615 (ins RC:$src1, RC:$src2),
3617 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3618 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3619 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3620 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3621 (ins RC:$src1, x86memop:$src2),
3623 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3624 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3625 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3626 (bitconvert (memop_frag addr:$src2)))))]>;
3628 } // ExeDomain = SSEPackedInt
3630 // 128-bit Integer Arithmetic
3632 let Predicates = [HasAVX] in {
3633 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3634 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3636 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3637 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3638 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3639 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3640 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3641 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3642 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3643 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3644 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3645 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3646 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3647 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3648 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3649 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3650 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3651 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3652 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3653 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3657 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3658 VR128, memopv2i64, i128mem,
3659 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3660 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3661 VR128, memopv2i64, i128mem,
3662 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3663 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3664 VR128, memopv2i64, i128mem,
3665 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3666 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3667 VR128, memopv2i64, i128mem,
3668 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3669 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3670 VR128, memopv2i64, i128mem,
3671 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3672 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3673 VR128, memopv2i64, i128mem,
3674 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3675 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3676 VR128, memopv2i64, i128mem,
3677 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3678 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3679 VR128, memopv2i64, i128mem,
3680 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3681 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3682 VR128, memopv2i64, i128mem,
3683 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3684 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3685 VR128, memopv2i64, i128mem,
3686 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3687 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3688 VR128, memopv2i64, i128mem,
3689 SSE_PMADD, 1, 0>, VEX_4V;
3690 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3691 VR128, memopv2i64, i128mem,
3692 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3693 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3694 VR128, memopv2i64, i128mem,
3695 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3696 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3697 VR128, memopv2i64, i128mem,
3698 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3699 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3700 VR128, memopv2i64, i128mem,
3701 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3702 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3703 VR128, memopv2i64, i128mem,
3704 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3705 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3706 VR128, memopv2i64, i128mem,
3707 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3708 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3709 VR128, memopv2i64, i128mem,
3710 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3713 let Predicates = [HasAVX2] in {
3714 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3715 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3716 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3717 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3718 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3719 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3720 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3721 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3722 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3723 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3724 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3725 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3726 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3727 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3728 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3729 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3730 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3731 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3732 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3733 VR256, memopv4i64, i256mem,
3734 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3737 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3738 VR256, memopv4i64, i256mem,
3739 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3740 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3741 VR256, memopv4i64, i256mem,
3742 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3743 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3744 VR256, memopv4i64, i256mem,
3745 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3746 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3747 VR256, memopv4i64, i256mem,
3748 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3749 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3750 VR256, memopv4i64, i256mem,
3751 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3752 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3753 VR256, memopv4i64, i256mem,
3754 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3755 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3756 VR256, memopv4i64, i256mem,
3757 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3758 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3759 VR256, memopv4i64, i256mem,
3760 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3761 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3762 VR256, memopv4i64, i256mem,
3763 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3764 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3765 VR256, memopv4i64, i256mem,
3766 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3767 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3768 VR256, memopv4i64, i256mem,
3769 SSE_PMADD, 1, 0>, VEX_4V;
3770 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3771 VR256, memopv4i64, i256mem,
3772 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3773 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3774 VR256, memopv4i64, i256mem,
3775 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3776 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3777 VR256, memopv4i64, i256mem,
3778 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3779 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3780 VR256, memopv4i64, i256mem,
3781 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3782 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3783 VR256, memopv4i64, i256mem,
3784 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3785 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3786 VR256, memopv4i64, i256mem,
3787 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3788 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3789 VR256, memopv4i64, i256mem,
3790 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3793 let Constraints = "$src1 = $dst" in {
3794 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3795 i128mem, SSE_INTALU_ITINS_P, 1>;
3796 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3797 i128mem, SSE_INTALU_ITINS_P, 1>;
3798 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3799 i128mem, SSE_INTALU_ITINS_P, 1>;
3800 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3801 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3802 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3803 i128mem, SSE_INTMUL_ITINS_P, 1>;
3804 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3805 i128mem, SSE_INTALU_ITINS_P>;
3806 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3807 i128mem, SSE_INTALU_ITINS_P>;
3808 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3809 i128mem, SSE_INTALU_ITINS_P>;
3810 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3811 i128mem, SSE_INTALUQ_ITINS_P>;
3812 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3813 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3816 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3817 VR128, memopv2i64, i128mem,
3818 SSE_INTALU_ITINS_P>;
3819 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3820 VR128, memopv2i64, i128mem,
3821 SSE_INTALU_ITINS_P>;
3822 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3823 VR128, memopv2i64, i128mem,
3824 SSE_INTALU_ITINS_P>;
3825 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3826 VR128, memopv2i64, i128mem,
3827 SSE_INTALU_ITINS_P>;
3828 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3829 VR128, memopv2i64, i128mem,
3830 SSE_INTALU_ITINS_P, 1>;
3831 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3832 VR128, memopv2i64, i128mem,
3833 SSE_INTALU_ITINS_P, 1>;
3834 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3835 VR128, memopv2i64, i128mem,
3836 SSE_INTALU_ITINS_P, 1>;
3837 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3838 VR128, memopv2i64, i128mem,
3839 SSE_INTALU_ITINS_P, 1>;
3840 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3841 VR128, memopv2i64, i128mem,
3842 SSE_INTMUL_ITINS_P, 1>;
3843 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3844 VR128, memopv2i64, i128mem,
3845 SSE_INTMUL_ITINS_P, 1>;
3846 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3847 VR128, memopv2i64, i128mem,
3849 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3850 VR128, memopv2i64, i128mem,
3851 SSE_INTALU_ITINS_P, 1>;
3852 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3853 VR128, memopv2i64, i128mem,
3854 SSE_INTALU_ITINS_P, 1>;
3855 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3856 VR128, memopv2i64, i128mem,
3857 SSE_INTALU_ITINS_P, 1>;
3858 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3859 VR128, memopv2i64, i128mem,
3860 SSE_INTALU_ITINS_P, 1>;
3861 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3862 VR128, memopv2i64, i128mem,
3863 SSE_INTALU_ITINS_P, 1>;
3864 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3865 VR128, memopv2i64, i128mem,
3866 SSE_INTALU_ITINS_P, 1>;
3867 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3868 VR128, memopv2i64, i128mem,
3869 SSE_INTALU_ITINS_P, 1>;
3871 } // Constraints = "$src1 = $dst"
3873 //===---------------------------------------------------------------------===//
3874 // SSE2 - Packed Integer Logical Instructions
3875 //===---------------------------------------------------------------------===//
3877 let Predicates = [HasAVX] in {
3878 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3879 VR128, v8i16, v8i16, bc_v8i16,
3880 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3881 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3882 VR128, v4i32, v4i32, bc_v4i32,
3883 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3884 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3885 VR128, v2i64, v2i64, bc_v2i64,
3886 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3888 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3889 VR128, v8i16, v8i16, bc_v8i16,
3890 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3891 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3892 VR128, v4i32, v4i32, bc_v4i32,
3893 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3894 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3895 VR128, v2i64, v2i64, bc_v2i64,
3896 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3898 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3899 VR128, v8i16, v8i16, bc_v8i16,
3900 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3901 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3902 VR128, v4i32, v4i32, bc_v4i32,
3903 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3905 let ExeDomain = SSEPackedInt in {
3906 // 128-bit logical shifts.
3907 def VPSLLDQri : PDIi8<0x73, MRM7r,
3908 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3909 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3911 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3913 def VPSRLDQri : PDIi8<0x73, MRM3r,
3914 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3915 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3917 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3919 // PSRADQri doesn't exist in SSE[1-3].
3921 } // Predicates = [HasAVX]
3923 let Predicates = [HasAVX2] in {
3924 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3925 VR256, v16i16, v8i16, bc_v8i16,
3926 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3927 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3928 VR256, v8i32, v4i32, bc_v4i32,
3929 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3930 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3931 VR256, v4i64, v2i64, bc_v2i64,
3932 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3934 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3935 VR256, v16i16, v8i16, bc_v8i16,
3936 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3937 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3938 VR256, v8i32, v4i32, bc_v4i32,
3939 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3940 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3941 VR256, v4i64, v2i64, bc_v2i64,
3942 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3944 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3945 VR256, v16i16, v8i16, bc_v8i16,
3946 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3947 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3948 VR256, v8i32, v4i32, bc_v4i32,
3949 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3951 let ExeDomain = SSEPackedInt in {
3952 // 256-bit logical shifts.
3953 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3954 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3955 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3957 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3959 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3960 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3961 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3963 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3965 // PSRADQYri doesn't exist in SSE[1-3].
3967 } // Predicates = [HasAVX2]
3969 let Constraints = "$src1 = $dst" in {
3970 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3971 VR128, v8i16, v8i16, bc_v8i16,
3972 SSE_INTSHIFT_ITINS_P>;
3973 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3974 VR128, v4i32, v4i32, bc_v4i32,
3975 SSE_INTSHIFT_ITINS_P>;
3976 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3977 VR128, v2i64, v2i64, bc_v2i64,
3978 SSE_INTSHIFT_ITINS_P>;
3980 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3981 VR128, v8i16, v8i16, bc_v8i16,
3982 SSE_INTSHIFT_ITINS_P>;
3983 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3984 VR128, v4i32, v4i32, bc_v4i32,
3985 SSE_INTSHIFT_ITINS_P>;
3986 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3987 VR128, v2i64, v2i64, bc_v2i64,
3988 SSE_INTSHIFT_ITINS_P>;
3990 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3991 VR128, v8i16, v8i16, bc_v8i16,
3992 SSE_INTSHIFT_ITINS_P>;
3993 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3994 VR128, v4i32, v4i32, bc_v4i32,
3995 SSE_INTSHIFT_ITINS_P>;
3997 let ExeDomain = SSEPackedInt in {
3998 // 128-bit logical shifts.
3999 def PSLLDQri : PDIi8<0x73, MRM7r,
4000 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4001 "pslldq\t{$src2, $dst|$dst, $src2}",
4003 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
4004 def PSRLDQri : PDIi8<0x73, MRM3r,
4005 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4006 "psrldq\t{$src2, $dst|$dst, $src2}",
4008 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
4009 // PSRADQri doesn't exist in SSE[1-3].
4011 } // Constraints = "$src1 = $dst"
4013 let Predicates = [HasAVX] in {
4014 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4015 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4016 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4017 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4018 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4019 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4021 // Shift up / down and insert zero's.
4022 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4023 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4024 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4025 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4028 let Predicates = [HasAVX2] in {
4029 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4030 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4031 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4032 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4035 let Predicates = [HasSSE2] in {
4036 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4037 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4038 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4039 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4040 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4041 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4043 // Shift up / down and insert zero's.
4044 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4045 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4046 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4047 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4050 //===---------------------------------------------------------------------===//
4051 // SSE2 - Packed Integer Comparison Instructions
4052 //===---------------------------------------------------------------------===//
4054 let Predicates = [HasAVX] in {
4055 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4056 VR128, memopv2i64, i128mem,
4057 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4058 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4059 VR128, memopv2i64, i128mem,
4060 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4061 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4062 VR128, memopv2i64, i128mem,
4063 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4064 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4065 VR128, memopv2i64, i128mem,
4066 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4067 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4068 VR128, memopv2i64, i128mem,
4069 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4070 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4071 VR128, memopv2i64, i128mem,
4072 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4075 let Predicates = [HasAVX2] in {
4076 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4077 VR256, memopv4i64, i256mem,
4078 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4079 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4080 VR256, memopv4i64, i256mem,
4081 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4082 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4083 VR256, memopv4i64, i256mem,
4084 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4085 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4086 VR256, memopv4i64, i256mem,
4087 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4088 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4089 VR256, memopv4i64, i256mem,
4090 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4091 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4092 VR256, memopv4i64, i256mem,
4093 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4096 let Constraints = "$src1 = $dst" in {
4097 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4098 VR128, memopv2i64, i128mem,
4099 SSE_INTALU_ITINS_P, 1>;
4100 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4101 VR128, memopv2i64, i128mem,
4102 SSE_INTALU_ITINS_P, 1>;
4103 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4104 VR128, memopv2i64, i128mem,
4105 SSE_INTALU_ITINS_P, 1>;
4106 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4107 VR128, memopv2i64, i128mem,
4108 SSE_INTALU_ITINS_P>;
4109 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4110 VR128, memopv2i64, i128mem,
4111 SSE_INTALU_ITINS_P>;
4112 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4113 VR128, memopv2i64, i128mem,
4114 SSE_INTALU_ITINS_P>;
4115 } // Constraints = "$src1 = $dst"
4117 //===---------------------------------------------------------------------===//
4118 // SSE2 - Packed Integer Pack Instructions
4119 //===---------------------------------------------------------------------===//
4121 let Predicates = [HasAVX] in {
4122 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4123 VR128, memopv2i64, i128mem,
4124 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4125 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4126 VR128, memopv2i64, i128mem,
4127 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4128 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4129 VR128, memopv2i64, i128mem,
4130 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4133 let Predicates = [HasAVX2] in {
4134 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4135 VR256, memopv4i64, i256mem,
4136 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4137 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4138 VR256, memopv4i64, i256mem,
4139 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4140 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4141 VR256, memopv4i64, i256mem,
4142 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4145 let Constraints = "$src1 = $dst" in {
4146 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4147 VR128, memopv2i64, i128mem,
4148 SSE_INTALU_ITINS_P>;
4149 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4150 VR128, memopv2i64, i128mem,
4151 SSE_INTALU_ITINS_P>;
4152 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4153 VR128, memopv2i64, i128mem,
4154 SSE_INTALU_ITINS_P>;
4155 } // Constraints = "$src1 = $dst"
4157 //===---------------------------------------------------------------------===//
4158 // SSE2 - Packed Integer Shuffle Instructions
4159 //===---------------------------------------------------------------------===//
4161 let ExeDomain = SSEPackedInt in {
4162 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4163 def ri : Ii8<0x70, MRMSrcReg,
4164 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4165 !strconcat(OpcodeStr,
4166 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4167 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4169 def mi : Ii8<0x70, MRMSrcMem,
4170 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4171 !strconcat(OpcodeStr,
4172 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4174 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4179 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4180 def Yri : Ii8<0x70, MRMSrcReg,
4181 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4182 !strconcat(OpcodeStr,
4183 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4184 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4185 def Ymi : Ii8<0x70, MRMSrcMem,
4186 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4187 !strconcat(OpcodeStr,
4188 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4190 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4191 (i8 imm:$src2))))]>;
4193 } // ExeDomain = SSEPackedInt
4195 let Predicates = [HasAVX] in {
4196 let AddedComplexity = 5 in
4197 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4199 // SSE2 with ImmT == Imm8 and XS prefix.
4200 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4202 // SSE2 with ImmT == Imm8 and XD prefix.
4203 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4205 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4206 (VPSHUFDmi addr:$src1, imm:$imm)>;
4207 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4208 (VPSHUFDri VR128:$src1, imm:$imm)>;
4211 let Predicates = [HasAVX2] in {
4212 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4213 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4214 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4217 let Predicates = [HasSSE2] in {
4218 let AddedComplexity = 5 in
4219 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4221 // SSE2 with ImmT == Imm8 and XS prefix.
4222 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4224 // SSE2 with ImmT == Imm8 and XD prefix.
4225 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4227 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4228 (PSHUFDmi addr:$src1, imm:$imm)>;
4229 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4230 (PSHUFDri VR128:$src1, imm:$imm)>;
4233 //===---------------------------------------------------------------------===//
4234 // SSE2 - Packed Integer Unpack Instructions
4235 //===---------------------------------------------------------------------===//
4237 let ExeDomain = SSEPackedInt in {
4238 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4239 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4240 def rr : PDI<opc, MRMSrcReg,
4241 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4243 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4244 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4245 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4247 def rm : PDI<opc, MRMSrcMem,
4248 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4250 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4251 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4252 [(set VR128:$dst, (OpNode VR128:$src1,
4253 (bc_frag (memopv2i64
4258 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4259 SDNode OpNode, PatFrag bc_frag> {
4260 def Yrr : PDI<opc, MRMSrcReg,
4261 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4262 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4263 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4264 def Yrm : PDI<opc, MRMSrcMem,
4265 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4266 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4267 [(set VR256:$dst, (OpNode VR256:$src1,
4268 (bc_frag (memopv4i64 addr:$src2))))]>;
4271 let Predicates = [HasAVX] in {
4272 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4273 bc_v16i8, 0>, VEX_4V;
4274 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4275 bc_v8i16, 0>, VEX_4V;
4276 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4277 bc_v4i32, 0>, VEX_4V;
4278 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4279 bc_v2i64, 0>, VEX_4V;
4281 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4282 bc_v16i8, 0>, VEX_4V;
4283 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4284 bc_v8i16, 0>, VEX_4V;
4285 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4286 bc_v4i32, 0>, VEX_4V;
4287 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4288 bc_v2i64, 0>, VEX_4V;
4291 let Predicates = [HasAVX2] in {
4292 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4294 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4296 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4298 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4301 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4303 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4305 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4307 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4311 let Constraints = "$src1 = $dst" in {
4312 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4314 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4316 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4318 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4321 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4323 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4325 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4327 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4330 } // ExeDomain = SSEPackedInt
4332 //===---------------------------------------------------------------------===//
4333 // SSE2 - Packed Integer Extract and Insert
4334 //===---------------------------------------------------------------------===//
4336 let ExeDomain = SSEPackedInt in {
4337 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4338 def rri : Ii8<0xC4, MRMSrcReg,
4339 (outs VR128:$dst), (ins VR128:$src1,
4340 GR32:$src2, i32i8imm:$src3),
4342 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4343 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4345 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4346 def rmi : Ii8<0xC4, MRMSrcMem,
4347 (outs VR128:$dst), (ins VR128:$src1,
4348 i16mem:$src2, i32i8imm:$src3),
4350 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4351 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4353 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4354 imm:$src3))], IIC_SSE_PINSRW>;
4358 let Predicates = [HasAVX] in
4359 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4360 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4361 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4362 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4363 imm:$src2))]>, TB, OpSize, VEX;
4364 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4365 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4366 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4367 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4368 imm:$src2))], IIC_SSE_PEXTRW>;
4371 let Predicates = [HasAVX] in {
4372 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4373 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4374 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4375 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4376 []>, TB, OpSize, VEX_4V;
4379 let Constraints = "$src1 = $dst" in
4380 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4382 } // ExeDomain = SSEPackedInt
4384 //===---------------------------------------------------------------------===//
4385 // SSE2 - Packed Mask Creation
4386 //===---------------------------------------------------------------------===//
4388 let ExeDomain = SSEPackedInt in {
4390 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4391 "pmovmskb\t{$src, $dst|$dst, $src}",
4392 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4393 IIC_SSE_MOVMSK>, VEX;
4394 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4395 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4397 let Predicates = [HasAVX2] in {
4398 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4399 "pmovmskb\t{$src, $dst|$dst, $src}",
4400 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4401 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4402 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4405 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4406 "pmovmskb\t{$src, $dst|$dst, $src}",
4407 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4410 } // ExeDomain = SSEPackedInt
4412 //===---------------------------------------------------------------------===//
4413 // SSE2 - Conditional Store
4414 //===---------------------------------------------------------------------===//
4416 let ExeDomain = SSEPackedInt in {
4419 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4420 (ins VR128:$src, VR128:$mask),
4421 "maskmovdqu\t{$mask, $src|$src, $mask}",
4422 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4423 IIC_SSE_MASKMOV>, VEX;
4425 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4426 (ins VR128:$src, VR128:$mask),
4427 "maskmovdqu\t{$mask, $src|$src, $mask}",
4428 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4429 IIC_SSE_MASKMOV>, VEX;
4432 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4433 "maskmovdqu\t{$mask, $src|$src, $mask}",
4434 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4437 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4438 "maskmovdqu\t{$mask, $src|$src, $mask}",
4439 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4442 } // ExeDomain = SSEPackedInt
4444 //===---------------------------------------------------------------------===//
4445 // SSE2 - Move Doubleword
4446 //===---------------------------------------------------------------------===//
4448 //===---------------------------------------------------------------------===//
4449 // Move Int Doubleword to Packed Double Int
4451 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4452 "movd\t{$src, $dst|$dst, $src}",
4454 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4456 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4457 "movd\t{$src, $dst|$dst, $src}",
4459 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4462 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4463 "mov{d|q}\t{$src, $dst|$dst, $src}",
4465 (v2i64 (scalar_to_vector GR64:$src)))],
4466 IIC_SSE_MOVDQ>, VEX;
4467 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4468 "mov{d|q}\t{$src, $dst|$dst, $src}",
4469 [(set FR64:$dst, (bitconvert GR64:$src))],
4470 IIC_SSE_MOVDQ>, VEX;
4472 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4473 "movd\t{$src, $dst|$dst, $src}",
4475 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4476 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4477 "movd\t{$src, $dst|$dst, $src}",
4479 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4481 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4482 "mov{d|q}\t{$src, $dst|$dst, $src}",
4484 (v2i64 (scalar_to_vector GR64:$src)))],
4486 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4487 "mov{d|q}\t{$src, $dst|$dst, $src}",
4488 [(set FR64:$dst, (bitconvert GR64:$src))],
4491 //===---------------------------------------------------------------------===//
4492 // Move Int Doubleword to Single Scalar
4494 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4495 "movd\t{$src, $dst|$dst, $src}",
4496 [(set FR32:$dst, (bitconvert GR32:$src))],
4497 IIC_SSE_MOVDQ>, VEX;
4499 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4500 "movd\t{$src, $dst|$dst, $src}",
4501 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4504 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4505 "movd\t{$src, $dst|$dst, $src}",
4506 [(set FR32:$dst, (bitconvert GR32:$src))],
4509 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4510 "movd\t{$src, $dst|$dst, $src}",
4511 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4514 //===---------------------------------------------------------------------===//
4515 // Move Packed Doubleword Int to Packed Double Int
4517 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4518 "movd\t{$src, $dst|$dst, $src}",
4519 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4520 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4521 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4522 (ins i32mem:$dst, VR128:$src),
4523 "movd\t{$src, $dst|$dst, $src}",
4524 [(store (i32 (vector_extract (v4i32 VR128:$src),
4525 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4527 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4528 "movd\t{$src, $dst|$dst, $src}",
4529 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4530 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4531 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4532 "movd\t{$src, $dst|$dst, $src}",
4533 [(store (i32 (vector_extract (v4i32 VR128:$src),
4534 (iPTR 0))), addr:$dst)],
4537 //===---------------------------------------------------------------------===//
4538 // Move Packed Doubleword Int first element to Doubleword Int
4540 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4541 "mov{d|q}\t{$src, $dst|$dst, $src}",
4542 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4545 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4547 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4548 "mov{d|q}\t{$src, $dst|$dst, $src}",
4549 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4553 //===---------------------------------------------------------------------===//
4554 // Bitcast FR64 <-> GR64
4556 let Predicates = [HasAVX] in
4557 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4558 "vmovq\t{$src, $dst|$dst, $src}",
4559 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4561 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4562 "mov{d|q}\t{$src, $dst|$dst, $src}",
4563 [(set GR64:$dst, (bitconvert FR64:$src))],
4564 IIC_SSE_MOVDQ>, VEX;
4565 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4566 "movq\t{$src, $dst|$dst, $src}",
4567 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4568 IIC_SSE_MOVDQ>, VEX;
4570 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4571 "movq\t{$src, $dst|$dst, $src}",
4572 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4574 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4575 "mov{d|q}\t{$src, $dst|$dst, $src}",
4576 [(set GR64:$dst, (bitconvert FR64:$src))],
4578 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4579 "movq\t{$src, $dst|$dst, $src}",
4580 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4583 //===---------------------------------------------------------------------===//
4584 // Move Scalar Single to Double Int
4586 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4587 "movd\t{$src, $dst|$dst, $src}",
4588 [(set GR32:$dst, (bitconvert FR32:$src))],
4589 IIC_SSE_MOVD_ToGP>, VEX;
4590 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4592 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4593 IIC_SSE_MOVDQ>, VEX;
4594 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4595 "movd\t{$src, $dst|$dst, $src}",
4596 [(set GR32:$dst, (bitconvert FR32:$src))],
4598 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4599 "movd\t{$src, $dst|$dst, $src}",
4600 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4603 //===---------------------------------------------------------------------===//
4604 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4606 let AddedComplexity = 15 in {
4607 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4608 "movd\t{$src, $dst|$dst, $src}",
4609 [(set VR128:$dst, (v4i32 (X86vzmovl
4610 (v4i32 (scalar_to_vector GR32:$src)))))],
4611 IIC_SSE_MOVDQ>, VEX;
4612 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4613 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4614 [(set VR128:$dst, (v2i64 (X86vzmovl
4615 (v2i64 (scalar_to_vector GR64:$src)))))],
4619 let AddedComplexity = 15 in {
4620 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4621 "movd\t{$src, $dst|$dst, $src}",
4622 [(set VR128:$dst, (v4i32 (X86vzmovl
4623 (v4i32 (scalar_to_vector GR32:$src)))))],
4625 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4626 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4627 [(set VR128:$dst, (v2i64 (X86vzmovl
4628 (v2i64 (scalar_to_vector GR64:$src)))))],
4632 let AddedComplexity = 20 in {
4633 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4634 "movd\t{$src, $dst|$dst, $src}",
4636 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4637 (loadi32 addr:$src))))))],
4638 IIC_SSE_MOVDQ>, VEX;
4639 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4640 "movd\t{$src, $dst|$dst, $src}",
4642 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4643 (loadi32 addr:$src))))))],
4647 let Predicates = [HasAVX] in {
4648 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4649 let AddedComplexity = 20 in {
4650 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4651 (VMOVZDI2PDIrm addr:$src)>;
4652 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4653 (VMOVZDI2PDIrm addr:$src)>;
4655 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4656 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4657 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4658 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4659 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4660 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4661 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4664 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4665 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4666 (MOVZDI2PDIrm addr:$src)>;
4667 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4668 (MOVZDI2PDIrm addr:$src)>;
4671 // These are the correct encodings of the instructions so that we know how to
4672 // read correct assembly, even though we continue to emit the wrong ones for
4673 // compatibility with Darwin's buggy assembler.
4674 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4675 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4676 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4677 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4678 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4679 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4680 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4681 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4682 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4683 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4684 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4685 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4687 //===---------------------------------------------------------------------===//
4688 // SSE2 - Move Quadword
4689 //===---------------------------------------------------------------------===//
4691 //===---------------------------------------------------------------------===//
4692 // Move Quadword Int to Packed Quadword Int
4694 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4695 "vmovq\t{$src, $dst|$dst, $src}",
4697 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4698 VEX, Requires<[HasAVX]>;
4699 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4700 "movq\t{$src, $dst|$dst, $src}",
4702 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4704 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4706 //===---------------------------------------------------------------------===//
4707 // Move Packed Quadword Int to Quadword Int
4709 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4710 "movq\t{$src, $dst|$dst, $src}",
4711 [(store (i64 (vector_extract (v2i64 VR128:$src),
4712 (iPTR 0))), addr:$dst)],
4713 IIC_SSE_MOVDQ>, VEX;
4714 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4715 "movq\t{$src, $dst|$dst, $src}",
4716 [(store (i64 (vector_extract (v2i64 VR128:$src),
4717 (iPTR 0))), addr:$dst)],
4720 //===---------------------------------------------------------------------===//
4721 // Store / copy lower 64-bits of a XMM register.
4723 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4724 "movq\t{$src, $dst|$dst, $src}",
4725 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4726 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4727 "movq\t{$src, $dst|$dst, $src}",
4728 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4731 let AddedComplexity = 20 in
4732 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4733 "vmovq\t{$src, $dst|$dst, $src}",
4735 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4736 (loadi64 addr:$src))))))],
4738 XS, VEX, Requires<[HasAVX]>;
4740 let AddedComplexity = 20 in
4741 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4742 "movq\t{$src, $dst|$dst, $src}",
4744 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4745 (loadi64 addr:$src))))))],
4747 XS, Requires<[HasSSE2]>;
4749 let Predicates = [HasAVX], AddedComplexity = 20 in {
4750 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4751 (VMOVZQI2PQIrm addr:$src)>;
4752 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4753 (VMOVZQI2PQIrm addr:$src)>;
4754 def : Pat<(v2i64 (X86vzload addr:$src)),
4755 (VMOVZQI2PQIrm addr:$src)>;
4758 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4759 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4760 (MOVZQI2PQIrm addr:$src)>;
4761 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4762 (MOVZQI2PQIrm addr:$src)>;
4763 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4766 let Predicates = [HasAVX] in {
4767 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4768 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4769 def : Pat<(v4i64 (X86vzload addr:$src)),
4770 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4773 //===---------------------------------------------------------------------===//
4774 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4775 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4777 let AddedComplexity = 15 in
4778 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4779 "vmovq\t{$src, $dst|$dst, $src}",
4780 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4782 XS, VEX, Requires<[HasAVX]>;
4783 let AddedComplexity = 15 in
4784 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4785 "movq\t{$src, $dst|$dst, $src}",
4786 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4788 XS, Requires<[HasSSE2]>;
4790 let AddedComplexity = 20 in
4791 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4792 "vmovq\t{$src, $dst|$dst, $src}",
4793 [(set VR128:$dst, (v2i64 (X86vzmovl
4794 (loadv2i64 addr:$src))))],
4796 XS, VEX, Requires<[HasAVX]>;
4797 let AddedComplexity = 20 in {
4798 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4799 "movq\t{$src, $dst|$dst, $src}",
4800 [(set VR128:$dst, (v2i64 (X86vzmovl
4801 (loadv2i64 addr:$src))))],
4803 XS, Requires<[HasSSE2]>;
4806 let AddedComplexity = 20 in {
4807 let Predicates = [HasAVX] in {
4808 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4809 (VMOVZPQILo2PQIrm addr:$src)>;
4810 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4811 (VMOVZPQILo2PQIrr VR128:$src)>;
4813 let Predicates = [HasSSE2] in {
4814 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4815 (MOVZPQILo2PQIrm addr:$src)>;
4816 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4817 (MOVZPQILo2PQIrr VR128:$src)>;
4821 // Instructions to match in the assembler
4822 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4823 "movq\t{$src, $dst|$dst, $src}", [],
4824 IIC_SSE_MOVDQ>, VEX, VEX_W;
4825 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4826 "movq\t{$src, $dst|$dst, $src}", [],
4827 IIC_SSE_MOVDQ>, VEX, VEX_W;
4828 // Recognize "movd" with GR64 destination, but encode as a "movq"
4829 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4830 "movd\t{$src, $dst|$dst, $src}", [],
4831 IIC_SSE_MOVDQ>, VEX, VEX_W;
4833 // Instructions for the disassembler
4834 // xr = XMM register
4837 let Predicates = [HasAVX] in
4838 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4839 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4840 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4841 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4843 //===---------------------------------------------------------------------===//
4844 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4845 //===---------------------------------------------------------------------===//
4846 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4847 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4848 X86MemOperand x86memop> {
4849 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4850 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4851 [(set RC:$dst, (vt (OpNode RC:$src)))],
4853 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4854 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4855 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4859 let Predicates = [HasAVX] in {
4860 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4861 v4f32, VR128, memopv4f32, f128mem>, VEX;
4862 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4863 v4f32, VR128, memopv4f32, f128mem>, VEX;
4864 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4865 v8f32, VR256, memopv8f32, f256mem>, VEX;
4866 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4867 v8f32, VR256, memopv8f32, f256mem>, VEX;
4869 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4870 memopv4f32, f128mem>;
4871 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4872 memopv4f32, f128mem>;
4874 let Predicates = [HasAVX] in {
4875 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4876 (VMOVSHDUPrr VR128:$src)>;
4877 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4878 (VMOVSHDUPrm addr:$src)>;
4879 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4880 (VMOVSLDUPrr VR128:$src)>;
4881 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4882 (VMOVSLDUPrm addr:$src)>;
4883 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4884 (VMOVSHDUPYrr VR256:$src)>;
4885 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4886 (VMOVSHDUPYrm addr:$src)>;
4887 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4888 (VMOVSLDUPYrr VR256:$src)>;
4889 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4890 (VMOVSLDUPYrm addr:$src)>;
4893 let Predicates = [HasSSE3] in {
4894 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4895 (MOVSHDUPrr VR128:$src)>;
4896 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4897 (MOVSHDUPrm addr:$src)>;
4898 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4899 (MOVSLDUPrr VR128:$src)>;
4900 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4901 (MOVSLDUPrm addr:$src)>;
4904 //===---------------------------------------------------------------------===//
4905 // SSE3 - Replicate Double FP - MOVDDUP
4906 //===---------------------------------------------------------------------===//
4908 multiclass sse3_replicate_dfp<string OpcodeStr> {
4909 let neverHasSideEffects = 1 in
4910 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4911 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4912 [], IIC_SSE_MOV_LH>;
4913 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4914 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4917 (scalar_to_vector (loadf64 addr:$src)))))],
4921 // FIXME: Merge with above classe when there're patterns for the ymm version
4922 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4923 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4924 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4925 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4926 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4927 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4930 (scalar_to_vector (loadf64 addr:$src)))))]>;
4933 let Predicates = [HasAVX] in {
4934 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4935 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4938 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4940 let Predicates = [HasAVX] in {
4941 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4942 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4943 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4944 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4945 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4946 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4947 def : Pat<(X86Movddup (bc_v2f64
4948 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4949 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4952 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4953 (VMOVDDUPYrm addr:$src)>;
4954 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4955 (VMOVDDUPYrm addr:$src)>;
4956 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4957 (VMOVDDUPYrm addr:$src)>;
4958 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4959 (VMOVDDUPYrr VR256:$src)>;
4962 let Predicates = [HasSSE3] in {
4963 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4964 (MOVDDUPrm addr:$src)>;
4965 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4966 (MOVDDUPrm addr:$src)>;
4967 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4968 (MOVDDUPrm addr:$src)>;
4969 def : Pat<(X86Movddup (bc_v2f64
4970 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4971 (MOVDDUPrm addr:$src)>;
4974 //===---------------------------------------------------------------------===//
4975 // SSE3 - Move Unaligned Integer
4976 //===---------------------------------------------------------------------===//
4978 let Predicates = [HasAVX] in {
4979 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4980 "vlddqu\t{$src, $dst|$dst, $src}",
4981 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4982 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4983 "vlddqu\t{$src, $dst|$dst, $src}",
4984 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4986 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4987 "lddqu\t{$src, $dst|$dst, $src}",
4988 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4991 //===---------------------------------------------------------------------===//
4992 // SSE3 - Arithmetic
4993 //===---------------------------------------------------------------------===//
4995 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4996 X86MemOperand x86memop, OpndItins itins,
4998 def rr : I<0xD0, MRMSrcReg,
4999 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5001 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5002 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5003 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
5004 def rm : I<0xD0, MRMSrcMem,
5005 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5007 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5009 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5012 let Predicates = [HasAVX] in {
5013 let ExeDomain = SSEPackedSingle in {
5014 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5015 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5016 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5017 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5019 let ExeDomain = SSEPackedDouble in {
5020 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5021 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5022 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5023 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5026 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5027 let ExeDomain = SSEPackedSingle in
5028 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5029 f128mem, SSE_ALU_F32P>, TB, XD;
5030 let ExeDomain = SSEPackedDouble in
5031 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5032 f128mem, SSE_ALU_F64P>, TB, OpSize;
5035 //===---------------------------------------------------------------------===//
5036 // SSE3 Instructions
5037 //===---------------------------------------------------------------------===//
5040 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5041 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5042 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5044 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5045 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5046 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5048 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5050 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5051 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5052 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5053 IIC_SSE_HADDSUB_RM>;
5055 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5056 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5057 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5059 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5060 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5061 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5063 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5065 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5066 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5067 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5068 IIC_SSE_HADDSUB_RM>;
5071 let Predicates = [HasAVX] in {
5072 let ExeDomain = SSEPackedSingle in {
5073 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5074 X86fhadd, 0>, VEX_4V;
5075 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5076 X86fhsub, 0>, VEX_4V;
5077 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5078 X86fhadd, 0>, VEX_4V;
5079 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5080 X86fhsub, 0>, VEX_4V;
5082 let ExeDomain = SSEPackedDouble in {
5083 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5084 X86fhadd, 0>, VEX_4V;
5085 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5086 X86fhsub, 0>, VEX_4V;
5087 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5088 X86fhadd, 0>, VEX_4V;
5089 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5090 X86fhsub, 0>, VEX_4V;
5094 let Constraints = "$src1 = $dst" in {
5095 let ExeDomain = SSEPackedSingle in {
5096 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5097 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5099 let ExeDomain = SSEPackedDouble in {
5100 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5101 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5105 //===---------------------------------------------------------------------===//
5106 // SSSE3 - Packed Absolute Instructions
5107 //===---------------------------------------------------------------------===//
5110 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5111 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5112 Intrinsic IntId128> {
5113 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5115 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5116 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5119 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5121 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5124 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5128 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5129 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5130 Intrinsic IntId256> {
5131 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5133 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5134 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5137 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5139 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5142 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5145 let Predicates = [HasAVX] in {
5146 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5147 int_x86_ssse3_pabs_b_128>, VEX;
5148 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5149 int_x86_ssse3_pabs_w_128>, VEX;
5150 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5151 int_x86_ssse3_pabs_d_128>, VEX;
5154 let Predicates = [HasAVX2] in {
5155 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5156 int_x86_avx2_pabs_b>, VEX;
5157 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5158 int_x86_avx2_pabs_w>, VEX;
5159 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5160 int_x86_avx2_pabs_d>, VEX;
5163 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5164 int_x86_ssse3_pabs_b_128>;
5165 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5166 int_x86_ssse3_pabs_w_128>;
5167 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5168 int_x86_ssse3_pabs_d_128>;
5170 //===---------------------------------------------------------------------===//
5171 // SSSE3 - Packed Binary Operator Instructions
5172 //===---------------------------------------------------------------------===//
5174 def SSE_PHADDSUBD : OpndItins<
5175 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5177 def SSE_PHADDSUBSW : OpndItins<
5178 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5180 def SSE_PHADDSUBW : OpndItins<
5181 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5183 def SSE_PSHUFB : OpndItins<
5184 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5186 def SSE_PSIGN : OpndItins<
5187 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5189 def SSE_PMULHRSW : OpndItins<
5190 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5193 /// SS3I_binop_rm - Simple SSSE3 bin op
5194 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5195 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5196 X86MemOperand x86memop, OpndItins itins,
5198 let isCommutable = 1 in
5199 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5200 (ins RC:$src1, RC:$src2),
5202 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5203 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5204 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5206 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5207 (ins RC:$src1, x86memop:$src2),
5209 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5210 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5212 (OpVT (OpNode RC:$src1,
5213 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5216 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5217 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5218 Intrinsic IntId128, OpndItins itins,
5220 let isCommutable = 1 in
5221 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5222 (ins VR128:$src1, VR128:$src2),
5224 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5225 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5226 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5228 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5229 (ins VR128:$src1, i128mem:$src2),
5231 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5232 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5234 (IntId128 VR128:$src1,
5235 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5238 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5239 Intrinsic IntId256> {
5240 let isCommutable = 1 in
5241 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5242 (ins VR256:$src1, VR256:$src2),
5243 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5244 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5246 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5247 (ins VR256:$src1, i256mem:$src2),
5248 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5250 (IntId256 VR256:$src1,
5251 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5254 let ImmT = NoImm, Predicates = [HasAVX] in {
5255 let isCommutable = 0 in {
5256 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5257 memopv2i64, i128mem,
5258 SSE_PHADDSUBW, 0>, VEX_4V;
5259 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5260 memopv2i64, i128mem,
5261 SSE_PHADDSUBD, 0>, VEX_4V;
5262 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5263 memopv2i64, i128mem,
5264 SSE_PHADDSUBW, 0>, VEX_4V;
5265 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5266 memopv2i64, i128mem,
5267 SSE_PHADDSUBD, 0>, VEX_4V;
5268 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5269 memopv2i64, i128mem,
5270 SSE_PSIGN, 0>, VEX_4V;
5271 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5272 memopv2i64, i128mem,
5273 SSE_PSIGN, 0>, VEX_4V;
5274 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5275 memopv2i64, i128mem,
5276 SSE_PSIGN, 0>, VEX_4V;
5277 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5278 memopv2i64, i128mem,
5279 SSE_PSHUFB, 0>, VEX_4V;
5280 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5281 int_x86_ssse3_phadd_sw_128,
5282 SSE_PHADDSUBSW, 0>, VEX_4V;
5283 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5284 int_x86_ssse3_phsub_sw_128,
5285 SSE_PHADDSUBSW, 0>, VEX_4V;
5286 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5287 int_x86_ssse3_pmadd_ub_sw_128,
5288 SSE_PMADD, 0>, VEX_4V;
5290 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5291 int_x86_ssse3_pmul_hr_sw_128,
5292 SSE_PMULHRSW, 0>, VEX_4V;
5295 let ImmT = NoImm, Predicates = [HasAVX2] in {
5296 let isCommutable = 0 in {
5297 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5298 memopv4i64, i256mem,
5299 SSE_PHADDSUBW, 0>, VEX_4V;
5300 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5301 memopv4i64, i256mem,
5302 SSE_PHADDSUBW, 0>, VEX_4V;
5303 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5304 memopv4i64, i256mem,
5305 SSE_PHADDSUBW, 0>, VEX_4V;
5306 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5307 memopv4i64, i256mem,
5308 SSE_PHADDSUBW, 0>, VEX_4V;
5309 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5310 memopv4i64, i256mem,
5311 SSE_PHADDSUBW, 0>, VEX_4V;
5312 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5313 memopv4i64, i256mem,
5314 SSE_PHADDSUBW, 0>, VEX_4V;
5315 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5316 memopv4i64, i256mem,
5317 SSE_PHADDSUBW, 0>, VEX_4V;
5318 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5319 memopv4i64, i256mem,
5320 SSE_PHADDSUBW, 0>, VEX_4V;
5321 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5322 int_x86_avx2_phadd_sw>, VEX_4V;
5323 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5324 int_x86_avx2_phsub_sw>, VEX_4V;
5325 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5326 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5328 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5329 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5332 // None of these have i8 immediate fields.
5333 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5334 let isCommutable = 0 in {
5335 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5336 memopv2i64, i128mem, SSE_PHADDSUBW>;
5337 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5338 memopv2i64, i128mem, SSE_PHADDSUBD>;
5339 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5340 memopv2i64, i128mem, SSE_PHADDSUBW>;
5341 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5342 memopv2i64, i128mem, SSE_PHADDSUBD>;
5343 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5344 memopv2i64, i128mem, SSE_PSIGN>;
5345 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5346 memopv2i64, i128mem, SSE_PSIGN>;
5347 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5348 memopv2i64, i128mem, SSE_PSIGN>;
5349 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5350 memopv2i64, i128mem, SSE_PSHUFB>;
5351 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5352 int_x86_ssse3_phadd_sw_128,
5354 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5355 int_x86_ssse3_phsub_sw_128,
5357 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5358 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5360 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5361 int_x86_ssse3_pmul_hr_sw_128,
5365 //===---------------------------------------------------------------------===//
5366 // SSSE3 - Packed Align Instruction Patterns
5367 //===---------------------------------------------------------------------===//
5369 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5370 let neverHasSideEffects = 1 in {
5371 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5372 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5374 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5376 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5377 [], IIC_SSE_PALIGNR>, OpSize;
5379 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5380 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5382 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5384 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5385 [], IIC_SSE_PALIGNR>, OpSize;
5389 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5390 let neverHasSideEffects = 1 in {
5391 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5392 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5394 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5397 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5398 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5400 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5405 let Predicates = [HasAVX] in
5406 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5407 let Predicates = [HasAVX2] in
5408 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5409 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5410 defm PALIGN : ssse3_palign<"palignr">;
5412 let Predicates = [HasAVX2] in {
5413 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5414 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5415 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5416 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5417 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5418 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5419 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5420 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5423 let Predicates = [HasAVX] in {
5424 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5425 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5426 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5427 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5428 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5429 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5430 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5431 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5434 let Predicates = [HasSSSE3] in {
5435 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5436 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5437 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5438 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5439 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5440 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5441 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5442 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5445 //===---------------------------------------------------------------------===//
5446 // SSSE3 - Thread synchronization
5447 //===---------------------------------------------------------------------===//
5449 let usesCustomInserter = 1 in {
5450 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5451 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5452 Requires<[HasSSE3]>;
5455 let Uses = [EAX, ECX, EDX] in
5456 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5457 TB, Requires<[HasSSE3]>;
5458 let Uses = [ECX, EAX] in
5459 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5460 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5461 TB, Requires<[HasSSE3]>;
5463 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5464 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5466 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5467 Requires<[In32BitMode]>;
5468 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5469 Requires<[In64BitMode]>;
5471 //===----------------------------------------------------------------------===//
5472 // SSE4.1 - Packed Move with Sign/Zero Extend
5473 //===----------------------------------------------------------------------===//
5475 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5476 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5477 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5478 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5480 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5481 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5483 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5487 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5489 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5491 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5493 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5495 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5498 let Predicates = [HasAVX] in {
5499 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5501 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5503 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5505 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5507 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5509 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5513 let Predicates = [HasAVX2] in {
5514 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5515 int_x86_avx2_pmovsxbw>, VEX;
5516 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5517 int_x86_avx2_pmovsxwd>, VEX;
5518 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5519 int_x86_avx2_pmovsxdq>, VEX;
5520 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5521 int_x86_avx2_pmovzxbw>, VEX;
5522 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5523 int_x86_avx2_pmovzxwd>, VEX;
5524 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5525 int_x86_avx2_pmovzxdq>, VEX;
5528 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5529 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5530 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5531 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5532 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5533 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5535 let Predicates = [HasAVX] in {
5536 // Common patterns involving scalar load.
5537 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5538 (VPMOVSXBWrm addr:$src)>;
5539 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5540 (VPMOVSXBWrm addr:$src)>;
5542 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5543 (VPMOVSXWDrm addr:$src)>;
5544 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5545 (VPMOVSXWDrm addr:$src)>;
5547 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5548 (VPMOVSXDQrm addr:$src)>;
5549 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5550 (VPMOVSXDQrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5553 (VPMOVZXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5555 (VPMOVZXBWrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5558 (VPMOVZXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5560 (VPMOVZXWDrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5563 (VPMOVZXDQrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5565 (VPMOVZXDQrm addr:$src)>;
5568 let Predicates = [HasSSE41] in {
5569 // Common patterns involving scalar load.
5570 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5571 (PMOVSXBWrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5573 (PMOVSXBWrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5576 (PMOVSXWDrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5578 (PMOVSXWDrm addr:$src)>;
5580 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5581 (PMOVSXDQrm addr:$src)>;
5582 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5583 (PMOVSXDQrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5586 (PMOVZXBWrm addr:$src)>;
5587 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5588 (PMOVZXBWrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5591 (PMOVZXWDrm addr:$src)>;
5592 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5593 (PMOVZXWDrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5596 (PMOVZXDQrm addr:$src)>;
5597 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5598 (PMOVZXDQrm addr:$src)>;
5601 let Predicates = [HasAVX2] in {
5602 let AddedComplexity = 15 in {
5603 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5604 (VPMOVZXDQYrr VR128:$src)>;
5605 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5606 (VPMOVZXWDYrr VR128:$src)>;
5609 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5610 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5613 let Predicates = [HasAVX] in {
5614 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5615 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5618 let Predicates = [HasSSE41] in {
5619 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5620 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5624 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5625 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5626 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5627 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5629 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5630 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5632 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5636 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5638 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5639 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5640 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5642 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5643 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5645 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5649 let Predicates = [HasAVX] in {
5650 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5652 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5654 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5656 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5660 let Predicates = [HasAVX2] in {
5661 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5662 int_x86_avx2_pmovsxbd>, VEX;
5663 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5664 int_x86_avx2_pmovsxwq>, VEX;
5665 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5666 int_x86_avx2_pmovzxbd>, VEX;
5667 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5668 int_x86_avx2_pmovzxwq>, VEX;
5671 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5672 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5673 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5674 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5676 let Predicates = [HasAVX] in {
5677 // Common patterns involving scalar load
5678 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5679 (VPMOVSXBDrm addr:$src)>;
5680 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5681 (VPMOVSXWQrm addr:$src)>;
5683 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5684 (VPMOVZXBDrm addr:$src)>;
5685 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5686 (VPMOVZXWQrm addr:$src)>;
5689 let Predicates = [HasSSE41] in {
5690 // Common patterns involving scalar load
5691 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5692 (PMOVSXBDrm addr:$src)>;
5693 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5694 (PMOVSXWQrm addr:$src)>;
5696 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5697 (PMOVZXBDrm addr:$src)>;
5698 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5699 (PMOVZXWQrm addr:$src)>;
5702 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5703 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5704 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5705 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5707 // Expecting a i16 load any extended to i32 value.
5708 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5709 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5710 [(set VR128:$dst, (IntId (bitconvert
5711 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5715 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5717 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5718 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5719 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5721 // Expecting a i16 load any extended to i32 value.
5722 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5723 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5724 [(set VR256:$dst, (IntId (bitconvert
5725 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5729 let Predicates = [HasAVX] in {
5730 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5732 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5735 let Predicates = [HasAVX2] in {
5736 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5737 int_x86_avx2_pmovsxbq>, VEX;
5738 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5739 int_x86_avx2_pmovzxbq>, VEX;
5741 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5742 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5744 let Predicates = [HasAVX] in {
5745 // Common patterns involving scalar load
5746 def : Pat<(int_x86_sse41_pmovsxbq
5747 (bitconvert (v4i32 (X86vzmovl
5748 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5749 (VPMOVSXBQrm addr:$src)>;
5751 def : Pat<(int_x86_sse41_pmovzxbq
5752 (bitconvert (v4i32 (X86vzmovl
5753 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5754 (VPMOVZXBQrm addr:$src)>;
5757 let Predicates = [HasSSE41] in {
5758 // Common patterns involving scalar load
5759 def : Pat<(int_x86_sse41_pmovsxbq
5760 (bitconvert (v4i32 (X86vzmovl
5761 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5762 (PMOVSXBQrm addr:$src)>;
5764 def : Pat<(int_x86_sse41_pmovzxbq
5765 (bitconvert (v4i32 (X86vzmovl
5766 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5767 (PMOVZXBQrm addr:$src)>;
5770 //===----------------------------------------------------------------------===//
5771 // SSE4.1 - Extract Instructions
5772 //===----------------------------------------------------------------------===//
5774 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5775 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5776 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5777 (ins VR128:$src1, i32i8imm:$src2),
5778 !strconcat(OpcodeStr,
5779 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5780 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5782 let neverHasSideEffects = 1, mayStore = 1 in
5783 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5784 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5785 !strconcat(OpcodeStr,
5786 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5789 // There's an AssertZext in the way of writing the store pattern
5790 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5793 let Predicates = [HasAVX] in {
5794 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5795 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5796 (ins VR128:$src1, i32i8imm:$src2),
5797 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5800 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5803 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5804 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5805 let neverHasSideEffects = 1, mayStore = 1 in
5806 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5807 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5808 !strconcat(OpcodeStr,
5809 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5812 // There's an AssertZext in the way of writing the store pattern
5813 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5816 let Predicates = [HasAVX] in
5817 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5819 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5822 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5823 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5824 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5825 (ins VR128:$src1, i32i8imm:$src2),
5826 !strconcat(OpcodeStr,
5827 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5829 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5830 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5831 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5832 !strconcat(OpcodeStr,
5833 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5834 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5835 addr:$dst)]>, OpSize;
5838 let Predicates = [HasAVX] in
5839 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5841 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5843 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5844 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5845 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5846 (ins VR128:$src1, i32i8imm:$src2),
5847 !strconcat(OpcodeStr,
5848 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5850 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5851 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5852 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5853 !strconcat(OpcodeStr,
5854 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5855 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5856 addr:$dst)]>, OpSize, REX_W;
5859 let Predicates = [HasAVX] in
5860 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5862 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5864 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5866 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5867 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5868 (ins VR128:$src1, i32i8imm:$src2),
5869 !strconcat(OpcodeStr,
5870 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5872 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5874 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5875 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5876 !strconcat(OpcodeStr,
5877 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5878 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5879 addr:$dst)]>, OpSize;
5882 let ExeDomain = SSEPackedSingle in {
5883 let Predicates = [HasAVX] in {
5884 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5885 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5886 (ins VR128:$src1, i32i8imm:$src2),
5887 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5890 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5893 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5894 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5897 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5899 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5902 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5903 Requires<[HasSSE41]>;
5905 //===----------------------------------------------------------------------===//
5906 // SSE4.1 - Insert Instructions
5907 //===----------------------------------------------------------------------===//
5909 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5910 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5911 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5913 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5915 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5917 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5918 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5919 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5921 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5923 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5925 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5926 imm:$src3))]>, OpSize;
5929 let Predicates = [HasAVX] in
5930 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5931 let Constraints = "$src1 = $dst" in
5932 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5934 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5935 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5936 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5938 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5940 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5942 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5944 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5945 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5947 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5949 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5951 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5952 imm:$src3)))]>, OpSize;
5955 let Predicates = [HasAVX] in
5956 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5957 let Constraints = "$src1 = $dst" in
5958 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5960 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5961 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5962 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5964 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5966 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5968 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5970 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5971 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5973 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5975 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5977 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5978 imm:$src3)))]>, OpSize;
5981 let Predicates = [HasAVX] in
5982 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5983 let Constraints = "$src1 = $dst" in
5984 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5986 // insertps has a few different modes, there's the first two here below which
5987 // are optimized inserts that won't zero arbitrary elements in the destination
5988 // vector. The next one matches the intrinsic and could zero arbitrary elements
5989 // in the target vector.
5990 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5991 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5992 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5994 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5996 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5998 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6000 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6001 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6003 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6005 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6007 (X86insrtps VR128:$src1,
6008 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6009 imm:$src3))]>, OpSize;
6012 let ExeDomain = SSEPackedSingle in {
6013 let Predicates = [HasAVX] in
6014 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6015 let Constraints = "$src1 = $dst" in
6016 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6019 //===----------------------------------------------------------------------===//
6020 // SSE4.1 - Round Instructions
6021 //===----------------------------------------------------------------------===//
6023 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6024 X86MemOperand x86memop, RegisterClass RC,
6025 PatFrag mem_frag32, PatFrag mem_frag64,
6026 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6027 let ExeDomain = SSEPackedSingle in {
6028 // Intrinsic operation, reg.
6029 // Vector intrinsic operation, reg
6030 def PSr : SS4AIi8<opcps, MRMSrcReg,
6031 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6032 !strconcat(OpcodeStr,
6033 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6034 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6037 // Vector intrinsic operation, mem
6038 def PSm : SS4AIi8<opcps, MRMSrcMem,
6039 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6040 !strconcat(OpcodeStr,
6041 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6043 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6045 } // ExeDomain = SSEPackedSingle
6047 let ExeDomain = SSEPackedDouble in {
6048 // Vector intrinsic operation, reg
6049 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6050 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6051 !strconcat(OpcodeStr,
6052 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6053 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6056 // Vector intrinsic operation, mem
6057 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6058 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6059 !strconcat(OpcodeStr,
6060 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6062 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6064 } // ExeDomain = SSEPackedDouble
6067 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6070 Intrinsic F64Int, bit Is2Addr = 1> {
6071 let ExeDomain = GenericDomain in {
6073 def SSr : SS4AIi8<opcss, MRMSrcReg,
6074 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6076 !strconcat(OpcodeStr,
6077 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6078 !strconcat(OpcodeStr,
6079 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6082 // Intrinsic operation, reg.
6083 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6084 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6086 !strconcat(OpcodeStr,
6087 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6088 !strconcat(OpcodeStr,
6089 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6090 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6093 // Intrinsic operation, mem.
6094 def SSm : SS4AIi8<opcss, MRMSrcMem,
6095 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6097 !strconcat(OpcodeStr,
6098 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6099 !strconcat(OpcodeStr,
6100 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6102 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6106 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6107 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6109 !strconcat(OpcodeStr,
6110 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6111 !strconcat(OpcodeStr,
6112 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6115 // Intrinsic operation, reg.
6116 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6117 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6119 !strconcat(OpcodeStr,
6120 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6121 !strconcat(OpcodeStr,
6122 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6123 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6126 // Intrinsic operation, mem.
6127 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6128 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6130 !strconcat(OpcodeStr,
6131 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6132 !strconcat(OpcodeStr,
6133 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6135 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6137 } // ExeDomain = GenericDomain
6140 // FP round - roundss, roundps, roundsd, roundpd
6141 let Predicates = [HasAVX] in {
6143 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6144 memopv4f32, memopv2f64,
6145 int_x86_sse41_round_ps,
6146 int_x86_sse41_round_pd>, VEX;
6147 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6148 memopv8f32, memopv4f64,
6149 int_x86_avx_round_ps_256,
6150 int_x86_avx_round_pd_256>, VEX;
6151 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6152 int_x86_sse41_round_ss,
6153 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6155 def : Pat<(ffloor FR32:$src),
6156 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6157 def : Pat<(f64 (ffloor FR64:$src)),
6158 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6159 def : Pat<(f32 (fnearbyint FR32:$src)),
6160 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6161 def : Pat<(f64 (fnearbyint FR64:$src)),
6162 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6163 def : Pat<(f32 (fceil FR32:$src)),
6164 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6165 def : Pat<(f64 (fceil FR64:$src)),
6166 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6167 def : Pat<(f32 (frint FR32:$src)),
6168 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6169 def : Pat<(f64 (frint FR64:$src)),
6170 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6171 def : Pat<(f32 (ftrunc FR32:$src)),
6172 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6173 def : Pat<(f64 (ftrunc FR64:$src)),
6174 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6177 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6178 memopv4f32, memopv2f64,
6179 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6180 let Constraints = "$src1 = $dst" in
6181 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6182 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6184 def : Pat<(ffloor FR32:$src),
6185 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6186 def : Pat<(f64 (ffloor FR64:$src)),
6187 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6188 def : Pat<(f32 (fnearbyint FR32:$src)),
6189 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6190 def : Pat<(f64 (fnearbyint FR64:$src)),
6191 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6192 def : Pat<(f32 (fceil FR32:$src)),
6193 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6194 def : Pat<(f64 (fceil FR64:$src)),
6195 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6196 def : Pat<(f32 (frint FR32:$src)),
6197 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6198 def : Pat<(f64 (frint FR64:$src)),
6199 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6200 def : Pat<(f32 (ftrunc FR32:$src)),
6201 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6202 def : Pat<(f64 (ftrunc FR64:$src)),
6203 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6205 //===----------------------------------------------------------------------===//
6206 // SSE4.1 - Packed Bit Test
6207 //===----------------------------------------------------------------------===//
6209 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6210 // the intel intrinsic that corresponds to this.
6211 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6212 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6213 "vptest\t{$src2, $src1|$src1, $src2}",
6214 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6216 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6217 "vptest\t{$src2, $src1|$src1, $src2}",
6218 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6221 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6222 "vptest\t{$src2, $src1|$src1, $src2}",
6223 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6225 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6226 "vptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6231 let Defs = [EFLAGS] in {
6232 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6233 "ptest\t{$src2, $src1|$src1, $src2}",
6234 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6236 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6237 "ptest\t{$src2, $src1|$src1, $src2}",
6238 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6242 // The bit test instructions below are AVX only
6243 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6244 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6245 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6246 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6247 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6248 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6249 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6250 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6254 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6255 let ExeDomain = SSEPackedSingle in {
6256 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6257 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6259 let ExeDomain = SSEPackedDouble in {
6260 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6261 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6265 //===----------------------------------------------------------------------===//
6266 // SSE4.1 - Misc Instructions
6267 //===----------------------------------------------------------------------===//
6269 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6270 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6271 "popcnt{w}\t{$src, $dst|$dst, $src}",
6272 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6274 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6275 "popcnt{w}\t{$src, $dst|$dst, $src}",
6276 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6277 (implicit EFLAGS)]>, OpSize, XS;
6279 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6280 "popcnt{l}\t{$src, $dst|$dst, $src}",
6281 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6283 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6284 "popcnt{l}\t{$src, $dst|$dst, $src}",
6285 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6286 (implicit EFLAGS)]>, XS;
6288 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6289 "popcnt{q}\t{$src, $dst|$dst, $src}",
6290 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6292 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6293 "popcnt{q}\t{$src, $dst|$dst, $src}",
6294 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6295 (implicit EFLAGS)]>, XS;
6300 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6301 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6302 Intrinsic IntId128> {
6303 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6305 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6306 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6307 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6309 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6312 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6315 let Predicates = [HasAVX] in
6316 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6317 int_x86_sse41_phminposuw>, VEX;
6318 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6319 int_x86_sse41_phminposuw>;
6321 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6322 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6323 Intrinsic IntId128, bit Is2Addr = 1> {
6324 let isCommutable = 1 in
6325 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6326 (ins VR128:$src1, VR128:$src2),
6328 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6330 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6331 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6332 (ins VR128:$src1, i128mem:$src2),
6334 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6335 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6337 (IntId128 VR128:$src1,
6338 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6341 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6342 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6343 Intrinsic IntId256> {
6344 let isCommutable = 1 in
6345 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6346 (ins VR256:$src1, VR256:$src2),
6347 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6348 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6349 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6350 (ins VR256:$src1, i256mem:$src2),
6351 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6353 (IntId256 VR256:$src1,
6354 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6357 let Predicates = [HasAVX] in {
6358 let isCommutable = 0 in
6359 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6361 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6363 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6365 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6367 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6369 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6371 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6373 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6375 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6377 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6381 let Predicates = [HasAVX2] in {
6382 let isCommutable = 0 in
6383 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6384 int_x86_avx2_packusdw>, VEX_4V;
6385 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6386 int_x86_avx2_pmins_b>, VEX_4V;
6387 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6388 int_x86_avx2_pmins_d>, VEX_4V;
6389 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6390 int_x86_avx2_pminu_d>, VEX_4V;
6391 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6392 int_x86_avx2_pminu_w>, VEX_4V;
6393 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6394 int_x86_avx2_pmaxs_b>, VEX_4V;
6395 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6396 int_x86_avx2_pmaxs_d>, VEX_4V;
6397 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6398 int_x86_avx2_pmaxu_d>, VEX_4V;
6399 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6400 int_x86_avx2_pmaxu_w>, VEX_4V;
6401 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6402 int_x86_avx2_pmul_dq>, VEX_4V;
6405 let Constraints = "$src1 = $dst" in {
6406 let isCommutable = 0 in
6407 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6408 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6409 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6410 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6411 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6412 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6413 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6414 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6415 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6416 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6419 /// SS48I_binop_rm - Simple SSE41 binary operator.
6420 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6421 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6422 X86MemOperand x86memop, bit Is2Addr = 1> {
6423 let isCommutable = 1 in
6424 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6425 (ins RC:$src1, RC:$src2),
6427 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6428 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6429 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6430 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6431 (ins RC:$src1, x86memop:$src2),
6433 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6434 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6436 (OpVT (OpNode RC:$src1,
6437 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6440 let Predicates = [HasAVX] in {
6441 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6442 memopv2i64, i128mem, 0>, VEX_4V;
6443 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6444 memopv2i64, i128mem, 0>, VEX_4V;
6446 let Predicates = [HasAVX2] in {
6447 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6448 memopv4i64, i256mem, 0>, VEX_4V;
6449 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6450 memopv4i64, i256mem, 0>, VEX_4V;
6453 let Constraints = "$src1 = $dst" in {
6454 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6455 memopv2i64, i128mem>;
6456 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6457 memopv2i64, i128mem>;
6460 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6461 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6462 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6463 X86MemOperand x86memop, bit Is2Addr = 1> {
6464 let isCommutable = 1 in
6465 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6466 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6468 !strconcat(OpcodeStr,
6469 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6470 !strconcat(OpcodeStr,
6471 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6472 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6474 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6475 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6477 !strconcat(OpcodeStr,
6478 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6479 !strconcat(OpcodeStr,
6480 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6483 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6487 let Predicates = [HasAVX] in {
6488 let isCommutable = 0 in {
6489 let ExeDomain = SSEPackedSingle in {
6490 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6491 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6492 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6493 int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
6495 let ExeDomain = SSEPackedDouble in {
6496 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6497 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6498 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6499 int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
6501 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6502 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6503 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6504 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6506 let ExeDomain = SSEPackedSingle in
6507 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6508 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6509 let ExeDomain = SSEPackedDouble in
6510 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6511 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6512 let ExeDomain = SSEPackedSingle in
6513 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6514 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6517 let Predicates = [HasAVX2] in {
6518 let isCommutable = 0 in {
6519 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6520 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6521 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6522 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6526 let Constraints = "$src1 = $dst" in {
6527 let isCommutable = 0 in {
6528 let ExeDomain = SSEPackedSingle in
6529 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6530 VR128, memopv4f32, f128mem>;
6531 let ExeDomain = SSEPackedDouble in
6532 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6533 VR128, memopv2f64, f128mem>;
6534 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6535 VR128, memopv2i64, i128mem>;
6536 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6537 VR128, memopv2i64, i128mem>;
6539 let ExeDomain = SSEPackedSingle in
6540 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6541 VR128, memopv4f32, f128mem>;
6542 let ExeDomain = SSEPackedDouble in
6543 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6544 VR128, memopv2f64, f128mem>;
6547 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6548 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6549 RegisterClass RC, X86MemOperand x86memop,
6550 PatFrag mem_frag, Intrinsic IntId> {
6551 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6552 (ins RC:$src1, RC:$src2, RC:$src3),
6553 !strconcat(OpcodeStr,
6554 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6555 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6556 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6558 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6559 (ins RC:$src1, x86memop:$src2, RC:$src3),
6560 !strconcat(OpcodeStr,
6561 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6563 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6565 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6568 let Predicates = [HasAVX] in {
6569 let ExeDomain = SSEPackedDouble in {
6570 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6571 memopv2f64, int_x86_sse41_blendvpd>;
6572 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6573 memopv4f64, int_x86_avx_blendv_pd_256>;
6574 } // ExeDomain = SSEPackedDouble
6575 let ExeDomain = SSEPackedSingle in {
6576 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6577 memopv4f32, int_x86_sse41_blendvps>;
6578 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6579 memopv8f32, int_x86_avx_blendv_ps_256>;
6580 } // ExeDomain = SSEPackedSingle
6581 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6582 memopv2i64, int_x86_sse41_pblendvb>;
6585 let Predicates = [HasAVX2] in {
6586 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6587 memopv4i64, int_x86_avx2_pblendvb>;
6590 let Predicates = [HasAVX] in {
6591 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6592 (v16i8 VR128:$src2))),
6593 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6594 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6595 (v4i32 VR128:$src2))),
6596 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6597 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6598 (v4f32 VR128:$src2))),
6599 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6600 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6601 (v2i64 VR128:$src2))),
6602 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6603 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6604 (v2f64 VR128:$src2))),
6605 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6606 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6607 (v8i32 VR256:$src2))),
6608 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6609 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6610 (v8f32 VR256:$src2))),
6611 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6612 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6613 (v4i64 VR256:$src2))),
6614 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6615 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6616 (v4f64 VR256:$src2))),
6617 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6619 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6621 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6622 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6624 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6626 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6628 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6629 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6631 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6632 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6634 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6637 let Predicates = [HasAVX2] in {
6638 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6639 (v32i8 VR256:$src2))),
6640 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6641 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6643 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6646 /// SS41I_ternary_int - SSE 4.1 ternary operator
6647 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6648 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6649 X86MemOperand x86memop, Intrinsic IntId> {
6650 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6651 (ins VR128:$src1, VR128:$src2),
6652 !strconcat(OpcodeStr,
6653 "\t{$src2, $dst|$dst, $src2}"),
6654 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6657 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6658 (ins VR128:$src1, x86memop:$src2),
6659 !strconcat(OpcodeStr,
6660 "\t{$src2, $dst|$dst, $src2}"),
6663 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6667 let ExeDomain = SSEPackedDouble in
6668 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6669 int_x86_sse41_blendvpd>;
6670 let ExeDomain = SSEPackedSingle in
6671 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6672 int_x86_sse41_blendvps>;
6673 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6674 int_x86_sse41_pblendvb>;
6676 // Aliases with the implicit xmm0 argument
6677 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6678 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6679 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6680 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6681 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6682 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6683 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6684 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6685 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6686 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6687 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6688 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6690 let Predicates = [HasSSE41] in {
6691 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6692 (v16i8 VR128:$src2))),
6693 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6694 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6695 (v4i32 VR128:$src2))),
6696 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6697 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6698 (v4f32 VR128:$src2))),
6699 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6700 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6701 (v2i64 VR128:$src2))),
6702 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6703 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6704 (v2f64 VR128:$src2))),
6705 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6707 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6709 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6710 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6712 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6713 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6715 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6719 let Predicates = [HasAVX] in
6720 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6721 "vmovntdqa\t{$src, $dst|$dst, $src}",
6722 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6724 let Predicates = [HasAVX2] in
6725 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6726 "vmovntdqa\t{$src, $dst|$dst, $src}",
6727 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6729 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6730 "movntdqa\t{$src, $dst|$dst, $src}",
6731 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6734 //===----------------------------------------------------------------------===//
6735 // SSE4.2 - Compare Instructions
6736 //===----------------------------------------------------------------------===//
6738 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6739 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6740 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6741 X86MemOperand x86memop, bit Is2Addr = 1> {
6742 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6743 (ins RC:$src1, RC:$src2),
6745 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6746 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6747 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6749 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6750 (ins RC:$src1, x86memop:$src2),
6752 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6753 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6755 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6758 let Predicates = [HasAVX] in
6759 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6760 memopv2i64, i128mem, 0>, VEX_4V;
6762 let Predicates = [HasAVX2] in
6763 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6764 memopv4i64, i256mem, 0>, VEX_4V;
6766 let Constraints = "$src1 = $dst" in
6767 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6768 memopv2i64, i128mem>;
6770 //===----------------------------------------------------------------------===//
6771 // SSE4.2 - String/text Processing Instructions
6772 //===----------------------------------------------------------------------===//
6774 // Packed Compare Implicit Length Strings, Return Mask
6775 multiclass pseudo_pcmpistrm<string asm> {
6776 def REG : PseudoI<(outs VR128:$dst),
6777 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6778 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6780 def MEM : PseudoI<(outs VR128:$dst),
6781 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6782 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6783 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6786 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6787 let AddedComplexity = 1 in
6788 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6789 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6792 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6793 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6794 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6795 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6797 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6798 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6799 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6802 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6803 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6804 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6805 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6807 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6808 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6809 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6812 // Packed Compare Explicit Length Strings, Return Mask
6813 multiclass pseudo_pcmpestrm<string asm> {
6814 def REG : PseudoI<(outs VR128:$dst),
6815 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6816 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6817 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6818 def MEM : PseudoI<(outs VR128:$dst),
6819 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6820 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6821 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6824 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6825 let AddedComplexity = 1 in
6826 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6827 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6830 let Predicates = [HasAVX],
6831 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6832 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6833 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6834 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6836 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6837 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6838 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6841 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6842 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6843 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6844 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6846 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6847 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6848 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6851 // Packed Compare Implicit Length Strings, Return Index
6852 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
6853 multiclass SS42AI_pcmpistri<string asm> {
6854 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6855 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6856 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6859 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6860 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6861 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6866 let Predicates = [HasAVX] in
6867 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
6868 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
6870 // Packed Compare Explicit Length Strings, Return Index
6871 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6872 multiclass SS42AI_pcmpestri<string asm> {
6873 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6874 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6875 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6878 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6879 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6880 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6885 let Predicates = [HasAVX] in
6886 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
6887 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
6889 //===----------------------------------------------------------------------===//
6890 // SSE4.2 - CRC Instructions
6891 //===----------------------------------------------------------------------===//
6893 // No CRC instructions have AVX equivalents
6895 // crc intrinsic instruction
6896 // This set of instructions are only rm, the only difference is the size
6898 let Constraints = "$src1 = $dst" in {
6899 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6900 (ins GR32:$src1, i8mem:$src2),
6901 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6903 (int_x86_sse42_crc32_32_8 GR32:$src1,
6904 (load addr:$src2)))]>;
6905 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6906 (ins GR32:$src1, GR8:$src2),
6907 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6909 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6910 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6911 (ins GR32:$src1, i16mem:$src2),
6912 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6914 (int_x86_sse42_crc32_32_16 GR32:$src1,
6915 (load addr:$src2)))]>,
6917 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6918 (ins GR32:$src1, GR16:$src2),
6919 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6921 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6923 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6924 (ins GR32:$src1, i32mem:$src2),
6925 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6927 (int_x86_sse42_crc32_32_32 GR32:$src1,
6928 (load addr:$src2)))]>;
6929 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6930 (ins GR32:$src1, GR32:$src2),
6931 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6933 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6934 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6935 (ins GR64:$src1, i8mem:$src2),
6936 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6938 (int_x86_sse42_crc32_64_8 GR64:$src1,
6939 (load addr:$src2)))]>,
6941 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6942 (ins GR64:$src1, GR8:$src2),
6943 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6945 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6947 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6948 (ins GR64:$src1, i64mem:$src2),
6949 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6951 (int_x86_sse42_crc32_64_64 GR64:$src1,
6952 (load addr:$src2)))]>,
6954 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6955 (ins GR64:$src1, GR64:$src2),
6956 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6958 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6962 //===----------------------------------------------------------------------===//
6963 // AES-NI Instructions
6964 //===----------------------------------------------------------------------===//
6966 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6967 Intrinsic IntId128, bit Is2Addr = 1> {
6968 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6969 (ins VR128:$src1, VR128:$src2),
6971 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6972 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6973 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6975 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6976 (ins VR128:$src1, i128mem:$src2),
6978 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6981 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6984 // Perform One Round of an AES Encryption/Decryption Flow
6985 let Predicates = [HasAVX, HasAES] in {
6986 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6987 int_x86_aesni_aesenc, 0>, VEX_4V;
6988 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6989 int_x86_aesni_aesenclast, 0>, VEX_4V;
6990 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6991 int_x86_aesni_aesdec, 0>, VEX_4V;
6992 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6993 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6996 let Constraints = "$src1 = $dst" in {
6997 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6998 int_x86_aesni_aesenc>;
6999 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7000 int_x86_aesni_aesenclast>;
7001 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7002 int_x86_aesni_aesdec>;
7003 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7004 int_x86_aesni_aesdeclast>;
7007 // Perform the AES InvMixColumn Transformation
7008 let Predicates = [HasAVX, HasAES] in {
7009 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7011 "vaesimc\t{$src1, $dst|$dst, $src1}",
7013 (int_x86_aesni_aesimc VR128:$src1))]>,
7015 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7016 (ins i128mem:$src1),
7017 "vaesimc\t{$src1, $dst|$dst, $src1}",
7018 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7021 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7023 "aesimc\t{$src1, $dst|$dst, $src1}",
7025 (int_x86_aesni_aesimc VR128:$src1))]>,
7027 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7028 (ins i128mem:$src1),
7029 "aesimc\t{$src1, $dst|$dst, $src1}",
7030 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7033 // AES Round Key Generation Assist
7034 let Predicates = [HasAVX, HasAES] in {
7035 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7036 (ins VR128:$src1, i8imm:$src2),
7037 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7039 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7041 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7042 (ins i128mem:$src1, i8imm:$src2),
7043 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7045 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7048 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7049 (ins VR128:$src1, i8imm:$src2),
7050 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7052 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7054 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7055 (ins i128mem:$src1, i8imm:$src2),
7056 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7058 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7061 //===----------------------------------------------------------------------===//
7062 // PCLMUL Instructions
7063 //===----------------------------------------------------------------------===//
7065 // AVX carry-less Multiplication instructions
7066 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7067 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7068 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7070 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7072 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7073 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7074 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7075 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7076 (memopv2i64 addr:$src2), imm:$src3))]>;
7078 // Carry-less Multiplication instructions
7079 let Constraints = "$src1 = $dst" in {
7080 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7081 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7082 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7084 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7086 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7087 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7088 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7089 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7090 (memopv2i64 addr:$src2), imm:$src3))]>;
7091 } // Constraints = "$src1 = $dst"
7094 multiclass pclmul_alias<string asm, int immop> {
7095 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7096 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7098 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7099 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7101 def : InstAlias<!strconcat("vpclmul", asm,
7102 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7103 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7105 def : InstAlias<!strconcat("vpclmul", asm,
7106 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7107 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7109 defm : pclmul_alias<"hqhq", 0x11>;
7110 defm : pclmul_alias<"hqlq", 0x01>;
7111 defm : pclmul_alias<"lqhq", 0x10>;
7112 defm : pclmul_alias<"lqlq", 0x00>;
7114 //===----------------------------------------------------------------------===//
7115 // SSE4A Instructions
7116 //===----------------------------------------------------------------------===//
7118 let Predicates = [HasSSE4A] in {
7120 let Constraints = "$src = $dst" in {
7121 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7122 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7123 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7124 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7125 imm:$idx))]>, TB, OpSize;
7126 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7127 (ins VR128:$src, VR128:$mask),
7128 "extrq\t{$mask, $src|$src, $mask}",
7129 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7130 VR128:$mask))]>, TB, OpSize;
7132 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7133 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7134 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7135 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7136 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7137 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7138 (ins VR128:$src, VR128:$mask),
7139 "insertq\t{$mask, $src|$src, $mask}",
7140 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7141 VR128:$mask))]>, XD;
7144 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7145 "movntss\t{$src, $dst|$dst, $src}",
7146 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7148 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7149 "movntsd\t{$src, $dst|$dst, $src}",
7150 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7153 //===----------------------------------------------------------------------===//
7155 //===----------------------------------------------------------------------===//
7157 //===----------------------------------------------------------------------===//
7158 // VBROADCAST - Load from memory and broadcast to all elements of the
7159 // destination operand
7161 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7162 X86MemOperand x86memop, Intrinsic Int> :
7163 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7164 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7165 [(set RC:$dst, (Int addr:$src))]>, VEX;
7167 // AVX2 adds register forms
7168 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7170 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7171 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7172 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7174 let ExeDomain = SSEPackedSingle in {
7175 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7176 int_x86_avx_vbroadcast_ss>;
7177 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7178 int_x86_avx_vbroadcast_ss_256>;
7180 let ExeDomain = SSEPackedDouble in
7181 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7182 int_x86_avx_vbroadcast_sd_256>;
7183 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7184 int_x86_avx_vbroadcastf128_pd_256>;
7186 let ExeDomain = SSEPackedSingle in {
7187 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7188 int_x86_avx2_vbroadcast_ss_ps>;
7189 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7190 int_x86_avx2_vbroadcast_ss_ps_256>;
7192 let ExeDomain = SSEPackedDouble in
7193 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7194 int_x86_avx2_vbroadcast_sd_pd_256>;
7196 let Predicates = [HasAVX2] in
7197 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7198 int_x86_avx2_vbroadcasti128>;
7200 let Predicates = [HasAVX] in
7201 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7202 (VBROADCASTF128 addr:$src)>;
7205 //===----------------------------------------------------------------------===//
7206 // VINSERTF128 - Insert packed floating-point values
7208 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7209 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7210 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7211 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7214 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7215 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7216 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7220 let Predicates = [HasAVX] in {
7221 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7223 (VINSERTF128rr VR256:$src1, VR128:$src2,
7224 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7225 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7227 (VINSERTF128rr VR256:$src1, VR128:$src2,
7228 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7230 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7232 (VINSERTF128rm VR256:$src1, addr:$src2,
7233 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7234 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7236 (VINSERTF128rm VR256:$src1, addr:$src2,
7237 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7240 let Predicates = [HasAVX1Only] in {
7241 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7243 (VINSERTF128rr VR256:$src1, VR128:$src2,
7244 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7245 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7247 (VINSERTF128rr VR256:$src1, VR128:$src2,
7248 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7249 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7251 (VINSERTF128rr VR256:$src1, VR128:$src2,
7252 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7253 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7255 (VINSERTF128rr VR256:$src1, VR128:$src2,
7256 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7258 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7260 (VINSERTF128rm VR256:$src1, addr:$src2,
7261 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7264 //===----------------------------------------------------------------------===//
7265 // VEXTRACTF128 - Extract packed floating-point values
7267 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7268 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7269 (ins VR256:$src1, i8imm:$src2),
7270 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7273 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7274 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7275 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7279 // Extract and store.
7280 let Predicates = [HasAVX] in {
7281 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1,
7282 imm:$src2), addr:$dst),
7283 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7284 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1,
7285 imm:$src2), addr:$dst),
7286 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7287 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1,
7288 imm:$src2), addr:$dst),
7289 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7293 let Predicates = [HasAVX] in {
7294 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7295 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7296 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7297 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7298 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7299 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7301 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7302 (v4f32 (VEXTRACTF128rr
7303 (v8f32 VR256:$src1),
7304 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7305 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7306 (v2f64 (VEXTRACTF128rr
7307 (v4f64 VR256:$src1),
7308 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7311 let Predicates = [HasAVX1Only] in {
7312 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7313 (v2i64 (VEXTRACTF128rr
7314 (v4i64 VR256:$src1),
7315 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7316 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7317 (v4i32 (VEXTRACTF128rr
7318 (v8i32 VR256:$src1),
7319 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7320 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7321 (v8i16 (VEXTRACTF128rr
7322 (v16i16 VR256:$src1),
7323 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7324 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7325 (v16i8 (VEXTRACTF128rr
7326 (v32i8 VR256:$src1),
7327 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7330 //===----------------------------------------------------------------------===//
7331 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7333 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7334 Intrinsic IntLd, Intrinsic IntLd256,
7335 Intrinsic IntSt, Intrinsic IntSt256> {
7336 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7337 (ins VR128:$src1, f128mem:$src2),
7338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7339 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7341 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7342 (ins VR256:$src1, f256mem:$src2),
7343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7344 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7346 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7347 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7349 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7350 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7351 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7352 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7353 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7356 let ExeDomain = SSEPackedSingle in
7357 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7358 int_x86_avx_maskload_ps,
7359 int_x86_avx_maskload_ps_256,
7360 int_x86_avx_maskstore_ps,
7361 int_x86_avx_maskstore_ps_256>;
7362 let ExeDomain = SSEPackedDouble in
7363 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7364 int_x86_avx_maskload_pd,
7365 int_x86_avx_maskload_pd_256,
7366 int_x86_avx_maskstore_pd,
7367 int_x86_avx_maskstore_pd_256>;
7369 //===----------------------------------------------------------------------===//
7370 // VPERMIL - Permute Single and Double Floating-Point Values
7372 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7373 RegisterClass RC, X86MemOperand x86memop_f,
7374 X86MemOperand x86memop_i, PatFrag i_frag,
7375 Intrinsic IntVar, ValueType vt> {
7376 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7377 (ins RC:$src1, RC:$src2),
7378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7379 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7380 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7381 (ins RC:$src1, x86memop_i:$src2),
7382 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7383 [(set RC:$dst, (IntVar RC:$src1,
7384 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7386 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7387 (ins RC:$src1, i8imm:$src2),
7388 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7389 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7390 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7391 (ins x86memop_f:$src1, i8imm:$src2),
7392 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7394 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7397 let ExeDomain = SSEPackedSingle in {
7398 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7399 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7400 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7401 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7403 let ExeDomain = SSEPackedDouble in {
7404 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7405 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7406 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7407 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7410 let Predicates = [HasAVX] in {
7411 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7412 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7413 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7414 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7415 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7417 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7418 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7419 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7421 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7422 (VPERMILPDri VR128:$src1, imm:$imm)>;
7423 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7424 (VPERMILPDmi addr:$src1, imm:$imm)>;
7427 //===----------------------------------------------------------------------===//
7428 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7430 let ExeDomain = SSEPackedSingle in {
7431 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7432 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7433 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7434 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7435 (i8 imm:$src3))))]>, VEX_4V;
7436 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7437 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7438 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7439 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7440 (i8 imm:$src3)))]>, VEX_4V;
7443 let Predicates = [HasAVX] in {
7444 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7445 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7446 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7447 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7448 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7451 let Predicates = [HasAVX1Only] in {
7452 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7453 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7454 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7455 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7456 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7457 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7458 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7459 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7461 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7462 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7463 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7464 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7465 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7466 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7467 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7468 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7469 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7470 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7471 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7472 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7475 //===----------------------------------------------------------------------===//
7476 // VZERO - Zero YMM registers
7478 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7479 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7480 // Zero All YMM registers
7481 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7482 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7484 // Zero Upper bits of YMM registers
7485 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7486 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7489 //===----------------------------------------------------------------------===//
7490 // Half precision conversion instructions
7491 //===----------------------------------------------------------------------===//
7492 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7493 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7494 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7495 [(set RC:$dst, (Int VR128:$src))]>,
7497 let neverHasSideEffects = 1, mayLoad = 1 in
7498 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7499 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7502 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7503 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7504 (ins RC:$src1, i32i8imm:$src2),
7505 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7506 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7508 let neverHasSideEffects = 1, mayStore = 1 in
7509 def mr : Ii8<0x1D, MRMDestMem, (outs),
7510 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7511 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7515 let Predicates = [HasAVX, HasF16C] in {
7516 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7517 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7518 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7519 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7522 //===----------------------------------------------------------------------===//
7523 // AVX2 Instructions
7524 //===----------------------------------------------------------------------===//
7526 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7527 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7528 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7529 X86MemOperand x86memop> {
7530 let isCommutable = 1 in
7531 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7532 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7533 !strconcat(OpcodeStr,
7534 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7535 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7537 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7538 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7539 !strconcat(OpcodeStr,
7540 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7543 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7547 let isCommutable = 0 in {
7548 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7549 VR128, memopv2i64, i128mem>;
7550 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7551 VR256, memopv4i64, i256mem>;
7554 //===----------------------------------------------------------------------===//
7555 // VPBROADCAST - Load from memory and broadcast to all elements of the
7556 // destination operand
7558 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7559 X86MemOperand x86memop, PatFrag ld_frag,
7560 Intrinsic Int128, Intrinsic Int256> {
7561 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7562 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7563 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7564 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7565 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7567 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7568 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7569 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7570 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7571 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7572 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7574 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7577 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7578 int_x86_avx2_pbroadcastb_128,
7579 int_x86_avx2_pbroadcastb_256>;
7580 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7581 int_x86_avx2_pbroadcastw_128,
7582 int_x86_avx2_pbroadcastw_256>;
7583 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7584 int_x86_avx2_pbroadcastd_128,
7585 int_x86_avx2_pbroadcastd_256>;
7586 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7587 int_x86_avx2_pbroadcastq_128,
7588 int_x86_avx2_pbroadcastq_256>;
7590 let Predicates = [HasAVX2] in {
7591 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7592 (VPBROADCASTBrm addr:$src)>;
7593 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7594 (VPBROADCASTBYrm addr:$src)>;
7595 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7596 (VPBROADCASTWrm addr:$src)>;
7597 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7598 (VPBROADCASTWYrm addr:$src)>;
7599 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7600 (VPBROADCASTDrm addr:$src)>;
7601 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7602 (VPBROADCASTDYrm addr:$src)>;
7603 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7604 (VPBROADCASTQrm addr:$src)>;
7605 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7606 (VPBROADCASTQYrm addr:$src)>;
7608 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7609 (VPBROADCASTBrr VR128:$src)>;
7610 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7611 (VPBROADCASTBYrr VR128:$src)>;
7612 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7613 (VPBROADCASTWrr VR128:$src)>;
7614 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7615 (VPBROADCASTWYrr VR128:$src)>;
7616 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7617 (VPBROADCASTDrr VR128:$src)>;
7618 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7619 (VPBROADCASTDYrr VR128:$src)>;
7620 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7621 (VPBROADCASTQrr VR128:$src)>;
7622 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7623 (VPBROADCASTQYrr VR128:$src)>;
7624 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7625 (VBROADCASTSSrr VR128:$src)>;
7626 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7627 (VBROADCASTSSYrr VR128:$src)>;
7628 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7629 (VPBROADCASTQrr VR128:$src)>;
7630 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7631 (VBROADCASTSDYrr VR128:$src)>;
7633 // Provide fallback in case the load node that is used in the patterns above
7634 // is used by additional users, which prevents the pattern selection.
7635 let AddedComplexity = 20 in {
7636 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7637 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7638 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7639 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7640 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7641 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7643 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7644 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7645 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7646 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7647 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7648 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7652 // AVX1 broadcast patterns
7653 let Predicates = [HasAVX1Only] in {
7654 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7655 (VBROADCASTSSYrm addr:$src)>;
7656 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7657 (VBROADCASTSDYrm addr:$src)>;
7658 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7659 (VBROADCASTSSrm addr:$src)>;
7662 let Predicates = [HasAVX] in {
7663 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7664 (VBROADCASTSSYrm addr:$src)>;
7665 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7666 (VBROADCASTSDYrm addr:$src)>;
7667 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7668 (VBROADCASTSSrm addr:$src)>;
7670 // Provide fallback in case the load node that is used in the patterns above
7671 // is used by additional users, which prevents the pattern selection.
7672 let AddedComplexity = 20 in {
7673 // 128bit broadcasts:
7674 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7675 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7676 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7677 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7678 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7679 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7680 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7681 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7682 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7683 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7685 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7686 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7687 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7688 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7689 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7690 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7691 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7692 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7693 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7694 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7698 //===----------------------------------------------------------------------===//
7699 // VPERM - Permute instructions
7702 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7704 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7705 (ins VR256:$src1, VR256:$src2),
7706 !strconcat(OpcodeStr,
7707 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7709 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7710 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7711 (ins VR256:$src1, i256mem:$src2),
7712 !strconcat(OpcodeStr,
7713 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7715 (OpVT (X86VPermv VR256:$src1,
7716 (bitconvert (mem_frag addr:$src2)))))]>,
7720 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7721 let ExeDomain = SSEPackedSingle in
7722 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7724 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7726 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7727 (ins VR256:$src1, i8imm:$src2),
7728 !strconcat(OpcodeStr,
7729 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7731 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7732 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7733 (ins i256mem:$src1, i8imm:$src2),
7734 !strconcat(OpcodeStr,
7735 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7737 (OpVT (X86VPermi (mem_frag addr:$src1),
7738 (i8 imm:$src2))))]>, VEX;
7741 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7742 let ExeDomain = SSEPackedDouble in
7743 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7745 //===----------------------------------------------------------------------===//
7746 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7748 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7749 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7750 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7751 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7752 (i8 imm:$src3))))]>, VEX_4V;
7753 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7754 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7755 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7756 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7757 (i8 imm:$src3)))]>, VEX_4V;
7759 let Predicates = [HasAVX2] in {
7760 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7761 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7762 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7763 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7764 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7765 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7767 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7769 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7770 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7771 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7772 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7773 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7775 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7779 //===----------------------------------------------------------------------===//
7780 // VINSERTI128 - Insert packed integer values
7782 let neverHasSideEffects = 1 in {
7783 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7784 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7785 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7788 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7789 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7790 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7794 let Predicates = [HasAVX2] in {
7795 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7797 (VINSERTI128rr VR256:$src1, VR128:$src2,
7798 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7799 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7801 (VINSERTI128rr VR256:$src1, VR128:$src2,
7802 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7803 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7805 (VINSERTI128rr VR256:$src1, VR128:$src2,
7806 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7807 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7809 (VINSERTI128rr VR256:$src1, VR128:$src2,
7810 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7812 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7814 (VINSERTI128rm VR256:$src1, addr:$src2,
7815 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7818 //===----------------------------------------------------------------------===//
7819 // VEXTRACTI128 - Extract packed integer values
7821 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7822 (ins VR256:$src1, i8imm:$src2),
7823 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7825 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7827 let neverHasSideEffects = 1, mayStore = 1 in
7828 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7829 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7830 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7832 let Predicates = [HasAVX2] in {
7833 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7834 (v2i64 (VEXTRACTI128rr
7835 (v4i64 VR256:$src1),
7836 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7837 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7838 (v4i32 (VEXTRACTI128rr
7839 (v8i32 VR256:$src1),
7840 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7841 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7842 (v8i16 (VEXTRACTI128rr
7843 (v16i16 VR256:$src1),
7844 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7845 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7846 (v16i8 (VEXTRACTI128rr
7847 (v32i8 VR256:$src1),
7848 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7851 //===----------------------------------------------------------------------===//
7852 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7854 multiclass avx2_pmovmask<string OpcodeStr,
7855 Intrinsic IntLd128, Intrinsic IntLd256,
7856 Intrinsic IntSt128, Intrinsic IntSt256> {
7857 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7858 (ins VR128:$src1, i128mem:$src2),
7859 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7860 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7861 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7862 (ins VR256:$src1, i256mem:$src2),
7863 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7864 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7865 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7866 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7867 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7868 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7869 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7870 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7871 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7872 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7875 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7876 int_x86_avx2_maskload_d,
7877 int_x86_avx2_maskload_d_256,
7878 int_x86_avx2_maskstore_d,
7879 int_x86_avx2_maskstore_d_256>;
7880 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7881 int_x86_avx2_maskload_q,
7882 int_x86_avx2_maskload_q_256,
7883 int_x86_avx2_maskstore_q,
7884 int_x86_avx2_maskstore_q_256>, VEX_W;
7887 //===----------------------------------------------------------------------===//
7888 // Variable Bit Shifts
7890 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7891 ValueType vt128, ValueType vt256> {
7892 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7893 (ins VR128:$src1, VR128:$src2),
7894 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7896 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7898 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7899 (ins VR128:$src1, i128mem:$src2),
7900 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7902 (vt128 (OpNode VR128:$src1,
7903 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7905 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7906 (ins VR256:$src1, VR256:$src2),
7907 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7909 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7911 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7912 (ins VR256:$src1, i256mem:$src2),
7913 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7915 (vt256 (OpNode VR256:$src1,
7916 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7920 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7921 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7922 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7923 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7924 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
7926 //===----------------------------------------------------------------------===//
7927 // VGATHER - GATHER Operations
7928 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
7929 X86MemOperand memop128, X86MemOperand memop256> {
7930 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
7931 (ins VR128:$src1, memop128:$src2, VR128:$mask),
7932 !strconcat(OpcodeStr,
7933 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7935 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
7936 (ins RC256:$src1, memop256:$src2, RC256:$mask),
7937 !strconcat(OpcodeStr,
7938 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7939 []>, VEX_4VOp3, VEX_L;
7942 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
7943 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
7944 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
7945 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
7946 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
7947 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
7948 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
7949 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
7950 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;