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:$src), (iPTR 0))),
248 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
249 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
250 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
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:$src)),
286 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
287 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
288 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
289 // Implicitly promote a 64-bit scalar to a vector.
290 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
291 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
292 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
293 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
295 // Bitcasts between 128-bit vector types. Return the original type since
296 // no instruction is needed for the conversion
297 let Predicates = [HasSSE2] in {
298 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
299 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
300 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
301 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
302 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
303 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
304 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
305 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
306 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
307 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
308 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
309 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
310 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
311 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
312 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
313 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
314 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
315 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
316 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
317 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
318 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
319 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
320 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
321 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
322 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
323 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
324 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
325 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
326 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
327 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
330 // Bitcasts between 256-bit vector types. Return the original type since
331 // no instruction is needed for the conversion
332 let Predicates = [HasAVX] in {
333 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
334 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
335 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
336 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
337 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
338 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
339 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
340 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
341 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
342 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
343 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
344 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
345 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
346 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
347 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
348 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
349 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
350 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
351 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
352 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
353 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
354 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
355 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
356 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
357 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
358 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
359 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
360 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
361 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
362 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
365 // Alias instructions that map fld0 to pxor for sse.
366 // This is expanded by ExpandPostRAPseudos.
367 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
369 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
370 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
371 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
372 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
375 //===----------------------------------------------------------------------===//
376 // AVX & SSE - Zero/One Vectors
377 //===----------------------------------------------------------------------===//
379 // Alias instruction that maps zero vector to pxor / xorp* for sse.
380 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
381 // swizzled by ExecutionDepsFix to pxor.
382 // We set canFoldAsLoad because this can be converted to a constant-pool
383 // load of an all-zeros value if folding it would be beneficial.
384 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
385 isPseudo = 1, neverHasSideEffects = 1 in {
386 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
389 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
390 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
391 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
392 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
393 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
394 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
397 // The same as done above but for AVX. The 256-bit ISA does not support PI,
398 // and doesn't need it because on sandy bridge the register is set to zero
399 // at the rename stage without using any execution unit, so SET0PSY
400 // and SET0PDY can be used for vector int instructions without penalty
401 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
402 // JIT implementatioan, it does not expand the instructions below like
403 // X86MCInstLower does.
404 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
405 isCodeGenOnly = 1 in {
406 let Predicates = [HasAVX] in {
407 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
408 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
409 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
410 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
412 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
413 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
417 let Predicates = [HasAVX2], AddedComplexity = 5 in {
418 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
419 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
420 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
421 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
424 // AVX has no support for 256-bit integer instructions, but since the 128-bit
425 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
426 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
427 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
428 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
430 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
431 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
432 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
434 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
435 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
436 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
438 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
439 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
440 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
442 // We set canFoldAsLoad because this can be converted to a constant-pool
443 // load of an all-ones value if folding it would be beneficial.
444 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
445 // JIT implementation, it does not expand the instructions below like
446 // X86MCInstLower does.
447 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
448 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
449 let Predicates = [HasAVX] in
450 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
451 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
452 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
453 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
454 let Predicates = [HasAVX2] in
455 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
456 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
460 //===----------------------------------------------------------------------===//
461 // SSE 1 & 2 - Move FP Scalar Instructions
463 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
464 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
465 // is used instead. Register-to-register movss/movsd is not modeled as an
466 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
467 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
468 //===----------------------------------------------------------------------===//
470 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
471 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
472 [(set VR128:$dst, (vt (OpNode VR128:$src1,
473 (scalar_to_vector RC:$src2))))],
476 // Loading from memory automatically zeroing upper bits.
477 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
478 PatFrag mem_pat, string OpcodeStr> :
479 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
481 [(set RC:$dst, (mem_pat addr:$src))],
485 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
486 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
488 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
489 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
492 // For the disassembler
493 let isCodeGenOnly = 1 in {
494 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
495 (ins VR128:$src1, FR32:$src2),
496 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
499 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
500 (ins VR128:$src1, FR64:$src2),
501 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
506 let canFoldAsLoad = 1, isReMaterializable = 1 in {
507 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
509 let AddedComplexity = 20 in
510 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
514 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
515 "movss\t{$src, $dst|$dst, $src}",
516 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
518 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
519 "movsd\t{$src, $dst|$dst, $src}",
520 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
524 let Constraints = "$src1 = $dst" in {
525 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
526 "movss\t{$src2, $dst|$dst, $src2}">, XS;
527 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
528 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
530 // For the disassembler
531 let isCodeGenOnly = 1 in {
532 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
533 (ins VR128:$src1, FR32:$src2),
534 "movss\t{$src2, $dst|$dst, $src2}", [],
535 IIC_SSE_MOV_S_RR>, XS;
536 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
537 (ins VR128:$src1, FR64:$src2),
538 "movsd\t{$src2, $dst|$dst, $src2}", [],
539 IIC_SSE_MOV_S_RR>, XD;
543 let canFoldAsLoad = 1, isReMaterializable = 1 in {
544 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
546 let AddedComplexity = 20 in
547 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
550 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
551 "movss\t{$src, $dst|$dst, $src}",
552 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
553 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
554 "movsd\t{$src, $dst|$dst, $src}",
555 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
558 let Predicates = [HasAVX] in {
559 let AddedComplexity = 15 in {
560 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
561 // MOVS{S,D} to the lower bits.
562 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
563 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
564 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
565 (VMOVSSrr (v4f32 (V_SET0)),
566 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
567 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
568 (VMOVSSrr (v4i32 (V_SET0)),
569 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
570 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
571 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
573 // Move low f32 and clear high bits.
574 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
575 (SUBREG_TO_REG (i32 0),
576 (VMOVSSrr (v4f32 (V_SET0)),
577 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
578 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
579 (SUBREG_TO_REG (i32 0),
580 (VMOVSSrr (v4i32 (V_SET0)),
581 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
584 let AddedComplexity = 20 in {
585 // MOVSSrm zeros the high parts of the register; represent this
586 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
587 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
588 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
589 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
590 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
591 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
592 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
594 // MOVSDrm zeros the high parts of the register; represent this
595 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
596 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
597 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
598 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
599 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
600 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
601 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
602 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
603 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
604 def : Pat<(v2f64 (X86vzload addr:$src)),
605 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
607 // Represent the same patterns above but in the form they appear for
609 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
610 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
611 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
612 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
613 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
614 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
615 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
616 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
617 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
619 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
620 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
621 (SUBREG_TO_REG (i32 0),
622 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
624 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
625 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
626 (SUBREG_TO_REG (i64 0),
627 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
629 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
630 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
631 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
633 // Move low f64 and clear high bits.
634 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
635 (SUBREG_TO_REG (i32 0),
636 (VMOVSDrr (v2f64 (V_SET0)),
637 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
639 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
640 (SUBREG_TO_REG (i32 0),
641 (VMOVSDrr (v2i64 (V_SET0)),
642 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
644 // Extract and store.
645 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
648 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
649 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
652 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
654 // Shuffle with VMOVSS
655 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
656 (VMOVSSrr (v4i32 VR128:$src1),
657 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
658 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
659 (VMOVSSrr (v4f32 VR128:$src1),
660 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
663 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
664 (SUBREG_TO_REG (i32 0),
665 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
666 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
667 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
668 (SUBREG_TO_REG (i32 0),
669 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
670 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
672 // Shuffle with VMOVSD
673 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
674 (VMOVSDrr (v2i64 VR128:$src1),
675 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
676 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
677 (VMOVSDrr (v2f64 VR128:$src1),
678 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
679 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
680 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
682 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
683 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
687 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
688 (SUBREG_TO_REG (i32 0),
689 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
690 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
691 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
692 (SUBREG_TO_REG (i32 0),
693 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
694 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
697 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
698 // is during lowering, where it's not possible to recognize the fold cause
699 // it has two uses through a bitcast. One use disappears at isel time and the
700 // fold opportunity reappears.
701 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
702 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
704 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
705 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
707 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
708 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
710 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
711 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
715 let Predicates = [HasSSE1] in {
716 let AddedComplexity = 15 in {
717 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
718 // MOVSS to the lower bits.
719 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
720 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
721 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
722 (MOVSSrr (v4f32 (V_SET0)),
723 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
724 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
725 (MOVSSrr (v4i32 (V_SET0)),
726 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
729 let AddedComplexity = 20 in {
730 // MOVSSrm zeros the high parts of the register; represent this
731 // with SUBREG_TO_REG.
732 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
733 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
734 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
735 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
736 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
737 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
740 // Extract and store.
741 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
744 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
746 // Shuffle with MOVSS
747 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
748 (MOVSSrr (v4i32 VR128:$src1),
749 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
750 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
751 (MOVSSrr (v4f32 VR128:$src1),
752 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
755 let Predicates = [HasSSE2] in {
756 let AddedComplexity = 15 in {
757 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
758 // MOVSD to the lower bits.
759 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
760 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
763 let AddedComplexity = 20 in {
764 // MOVSDrm zeros the high parts of the register; represent this
765 // with SUBREG_TO_REG.
766 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
767 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
768 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
769 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
770 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
771 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
772 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
773 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
774 def : Pat<(v2f64 (X86vzload addr:$src)),
775 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
778 // Extract and store.
779 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
782 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
784 // Shuffle with MOVSD
785 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
786 (MOVSDrr (v2i64 VR128:$src1),
787 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
788 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
789 (MOVSDrr (v2f64 VR128:$src1),
790 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
791 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
792 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
793 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
794 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
796 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
797 // is during lowering, where it's not possible to recognize the fold cause
798 // it has two uses through a bitcast. One use disappears at isel time and the
799 // fold opportunity reappears.
800 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
801 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
802 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
803 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
804 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
805 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
806 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
807 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
810 //===----------------------------------------------------------------------===//
811 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
812 //===----------------------------------------------------------------------===//
814 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
815 X86MemOperand x86memop, PatFrag ld_frag,
816 string asm, Domain d,
818 bit IsReMaterializable = 1> {
819 let neverHasSideEffects = 1 in
820 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
821 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
822 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
823 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
824 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
825 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
828 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
829 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
831 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
832 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
834 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
835 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
837 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
838 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
841 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
842 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
844 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
845 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
847 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
848 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
850 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
851 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
853 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
854 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
856 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
857 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
859 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
860 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
862 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
863 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
866 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
867 "movaps\t{$src, $dst|$dst, $src}",
868 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
869 IIC_SSE_MOVA_P_MR>, VEX;
870 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
871 "movapd\t{$src, $dst|$dst, $src}",
872 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
873 IIC_SSE_MOVA_P_MR>, VEX;
874 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
875 "movups\t{$src, $dst|$dst, $src}",
876 [(store (v4f32 VR128:$src), addr:$dst)],
877 IIC_SSE_MOVU_P_MR>, VEX;
878 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
879 "movupd\t{$src, $dst|$dst, $src}",
880 [(store (v2f64 VR128:$src), addr:$dst)],
881 IIC_SSE_MOVU_P_MR>, VEX;
882 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
883 "movaps\t{$src, $dst|$dst, $src}",
884 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
885 IIC_SSE_MOVA_P_MR>, VEX;
886 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
887 "movapd\t{$src, $dst|$dst, $src}",
888 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
889 IIC_SSE_MOVA_P_MR>, VEX;
890 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
891 "movups\t{$src, $dst|$dst, $src}",
892 [(store (v8f32 VR256:$src), addr:$dst)],
893 IIC_SSE_MOVU_P_MR>, VEX;
894 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
895 "movupd\t{$src, $dst|$dst, $src}",
896 [(store (v4f64 VR256:$src), addr:$dst)],
897 IIC_SSE_MOVU_P_MR>, VEX;
900 let isCodeGenOnly = 1 in {
901 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
903 "movaps\t{$src, $dst|$dst, $src}", [],
904 IIC_SSE_MOVA_P_RR>, VEX;
905 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
907 "movapd\t{$src, $dst|$dst, $src}", [],
908 IIC_SSE_MOVA_P_RR>, VEX;
909 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
911 "movups\t{$src, $dst|$dst, $src}", [],
912 IIC_SSE_MOVU_P_RR>, VEX;
913 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
915 "movupd\t{$src, $dst|$dst, $src}", [],
916 IIC_SSE_MOVU_P_RR>, VEX;
917 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
919 "movaps\t{$src, $dst|$dst, $src}", [],
920 IIC_SSE_MOVA_P_RR>, VEX;
921 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
923 "movapd\t{$src, $dst|$dst, $src}", [],
924 IIC_SSE_MOVA_P_RR>, VEX;
925 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
927 "movups\t{$src, $dst|$dst, $src}", [],
928 IIC_SSE_MOVU_P_RR>, VEX;
929 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
931 "movupd\t{$src, $dst|$dst, $src}", [],
932 IIC_SSE_MOVU_P_RR>, VEX;
935 let Predicates = [HasAVX] in {
936 def : Pat<(v8i32 (X86vzmovl
937 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
938 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
939 def : Pat<(v4i64 (X86vzmovl
940 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
941 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
942 def : Pat<(v8f32 (X86vzmovl
943 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
944 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
945 def : Pat<(v4f64 (X86vzmovl
946 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
947 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
951 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
952 (VMOVUPSYmr addr:$dst, VR256:$src)>;
953 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
954 (VMOVUPDYmr addr:$dst, VR256:$src)>;
956 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
957 "movaps\t{$src, $dst|$dst, $src}",
958 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
960 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
961 "movapd\t{$src, $dst|$dst, $src}",
962 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
964 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
965 "movups\t{$src, $dst|$dst, $src}",
966 [(store (v4f32 VR128:$src), addr:$dst)],
968 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
969 "movupd\t{$src, $dst|$dst, $src}",
970 [(store (v2f64 VR128:$src), addr:$dst)],
974 let isCodeGenOnly = 1 in {
975 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
976 "movaps\t{$src, $dst|$dst, $src}", [],
978 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
979 "movapd\t{$src, $dst|$dst, $src}", [],
981 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
982 "movups\t{$src, $dst|$dst, $src}", [],
984 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
985 "movupd\t{$src, $dst|$dst, $src}", [],
989 let Predicates = [HasAVX] in {
990 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
991 (VMOVUPSmr addr:$dst, VR128:$src)>;
992 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
993 (VMOVUPDmr addr:$dst, VR128:$src)>;
996 let Predicates = [HasSSE1] in
997 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
998 (MOVUPSmr addr:$dst, VR128:$src)>;
999 let Predicates = [HasSSE2] in
1000 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
1001 (MOVUPDmr addr:$dst, VR128:$src)>;
1003 // Use vmovaps/vmovups for AVX integer load/store.
1004 let Predicates = [HasAVX] in {
1005 // 128-bit load/store
1006 def : Pat<(alignedloadv2i64 addr:$src),
1007 (VMOVAPSrm addr:$src)>;
1008 def : Pat<(loadv2i64 addr:$src),
1009 (VMOVUPSrm addr:$src)>;
1011 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1012 (VMOVAPSmr addr:$dst, VR128:$src)>;
1013 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1014 (VMOVAPSmr addr:$dst, VR128:$src)>;
1015 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1016 (VMOVAPSmr addr:$dst, VR128:$src)>;
1017 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1018 (VMOVAPSmr addr:$dst, VR128:$src)>;
1019 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1020 (VMOVUPSmr addr:$dst, VR128:$src)>;
1021 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1022 (VMOVUPSmr addr:$dst, VR128:$src)>;
1023 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1024 (VMOVUPSmr addr:$dst, VR128:$src)>;
1025 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1026 (VMOVUPSmr addr:$dst, VR128:$src)>;
1028 // 256-bit load/store
1029 def : Pat<(alignedloadv4i64 addr:$src),
1030 (VMOVAPSYrm addr:$src)>;
1031 def : Pat<(loadv4i64 addr:$src),
1032 (VMOVUPSYrm addr:$src)>;
1033 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1034 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1035 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1036 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1037 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1038 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1039 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1040 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1041 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1042 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1043 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1044 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1045 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1046 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1047 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1048 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1051 // Use movaps / movups for SSE integer load / store (one byte shorter).
1052 // The instructions selected below are then converted to MOVDQA/MOVDQU
1053 // during the SSE domain pass.
1054 let Predicates = [HasSSE1] in {
1055 def : Pat<(alignedloadv2i64 addr:$src),
1056 (MOVAPSrm addr:$src)>;
1057 def : Pat<(loadv2i64 addr:$src),
1058 (MOVUPSrm addr:$src)>;
1060 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1061 (MOVAPSmr addr:$dst, VR128:$src)>;
1062 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1063 (MOVAPSmr addr:$dst, VR128:$src)>;
1064 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1065 (MOVAPSmr addr:$dst, VR128:$src)>;
1066 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1067 (MOVAPSmr addr:$dst, VR128:$src)>;
1068 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1069 (MOVUPSmr addr:$dst, VR128:$src)>;
1070 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1071 (MOVUPSmr addr:$dst, VR128:$src)>;
1072 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1073 (MOVUPSmr addr:$dst, VR128:$src)>;
1074 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1075 (MOVUPSmr addr:$dst, VR128:$src)>;
1078 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1079 // bits are disregarded. FIXME: Set encoding to pseudo!
1080 let neverHasSideEffects = 1 in {
1081 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1082 "movaps\t{$src, $dst|$dst, $src}", [],
1083 IIC_SSE_MOVA_P_RR>, VEX;
1084 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1085 "movapd\t{$src, $dst|$dst, $src}", [],
1086 IIC_SSE_MOVA_P_RR>, VEX;
1087 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1088 "movaps\t{$src, $dst|$dst, $src}", [],
1090 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1091 "movapd\t{$src, $dst|$dst, $src}", [],
1095 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1096 // bits are disregarded. FIXME: Set encoding to pseudo!
1097 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1098 let isCodeGenOnly = 1 in {
1099 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1100 "movaps\t{$src, $dst|$dst, $src}",
1101 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1102 IIC_SSE_MOVA_P_RM>, VEX;
1103 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1104 "movapd\t{$src, $dst|$dst, $src}",
1105 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1106 IIC_SSE_MOVA_P_RM>, VEX;
1108 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1109 "movaps\t{$src, $dst|$dst, $src}",
1110 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1112 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1113 "movapd\t{$src, $dst|$dst, $src}",
1114 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1118 //===----------------------------------------------------------------------===//
1119 // SSE 1 & 2 - Move Low packed FP Instructions
1120 //===----------------------------------------------------------------------===//
1122 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1123 SDNode psnode, SDNode pdnode, string base_opc,
1124 string asm_opr, InstrItinClass itin> {
1125 def PSrm : PI<opc, MRMSrcMem,
1126 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1127 !strconcat(base_opc, "s", asm_opr),
1130 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1131 itin, SSEPackedSingle>, TB;
1133 def PDrm : PI<opc, MRMSrcMem,
1134 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1135 !strconcat(base_opc, "d", asm_opr),
1136 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1137 (scalar_to_vector (loadf64 addr:$src2)))))],
1138 itin, SSEPackedDouble>, TB, OpSize;
1141 let AddedComplexity = 20 in {
1142 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1143 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1144 IIC_SSE_MOV_LH>, VEX_4V;
1146 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1147 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1148 "\t{$src2, $dst|$dst, $src2}",
1152 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1153 "movlps\t{$src, $dst|$dst, $src}",
1154 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1155 (iPTR 0))), addr:$dst)],
1156 IIC_SSE_MOV_LH>, VEX;
1157 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1158 "movlpd\t{$src, $dst|$dst, $src}",
1159 [(store (f64 (vector_extract (v2f64 VR128:$src),
1160 (iPTR 0))), addr:$dst)],
1161 IIC_SSE_MOV_LH>, VEX;
1162 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1163 "movlps\t{$src, $dst|$dst, $src}",
1164 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1165 (iPTR 0))), addr:$dst)],
1167 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1168 "movlpd\t{$src, $dst|$dst, $src}",
1169 [(store (f64 (vector_extract (v2f64 VR128:$src),
1170 (iPTR 0))), addr:$dst)],
1173 let Predicates = [HasAVX] in {
1174 // Shuffle with VMOVLPS
1175 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1176 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1177 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1178 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1180 // Shuffle with VMOVLPD
1181 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1182 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1183 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1184 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1187 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1189 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1190 def : Pat<(store (v4i32 (X86Movlps
1191 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1192 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1193 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1195 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1196 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1198 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1201 let Predicates = [HasSSE1] in {
1202 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1203 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1204 (iPTR 0))), addr:$src1),
1205 (MOVLPSmr addr:$src1, VR128:$src2)>;
1207 // Shuffle with MOVLPS
1208 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1209 (MOVLPSrm VR128:$src1, addr:$src2)>;
1210 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1211 (MOVLPSrm VR128:$src1, addr:$src2)>;
1212 def : Pat<(X86Movlps VR128:$src1,
1213 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1214 (MOVLPSrm VR128:$src1, addr:$src2)>;
1217 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1219 (MOVLPSmr addr:$src1, VR128:$src2)>;
1220 def : Pat<(store (v4i32 (X86Movlps
1221 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1223 (MOVLPSmr addr:$src1, VR128:$src2)>;
1226 let Predicates = [HasSSE2] in {
1227 // Shuffle with MOVLPD
1228 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1229 (MOVLPDrm VR128:$src1, addr:$src2)>;
1230 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1231 (MOVLPDrm VR128:$src1, addr:$src2)>;
1234 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1236 (MOVLPDmr addr:$src1, VR128:$src2)>;
1237 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1239 (MOVLPDmr addr:$src1, VR128:$src2)>;
1242 //===----------------------------------------------------------------------===//
1243 // SSE 1 & 2 - Move Hi packed FP Instructions
1244 //===----------------------------------------------------------------------===//
1246 let AddedComplexity = 20 in {
1247 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1248 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1249 IIC_SSE_MOV_LH>, VEX_4V;
1251 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1252 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1253 "\t{$src2, $dst|$dst, $src2}",
1257 // v2f64 extract element 1 is always custom lowered to unpack high to low
1258 // and extract element 0 so the non-store version isn't too horrible.
1259 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1260 "movhps\t{$src, $dst|$dst, $src}",
1261 [(store (f64 (vector_extract
1262 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1263 (bc_v2f64 (v4f32 VR128:$src))),
1264 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1265 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1266 "movhpd\t{$src, $dst|$dst, $src}",
1267 [(store (f64 (vector_extract
1268 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1269 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1270 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1271 "movhps\t{$src, $dst|$dst, $src}",
1272 [(store (f64 (vector_extract
1273 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1274 (bc_v2f64 (v4f32 VR128:$src))),
1275 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1276 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1277 "movhpd\t{$src, $dst|$dst, $src}",
1278 [(store (f64 (vector_extract
1279 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1280 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1282 let Predicates = [HasAVX] in {
1284 def : Pat<(X86Movlhps VR128:$src1,
1285 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1286 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1287 def : Pat<(X86Movlhps VR128:$src1,
1288 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1289 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1291 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1292 // is during lowering, where it's not possible to recognize the load fold
1293 // cause it has two uses through a bitcast. One use disappears at isel time
1294 // and the fold opportunity reappears.
1295 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1296 (scalar_to_vector (loadf64 addr:$src2)))),
1297 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1300 let Predicates = [HasSSE1] in {
1302 def : Pat<(X86Movlhps VR128:$src1,
1303 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1304 (MOVHPSrm VR128:$src1, addr:$src2)>;
1305 def : Pat<(X86Movlhps VR128:$src1,
1306 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1307 (MOVHPSrm VR128:$src1, addr:$src2)>;
1310 let Predicates = [HasSSE2] in {
1311 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1312 // is during lowering, where it's not possible to recognize the load fold
1313 // cause it has two uses through a bitcast. One use disappears at isel time
1314 // and the fold opportunity reappears.
1315 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1316 (scalar_to_vector (loadf64 addr:$src2)))),
1317 (MOVHPDrm VR128:$src1, addr:$src2)>;
1320 //===----------------------------------------------------------------------===//
1321 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1322 //===----------------------------------------------------------------------===//
1324 let AddedComplexity = 20 in {
1325 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1326 (ins VR128:$src1, VR128:$src2),
1327 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1329 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1332 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1333 (ins VR128:$src1, VR128:$src2),
1334 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1336 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1340 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1341 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1342 (ins VR128:$src1, VR128:$src2),
1343 "movlhps\t{$src2, $dst|$dst, $src2}",
1345 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1347 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1348 (ins VR128:$src1, VR128:$src2),
1349 "movhlps\t{$src2, $dst|$dst, $src2}",
1351 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1355 let Predicates = [HasAVX] in {
1357 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1358 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1359 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1360 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1363 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1364 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1367 let Predicates = [HasSSE1] in {
1369 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1370 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1371 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1372 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1375 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1376 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1379 //===----------------------------------------------------------------------===//
1380 // SSE 1 & 2 - Conversion Instructions
1381 //===----------------------------------------------------------------------===//
1383 def SSE_CVT_PD : OpndItins<
1384 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1387 def SSE_CVT_PS : OpndItins<
1388 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1391 def SSE_CVT_Scalar : OpndItins<
1392 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1395 def SSE_CVT_SS2SI_32 : OpndItins<
1396 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1399 def SSE_CVT_SS2SI_64 : OpndItins<
1400 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1403 def SSE_CVT_SD2SI : OpndItins<
1404 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1407 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1408 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1409 string asm, OpndItins itins> {
1410 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1411 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1413 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1414 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1418 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1419 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1420 string asm, Domain d, OpndItins itins> {
1421 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1422 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1424 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1425 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1429 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 X86MemOperand x86memop, string asm> {
1431 let neverHasSideEffects = 1 in {
1432 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1433 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1435 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1436 (ins DstRC:$src1, x86memop:$src),
1437 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1438 } // neverHasSideEffects = 1
1441 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1442 "cvttss2si\t{$src, $dst|$dst, $src}",
1445 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1446 "cvttss2si\t{$src, $dst|$dst, $src}",
1448 XS, VEX, VEX_W, VEX_LIG;
1449 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1450 "cvttsd2si\t{$src, $dst|$dst, $src}",
1453 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1454 "cvttsd2si\t{$src, $dst|$dst, $src}",
1456 XD, VEX, VEX_W, VEX_LIG;
1458 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1459 // register, but the same isn't true when only using memory operands,
1460 // provide other assembly "l" and "q" forms to address this explicitly
1461 // where appropriate to do so.
1462 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1463 XS, VEX_4V, VEX_LIG;
1464 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1465 XS, VEX_4V, VEX_W, VEX_LIG;
1466 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1467 XD, VEX_4V, VEX_LIG;
1468 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1469 XD, VEX_4V, VEX_LIG;
1470 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1471 XD, VEX_4V, VEX_W, VEX_LIG;
1473 let Predicates = [HasAVX], AddedComplexity = 1 in {
1474 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1475 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1476 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1477 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1478 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1479 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1480 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1481 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1483 def : Pat<(f32 (sint_to_fp GR32:$src)),
1484 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1485 def : Pat<(f32 (sint_to_fp GR64:$src)),
1486 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1487 def : Pat<(f64 (sint_to_fp GR32:$src)),
1488 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1489 def : Pat<(f64 (sint_to_fp GR64:$src)),
1490 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1493 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1494 "cvttss2si\t{$src, $dst|$dst, $src}",
1495 SSE_CVT_SS2SI_32>, XS;
1496 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1497 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1498 SSE_CVT_SS2SI_64>, XS, REX_W;
1499 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1500 "cvttsd2si\t{$src, $dst|$dst, $src}",
1502 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1503 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1504 SSE_CVT_SD2SI>, XD, REX_W;
1505 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1506 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1507 SSE_CVT_Scalar>, XS;
1508 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1509 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1510 SSE_CVT_Scalar>, XS, REX_W;
1511 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1512 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1513 SSE_CVT_Scalar>, XD;
1514 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1515 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1516 SSE_CVT_Scalar>, XD, REX_W;
1518 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1519 // and/or XMM operand(s).
1521 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1522 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1523 string asm, OpndItins itins> {
1524 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1525 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1526 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1527 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1528 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1529 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm>;
1532 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1533 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1534 PatFrag ld_frag, string asm, OpndItins itins,
1536 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1538 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1539 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1540 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1542 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1543 (ins DstRC:$src1, x86memop:$src2),
1545 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1546 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1547 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1551 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1552 f128mem, load, "cvtsd2si", SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1553 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1554 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si",
1555 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1557 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1558 f128mem, load, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1559 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1560 f128mem, load, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1563 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1564 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1565 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1566 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1567 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss",
1568 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1570 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1571 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1572 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1573 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1574 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd",
1575 SSE_CVT_Scalar, 0>, XD,
1578 let Constraints = "$src1 = $dst" in {
1579 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1580 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1581 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1582 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1583 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1584 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1585 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1586 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1587 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1588 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1589 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1590 "cvtsi2sd", SSE_CVT_Scalar>, XD, REX_W;
1595 // Aliases for intrinsics
1596 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1597 f32mem, load, "cvttss2si",
1598 SSE_CVT_SS2SI_32>, XS, VEX;
1599 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1600 int_x86_sse_cvttss2si64, f32mem, load,
1601 "cvttss2si", SSE_CVT_SS2SI_64>,
1603 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1604 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1606 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1607 int_x86_sse2_cvttsd2si64, f128mem, load,
1608 "cvttsd2si", SSE_CVT_SD2SI>,
1610 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1611 f32mem, load, "cvttss2si",
1612 SSE_CVT_SS2SI_32>, XS;
1613 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1614 int_x86_sse_cvttss2si64, f32mem, load,
1615 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1617 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1618 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1620 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1621 int_x86_sse2_cvttsd2si64, f128mem, load,
1622 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1625 let Pattern = []<dag> in {
1626 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1627 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1628 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1629 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1630 "cvtss2si\t{$src, $dst|$dst, $src}",
1631 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1632 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1633 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1634 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1636 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1637 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1638 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1642 let Pattern = []<dag> in {
1643 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1644 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1645 SSE_CVT_SS2SI_32>, XS;
1646 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1647 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1648 SSE_CVT_SS2SI_64>, XS, REX_W;
1649 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1650 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1651 SSEPackedSingle, SSE_CVT_PS>, TB,
1652 Requires<[HasSSE2]>;
1655 let Predicates = [HasAVX] in {
1656 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1657 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1658 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1659 (VCVTSS2SIrm addr:$src)>;
1660 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1661 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1662 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1663 (VCVTSS2SI64rm addr:$src)>;
1666 let Predicates = [HasSSE1] in {
1667 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1668 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1669 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1670 (CVTSS2SIrm addr:$src)>;
1671 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1672 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1673 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1674 (CVTSS2SI64rm addr:$src)>;
1679 // Convert scalar double to scalar single
1680 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1681 (ins FR64:$src1, FR64:$src2),
1682 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1683 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1685 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1686 (ins FR64:$src1, f64mem:$src2),
1687 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1688 [], IIC_SSE_CVT_Scalar_RM>,
1689 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1691 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1694 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1695 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1696 [(set FR32:$dst, (fround FR64:$src))],
1697 IIC_SSE_CVT_Scalar_RR>;
1698 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1699 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1700 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1701 IIC_SSE_CVT_Scalar_RM>,
1703 Requires<[HasSSE2, OptForSize]>;
1705 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1706 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1709 let Constraints = "$src1 = $dst" in
1710 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1711 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1712 SSE_CVT_Scalar>, XS;
1714 // Convert scalar single to scalar double
1715 // SSE2 instructions with XS prefix
1716 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1717 (ins FR32:$src1, FR32:$src2),
1718 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1719 [], IIC_SSE_CVT_Scalar_RR>,
1720 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1722 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1723 (ins FR32:$src1, f32mem:$src2),
1724 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1725 [], IIC_SSE_CVT_Scalar_RM>,
1726 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1728 let Predicates = [HasAVX] in {
1729 def : Pat<(f64 (fextend FR32:$src)),
1730 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1731 def : Pat<(fextend (loadf32 addr:$src)),
1732 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1733 def : Pat<(extloadf32 addr:$src),
1734 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1737 def : Pat<(extloadf32 addr:$src),
1738 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1739 Requires<[HasAVX, OptForSpeed]>;
1741 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1742 "cvtss2sd\t{$src, $dst|$dst, $src}",
1743 [(set FR64:$dst, (fextend FR32:$src))],
1744 IIC_SSE_CVT_Scalar_RR>, XS,
1745 Requires<[HasSSE2]>;
1746 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1747 "cvtss2sd\t{$src, $dst|$dst, $src}",
1748 [(set FR64:$dst, (extloadf32 addr:$src))],
1749 IIC_SSE_CVT_Scalar_RM>, XS,
1750 Requires<[HasSSE2, OptForSize]>;
1752 // extload f32 -> f64. This matches load+fextend because we have a hack in
1753 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1755 // Since these loads aren't folded into the fextend, we have to match it
1757 def : Pat<(fextend (loadf32 addr:$src)),
1758 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1759 def : Pat<(extloadf32 addr:$src),
1760 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1762 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1763 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1764 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1767 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V,
1769 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1770 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1771 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1772 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1773 (load addr:$src2)))],
1774 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V,
1776 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1777 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1778 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1779 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1780 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1782 IIC_SSE_CVT_Scalar_RR>, XS,
1783 Requires<[HasSSE2]>;
1784 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1785 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1786 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1787 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1788 (load addr:$src2)))],
1789 IIC_SSE_CVT_Scalar_RM>, XS,
1790 Requires<[HasSSE2]>;
1793 // Convert doubleword to packed single/double fp
1794 // SSE2 instructions without OpSize prefix
1795 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1796 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1797 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))],
1799 TB, VEX, Requires<[HasAVX]>;
1800 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1801 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1802 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1803 (bitconvert (memopv2i64 addr:$src))))],
1805 TB, VEX, Requires<[HasAVX]>;
1806 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1807 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1808 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))],
1810 TB, Requires<[HasSSE2]>;
1811 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1812 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1813 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1814 (bitconvert (memopv2i64 addr:$src))))],
1816 TB, Requires<[HasSSE2]>;
1819 // Convert packed single/double fp to doubleword
1820 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1821 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1822 IIC_SSE_CVT_PS_RR>, VEX;
1823 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1824 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1825 IIC_SSE_CVT_PS_RM>, VEX;
1826 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1827 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1828 IIC_SSE_CVT_PS_RR>, VEX;
1829 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1830 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1831 IIC_SSE_CVT_PS_RM>, VEX;
1832 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1833 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1835 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1836 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1839 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1840 "cvtps2dq\t{$src, $dst|$dst, $src}",
1841 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1844 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1846 "cvtps2dq\t{$src, $dst|$dst, $src}",
1847 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1848 (memop addr:$src)))],
1849 IIC_SSE_CVT_PS_RM>, VEX;
1850 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1851 "cvtps2dq\t{$src, $dst|$dst, $src}",
1852 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1854 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1855 "cvtps2dq\t{$src, $dst|$dst, $src}",
1856 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1857 (memop addr:$src)))],
1860 // Convert Packed Double FP to Packed DW Integers
1861 let Predicates = [HasAVX] in {
1862 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1863 // register, but the same isn't true when using memory operands instead.
1864 // Provide other assembly rr and rm forms to address this explicitly.
1865 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1866 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1867 def VCVTPD2DQXrYr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1868 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1871 def VCVTPD2DQXrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1872 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1873 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1874 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1877 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1878 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX;
1879 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1880 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1883 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1884 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1886 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1887 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1890 // SSE2 packed instructions with XD prefix
1891 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1892 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1893 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1895 XD, VEX, Requires<[HasAVX]>;
1896 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1897 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1898 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1899 (memop addr:$src)))],
1901 XD, VEX, Requires<[HasAVX]>;
1902 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1903 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1904 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1906 XD, Requires<[HasSSE2]>;
1907 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1908 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1909 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1910 (memop addr:$src)))],
1912 XD, Requires<[HasSSE2]>;
1915 // Convert with truncation packed single/double fp to doubleword
1916 // SSE2 packed instructions with XS prefix
1917 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1918 "cvttps2dq\t{$src, $dst|$dst, $src}",
1920 (int_x86_sse2_cvttps2dq VR128:$src))],
1921 IIC_SSE_CVT_PS_RR>, VEX;
1922 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1923 "cvttps2dq\t{$src, $dst|$dst, $src}",
1924 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1925 (memop addr:$src)))],
1926 IIC_SSE_CVT_PS_RM>, VEX;
1927 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1928 "cvttps2dq\t{$src, $dst|$dst, $src}",
1930 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1931 IIC_SSE_CVT_PS_RR>, VEX;
1932 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1933 "cvttps2dq\t{$src, $dst|$dst, $src}",
1934 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1935 (memopv8f32 addr:$src)))],
1936 IIC_SSE_CVT_PS_RM>, VEX;
1938 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1939 "cvttps2dq\t{$src, $dst|$dst, $src}",
1941 (int_x86_sse2_cvttps2dq VR128:$src))],
1943 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1944 "cvttps2dq\t{$src, $dst|$dst, $src}",
1946 (int_x86_sse2_cvttps2dq (memop addr:$src)))],
1949 let Predicates = [HasAVX] in {
1950 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1951 (Int_VCVTDQ2PSrr VR128:$src)>;
1952 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1953 (Int_VCVTDQ2PSrm addr:$src)>;
1955 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1956 (VCVTTPS2DQrr VR128:$src)>;
1957 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1958 (VCVTTPS2DQrm addr:$src)>;
1960 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1961 (VCVTDQ2PSYrr VR256:$src)>;
1962 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1963 (VCVTDQ2PSYrm addr:$src)>;
1965 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1966 (VCVTTPS2DQYrr VR256:$src)>;
1967 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1968 (VCVTTPS2DQYrm addr:$src)>;
1971 let Predicates = [HasSSE2] in {
1972 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1973 (Int_CVTDQ2PSrr VR128:$src)>;
1974 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1975 (Int_CVTDQ2PSrm addr:$src)>;
1977 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1978 (CVTTPS2DQrr VR128:$src)>;
1979 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1980 (CVTTPS2DQrm addr:$src)>;
1983 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1984 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1986 (int_x86_sse2_cvttpd2dq VR128:$src))],
1987 IIC_SSE_CVT_PD_RR>, VEX;
1988 let isCodeGenOnly = 1 in
1989 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1990 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1991 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1992 (memop addr:$src)))],
1993 IIC_SSE_CVT_PD_RM>, VEX;
1994 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1995 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1996 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1998 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1999 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2000 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2001 (memop addr:$src)))],
2004 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2005 // register, but the same isn't true when using memory operands instead.
2006 // Provide other assembly rr and rm forms to address this explicitly.
2007 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2008 "cvttpd2dq\t{$src, $dst|$dst, $src}", [],
2009 IIC_SSE_CVT_PD_RR>, VEX;
2012 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2013 "cvttpd2dqx\t{$src, $dst|$dst, $src}", [],
2014 IIC_SSE_CVT_PD_RR>, VEX;
2015 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2016 "cvttpd2dqx\t{$src, $dst|$dst, $src}", [],
2017 IIC_SSE_CVT_PD_RM>, VEX;
2020 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2021 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
2022 IIC_SSE_CVT_PD_RR>, VEX;
2023 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2024 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
2025 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2027 let Predicates = [HasAVX] in {
2028 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2029 (VCVTTPD2DQYrr VR256:$src)>;
2030 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2031 (VCVTTPD2DQYrm addr:$src)>;
2032 } // Predicates = [HasAVX]
2034 // Convert packed single to packed double
2035 let Predicates = [HasAVX] in {
2036 // SSE2 instructions without OpSize prefix
2037 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2038 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2039 IIC_SSE_CVT_PD_RR>, TB, VEX;
2040 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2041 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2042 IIC_SSE_CVT_PD_RM>, TB, VEX;
2043 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2044 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2045 IIC_SSE_CVT_PD_RR>, TB, VEX;
2046 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2047 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2048 IIC_SSE_CVT_PD_RM>, TB, VEX;
2050 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2051 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2052 IIC_SSE_CVT_PD_RR>, TB;
2053 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2054 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2055 IIC_SSE_CVT_PD_RM>, TB;
2057 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2058 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2059 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2061 TB, VEX, Requires<[HasAVX]>;
2062 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2063 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2064 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
2065 (load addr:$src)))],
2067 TB, VEX, Requires<[HasAVX]>;
2068 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2069 "cvtps2pd\t{$src, $dst|$dst, $src}",
2070 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2072 TB, Requires<[HasSSE2]>;
2073 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2074 "cvtps2pd\t{$src, $dst|$dst, $src}",
2075 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
2076 (load addr:$src)))],
2078 TB, Requires<[HasSSE2]>;
2080 // Convert Packed DW Integers to Packed Double FP
2081 let Predicates = [HasAVX] in {
2082 def VCVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2083 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2084 def VCVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2085 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2086 def VCVTDQ2PDYrm : SSDI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2087 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2088 def VCVTDQ2PDYrr : SSDI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2089 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2092 def CVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2093 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2095 def CVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2096 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2099 // 128 bit register conversion intrinsics
2100 let Predicates = [HasAVX] in
2101 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2102 (VCVTDQ2PDrr VR128:$src)>;
2104 let Predicates = [HasSSE2] in
2105 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2106 (CVTDQ2PDrr VR128:$src)>;
2108 // AVX 256-bit register conversion intrinsics
2109 let Predicates = [HasAVX] in {
2110 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
2111 (VCVTDQ2PDYrr VR128:$src)>;
2112 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
2113 (VCVTDQ2PDYrm addr:$src)>;
2115 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
2116 (VCVTPD2DQYrr VR256:$src)>;
2117 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
2118 (VCVTPD2DQYrm addr:$src)>;
2120 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2121 (VCVTDQ2PDYrr VR128:$src)>;
2122 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2123 (VCVTDQ2PDYrm addr:$src)>;
2124 } // Predicates = [HasAVX]
2126 // Convert packed double to packed single
2127 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2128 // register, but the same isn't true when using memory operands instead.
2129 // Provide other assembly rr and rm forms to address this explicitly.
2130 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2131 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2132 IIC_SSE_CVT_PD_RR>, VEX;
2133 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2134 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2135 IIC_SSE_CVT_PD_RR>, VEX;
2138 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2139 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2140 IIC_SSE_CVT_PD_RR>, VEX;
2141 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2142 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2143 IIC_SSE_CVT_PD_RM>, VEX;
2146 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2147 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2148 IIC_SSE_CVT_PD_RR>, VEX;
2149 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2150 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2151 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2152 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2153 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2155 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2156 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2160 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2161 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2162 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2164 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2166 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2167 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2168 (memop addr:$src)))],
2170 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2171 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2172 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2174 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2175 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2176 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2177 (memop addr:$src)))],
2180 // AVX 256-bit register conversion intrinsics
2181 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2182 // whenever possible to avoid declaring two versions of each one.
2183 let Predicates = [HasAVX] in {
2184 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2185 (VCVTDQ2PSYrr VR256:$src)>;
2186 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2187 (VCVTDQ2PSYrm addr:$src)>;
2189 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2190 (VCVTPD2PSYrr VR256:$src)>;
2191 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2192 (VCVTPD2PSYrm addr:$src)>;
2194 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2195 (VCVTPS2DQYrr VR256:$src)>;
2196 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2197 (VCVTPS2DQYrm addr:$src)>;
2199 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2200 (VCVTPS2PDYrr VR128:$src)>;
2201 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2202 (VCVTPS2PDYrm addr:$src)>;
2204 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2205 (VCVTTPD2DQYrr VR256:$src)>;
2206 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2207 (VCVTTPD2DQYrm addr:$src)>;
2209 // Match fround and fextend for 128/256-bit conversions
2210 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2211 (VCVTPD2PSYrr VR256:$src)>;
2212 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2213 (VCVTPD2PSYrm addr:$src)>;
2215 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2216 (VCVTPS2PDYrr VR128:$src)>;
2217 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2218 (VCVTPS2PDYrm addr:$src)>;
2221 //===----------------------------------------------------------------------===//
2222 // SSE 1 & 2 - Compare Instructions
2223 //===----------------------------------------------------------------------===//
2225 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2226 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2227 Operand CC, SDNode OpNode, ValueType VT,
2228 PatFrag ld_frag, string asm, string asm_alt,
2230 def rr : SIi8<0xC2, MRMSrcReg,
2231 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2232 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2234 def rm : SIi8<0xC2, MRMSrcMem,
2235 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2236 [(set RC:$dst, (OpNode (VT RC:$src1),
2237 (ld_frag addr:$src2), imm:$cc))],
2240 // Accept explicit immediate argument form instead of comparison code.
2241 let neverHasSideEffects = 1 in {
2242 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2243 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2244 IIC_SSE_ALU_F32S_RR>;
2246 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2247 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2248 IIC_SSE_ALU_F32S_RM>;
2252 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2253 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2254 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2256 XS, VEX_4V, VEX_LIG;
2257 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2258 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2259 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2260 SSE_ALU_F32S>, // same latency as 32 bit compare
2261 XD, VEX_4V, VEX_LIG;
2263 let Constraints = "$src1 = $dst" in {
2264 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2265 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2266 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2268 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2269 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2270 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2271 SSE_ALU_F32S>, // same latency as 32 bit compare
2275 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2276 Intrinsic Int, string asm, OpndItins itins> {
2277 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2278 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2279 [(set VR128:$dst, (Int VR128:$src1,
2280 VR128:$src, imm:$cc))],
2282 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2283 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2284 [(set VR128:$dst, (Int VR128:$src1,
2285 (load addr:$src), imm:$cc))],
2289 // Aliases to match intrinsics which expect XMM operand(s).
2290 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2291 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2294 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2295 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2296 SSE_ALU_F32S>, // same latency as f32
2298 let Constraints = "$src1 = $dst" in {
2299 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2300 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2302 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2303 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2304 SSE_ALU_F32S>, // same latency as f32
2309 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2310 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2311 ValueType vt, X86MemOperand x86memop,
2312 PatFrag ld_frag, string OpcodeStr, Domain d> {
2313 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2314 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2315 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2316 IIC_SSE_COMIS_RR, d>;
2317 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2318 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2319 [(set EFLAGS, (OpNode (vt RC:$src1),
2320 (ld_frag addr:$src2)))],
2321 IIC_SSE_COMIS_RM, d>;
2324 let Defs = [EFLAGS] in {
2325 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2326 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2327 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2328 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2330 let Pattern = []<dag> in {
2331 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2332 "comiss", SSEPackedSingle>, TB, VEX,
2334 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2335 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2339 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2340 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2341 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2342 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2344 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2345 load, "comiss", SSEPackedSingle>, TB, VEX;
2346 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2347 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2348 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2349 "ucomiss", SSEPackedSingle>, TB;
2350 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2351 "ucomisd", SSEPackedDouble>, TB, OpSize;
2353 let Pattern = []<dag> in {
2354 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2355 "comiss", SSEPackedSingle>, TB;
2356 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2357 "comisd", SSEPackedDouble>, TB, OpSize;
2360 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2361 load, "ucomiss", SSEPackedSingle>, TB;
2362 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2363 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2365 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2366 "comiss", SSEPackedSingle>, TB;
2367 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2368 "comisd", SSEPackedDouble>, TB, OpSize;
2369 } // Defs = [EFLAGS]
2371 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2372 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2373 Operand CC, Intrinsic Int, string asm,
2374 string asm_alt, Domain d> {
2375 def rri : PIi8<0xC2, MRMSrcReg,
2376 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2377 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2378 IIC_SSE_CMPP_RR, d>;
2379 def rmi : PIi8<0xC2, MRMSrcMem,
2380 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2381 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2382 IIC_SSE_CMPP_RM, d>;
2384 // Accept explicit immediate argument form instead of comparison code.
2385 let neverHasSideEffects = 1 in {
2386 def rri_alt : PIi8<0xC2, MRMSrcReg,
2387 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2388 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2389 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2390 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2391 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2395 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2396 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2397 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2398 SSEPackedSingle>, TB, VEX_4V;
2399 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2400 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2401 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2402 SSEPackedDouble>, TB, OpSize, VEX_4V;
2403 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2404 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2405 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2406 SSEPackedSingle>, TB, VEX_4V;
2407 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2408 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2409 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2410 SSEPackedDouble>, TB, OpSize, VEX_4V;
2411 let Constraints = "$src1 = $dst" in {
2412 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2413 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2414 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2415 SSEPackedSingle>, TB;
2416 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2417 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2418 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2419 SSEPackedDouble>, TB, OpSize;
2422 let Predicates = [HasAVX] in {
2423 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2424 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2425 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2426 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2427 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2428 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2429 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2430 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2432 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2433 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2434 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2435 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2436 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2437 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2438 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2439 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2442 let Predicates = [HasSSE1] in {
2443 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2444 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2445 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2446 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2449 let Predicates = [HasSSE2] in {
2450 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2451 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2452 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2453 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2456 //===----------------------------------------------------------------------===//
2457 // SSE 1 & 2 - Shuffle Instructions
2458 //===----------------------------------------------------------------------===//
2460 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2461 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2462 ValueType vt, string asm, PatFrag mem_frag,
2463 Domain d, bit IsConvertibleToThreeAddress = 0> {
2464 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2465 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2466 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2467 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2468 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2469 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2470 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2471 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2472 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2475 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2476 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2477 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2478 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2479 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2480 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2481 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2482 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2483 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2484 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2485 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2486 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2488 let Constraints = "$src1 = $dst" in {
2489 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2490 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2491 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2493 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2494 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2495 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2499 let Predicates = [HasAVX] in {
2500 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2501 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2502 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2503 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2504 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2506 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2507 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2508 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2509 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2510 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2513 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2514 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2515 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2516 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2517 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2519 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2520 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2521 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2522 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2523 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2526 let Predicates = [HasSSE1] in {
2527 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2528 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2529 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2530 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2531 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2534 let Predicates = [HasSSE2] in {
2535 // Generic SHUFPD patterns
2536 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2537 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2538 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2539 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2540 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2543 //===----------------------------------------------------------------------===//
2544 // SSE 1 & 2 - Unpack Instructions
2545 //===----------------------------------------------------------------------===//
2547 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2548 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2549 PatFrag mem_frag, RegisterClass RC,
2550 X86MemOperand x86memop, string asm,
2552 def rr : PI<opc, MRMSrcReg,
2553 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2555 (vt (OpNode RC:$src1, RC:$src2)))],
2557 def rm : PI<opc, MRMSrcMem,
2558 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2560 (vt (OpNode RC:$src1,
2561 (mem_frag addr:$src2))))],
2565 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2566 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2567 SSEPackedSingle>, TB, VEX_4V;
2568 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2569 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2570 SSEPackedDouble>, TB, OpSize, VEX_4V;
2571 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2572 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2573 SSEPackedSingle>, TB, VEX_4V;
2574 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2575 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2576 SSEPackedDouble>, TB, OpSize, VEX_4V;
2578 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2579 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2580 SSEPackedSingle>, TB, VEX_4V;
2581 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2582 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2583 SSEPackedDouble>, TB, OpSize, VEX_4V;
2584 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2585 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2586 SSEPackedSingle>, TB, VEX_4V;
2587 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2588 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2589 SSEPackedDouble>, TB, OpSize, VEX_4V;
2591 let Constraints = "$src1 = $dst" in {
2592 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2593 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2594 SSEPackedSingle>, TB;
2595 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2596 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2597 SSEPackedDouble>, TB, OpSize;
2598 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2599 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2600 SSEPackedSingle>, TB;
2601 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2602 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2603 SSEPackedDouble>, TB, OpSize;
2604 } // Constraints = "$src1 = $dst"
2606 let Predicates = [HasAVX], AddedComplexity = 1 in {
2607 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2608 // problem is during lowering, where it's not possible to recognize the load
2609 // fold cause it has two uses through a bitcast. One use disappears at isel
2610 // time and the fold opportunity reappears.
2611 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2612 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2615 let Predicates = [HasSSE2] in {
2616 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2617 // problem is during lowering, where it's not possible to recognize the load
2618 // fold cause it has two uses through a bitcast. One use disappears at isel
2619 // time and the fold opportunity reappears.
2620 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2621 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2624 //===----------------------------------------------------------------------===//
2625 // SSE 1 & 2 - Extract Floating-Point Sign mask
2626 //===----------------------------------------------------------------------===//
2628 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2629 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2631 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2632 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2633 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2634 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2635 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2636 IIC_SSE_MOVMSK, d>, REX_W;
2639 let Predicates = [HasAVX] in {
2640 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2641 "movmskps", SSEPackedSingle>, TB, VEX;
2642 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2643 "movmskpd", SSEPackedDouble>, TB,
2645 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2646 "movmskps", SSEPackedSingle>, TB, VEX;
2647 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2648 "movmskpd", SSEPackedDouble>, TB,
2651 def : Pat<(i32 (X86fgetsign FR32:$src)),
2652 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2654 def : Pat<(i64 (X86fgetsign FR32:$src)),
2655 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2657 def : Pat<(i32 (X86fgetsign FR64:$src)),
2658 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2660 def : Pat<(i64 (X86fgetsign FR64:$src)),
2661 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2665 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2666 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2667 SSEPackedSingle>, TB, VEX;
2668 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2669 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2670 SSEPackedDouble>, TB,
2672 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2673 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2674 SSEPackedSingle>, TB, VEX;
2675 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2676 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2677 SSEPackedDouble>, TB,
2681 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2682 SSEPackedSingle>, TB;
2683 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2684 SSEPackedDouble>, TB, OpSize;
2686 def : Pat<(i32 (X86fgetsign FR32:$src)),
2687 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2688 sub_ss))>, Requires<[HasSSE1]>;
2689 def : Pat<(i64 (X86fgetsign FR32:$src)),
2690 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2691 sub_ss))>, Requires<[HasSSE1]>;
2692 def : Pat<(i32 (X86fgetsign FR64:$src)),
2693 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2694 sub_sd))>, Requires<[HasSSE2]>;
2695 def : Pat<(i64 (X86fgetsign FR64:$src)),
2696 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2697 sub_sd))>, Requires<[HasSSE2]>;
2699 //===---------------------------------------------------------------------===//
2700 // SSE2 - Packed Integer Logical Instructions
2701 //===---------------------------------------------------------------------===//
2703 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2705 /// PDI_binop_rm - Simple SSE2 binary operator.
2706 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2707 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2708 X86MemOperand x86memop,
2710 bit IsCommutable = 0,
2712 let isCommutable = IsCommutable in
2713 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2714 (ins RC:$src1, RC:$src2),
2716 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2717 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2718 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2719 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2720 (ins RC:$src1, x86memop:$src2),
2722 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2723 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2724 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2725 (bitconvert (memop_frag addr:$src2)))))],
2728 } // ExeDomain = SSEPackedInt
2730 // These are ordered here for pattern ordering requirements with the fp versions
2732 let Predicates = [HasAVX] in {
2733 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2734 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2735 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2736 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2737 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2738 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2739 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2740 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2743 let Constraints = "$src1 = $dst" in {
2744 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2745 i128mem, SSE_BIT_ITINS_P, 1>;
2746 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2747 i128mem, SSE_BIT_ITINS_P, 1>;
2748 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2749 i128mem, SSE_BIT_ITINS_P, 1>;
2750 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2751 i128mem, SSE_BIT_ITINS_P, 0>;
2752 } // Constraints = "$src1 = $dst"
2754 let Predicates = [HasAVX2] in {
2755 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2756 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2757 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2758 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2759 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2760 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2761 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2762 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2765 //===----------------------------------------------------------------------===//
2766 // SSE 1 & 2 - Logical Instructions
2767 //===----------------------------------------------------------------------===//
2769 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2771 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2772 SDNode OpNode, OpndItins itins> {
2773 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2774 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2777 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2778 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2781 let Constraints = "$src1 = $dst" in {
2782 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2783 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2786 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2787 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2792 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2793 let mayLoad = 0 in {
2794 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2796 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2798 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2802 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2803 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2806 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2808 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2810 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2811 // are all promoted to v2i64, and the patterns are covered by the int
2812 // version. This is needed in SSE only, because v2i64 isn't supported on
2813 // SSE1, but only on SSE2.
2814 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2815 !strconcat(OpcodeStr, "ps"), f128mem, [],
2816 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2817 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2819 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2820 !strconcat(OpcodeStr, "pd"), f128mem,
2821 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2822 (bc_v2i64 (v2f64 VR128:$src2))))],
2823 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2824 (memopv2i64 addr:$src2)))], 0>,
2826 let Constraints = "$src1 = $dst" in {
2827 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2828 !strconcat(OpcodeStr, "ps"), f128mem,
2829 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2830 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2831 (memopv2i64 addr:$src2)))]>, TB;
2833 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2834 !strconcat(OpcodeStr, "pd"), f128mem,
2835 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2836 (bc_v2i64 (v2f64 VR128:$src2))))],
2837 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2838 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2842 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2844 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2846 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2847 !strconcat(OpcodeStr, "ps"), f256mem,
2848 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2849 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2850 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2852 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2853 !strconcat(OpcodeStr, "pd"), f256mem,
2854 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2855 (bc_v4i64 (v4f64 VR256:$src2))))],
2856 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2857 (memopv4i64 addr:$src2)))], 0>,
2861 // AVX 256-bit packed logical ops forms
2862 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2863 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2864 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2865 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2867 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2868 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2869 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2870 let isCommutable = 0 in
2871 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2873 //===----------------------------------------------------------------------===//
2874 // SSE 1 & 2 - Arithmetic Instructions
2875 //===----------------------------------------------------------------------===//
2877 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2880 /// In addition, we also have a special variant of the scalar form here to
2881 /// represent the associated intrinsic operation. This form is unlike the
2882 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2883 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2885 /// These three forms can each be reg+reg or reg+mem.
2888 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2890 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2893 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2894 OpNode, FR32, f32mem,
2895 itins.s, Is2Addr>, XS;
2896 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2897 OpNode, FR64, f64mem,
2898 itins.d, Is2Addr>, XD;
2901 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2904 let mayLoad = 0 in {
2905 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2906 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2908 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2909 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2914 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2917 let mayLoad = 0 in {
2918 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2919 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2921 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2922 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2927 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2930 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2931 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2932 itins.s, Is2Addr>, XS;
2933 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2934 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2935 itins.d, Is2Addr>, XD;
2938 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2941 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2942 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2943 SSEPackedSingle, itins.s, Is2Addr>,
2946 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2947 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2948 SSEPackedDouble, itins.d, Is2Addr>,
2952 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2954 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2955 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2956 SSEPackedSingle, itins.s, 0>, TB;
2958 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2959 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2960 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2963 // Binary Arithmetic instructions
2964 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2965 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2967 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2968 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2970 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2971 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2973 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2974 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2977 let isCommutable = 0 in {
2978 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2979 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2981 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2982 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2983 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2984 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2986 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2987 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2989 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2990 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2992 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2993 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2994 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2995 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2997 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2998 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
3000 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
3001 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
3002 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
3003 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3007 let Constraints = "$src1 = $dst" in {
3008 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
3009 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
3010 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
3011 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
3012 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
3013 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
3015 let isCommutable = 0 in {
3016 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3017 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3018 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3019 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3020 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3021 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3022 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3023 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3024 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
3025 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
3026 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3027 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3028 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
3029 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
3034 /// In addition, we also have a special variant of the scalar form here to
3035 /// represent the associated intrinsic operation. This form is unlike the
3036 /// plain scalar form, in that it takes an entire vector (instead of a
3037 /// scalar) and leaves the top elements undefined.
3039 /// And, we have a special variant form for a full-vector intrinsic form.
3041 def SSE_SQRTP : OpndItins<
3042 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
3045 def SSE_SQRTS : OpndItins<
3046 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
3049 def SSE_RCPP : OpndItins<
3050 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3053 def SSE_RCPS : OpndItins<
3054 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3057 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3058 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3059 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3060 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3061 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3062 [(set FR32:$dst, (OpNode FR32:$src))]>;
3063 // For scalar unary operations, fold a load into the operation
3064 // only in OptForSize mode. It eliminates an instruction, but it also
3065 // eliminates a whole-register clobber (the load), so it introduces a
3066 // partial register update condition.
3067 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3068 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3069 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3070 Requires<[HasSSE1, OptForSize]>;
3071 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3072 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3073 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
3074 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3075 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3076 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
3079 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3080 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3081 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3082 !strconcat(OpcodeStr,
3083 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3085 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3086 !strconcat(OpcodeStr,
3087 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3088 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3089 (ins VR128:$src1, ssmem:$src2),
3090 !strconcat(OpcodeStr,
3091 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3094 /// sse1_fp_unop_p - SSE1 unops in packed form.
3095 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3097 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3098 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3099 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3100 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3101 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3102 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3105 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3106 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3108 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3109 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3110 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3112 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3113 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3114 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3118 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3119 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3120 Intrinsic V4F32Int, OpndItins itins> {
3121 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3122 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3123 [(set VR128:$dst, (V4F32Int VR128:$src))],
3125 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3126 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3127 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3131 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3132 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3133 Intrinsic V4F32Int, OpndItins itins> {
3134 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3135 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3136 [(set VR256:$dst, (V4F32Int VR256:$src))],
3138 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3139 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3140 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3144 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3145 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3146 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3147 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3148 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3149 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3150 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3151 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3152 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3153 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3154 Requires<[HasSSE2, OptForSize]>;
3155 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3156 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3157 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3158 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3159 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3160 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3163 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3164 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3165 let neverHasSideEffects = 1 in {
3166 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3167 !strconcat(OpcodeStr,
3168 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3170 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3171 !strconcat(OpcodeStr,
3172 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3174 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3175 (ins VR128:$src1, sdmem:$src2),
3176 !strconcat(OpcodeStr,
3177 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3180 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3181 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3182 SDNode OpNode, OpndItins itins> {
3183 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3184 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3185 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3186 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3187 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3188 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3191 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3192 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3194 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3195 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3196 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3198 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3199 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3200 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3204 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3205 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3206 Intrinsic V2F64Int, OpndItins itins> {
3207 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3208 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3209 [(set VR128:$dst, (V2F64Int VR128:$src))],
3211 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3212 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3213 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3217 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3218 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3219 Intrinsic V2F64Int, OpndItins itins> {
3220 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3221 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3222 [(set VR256:$dst, (V2F64Int VR256:$src))],
3224 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3225 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3226 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3230 let Predicates = [HasAVX] in {
3232 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3233 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3235 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3236 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3237 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3238 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3239 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3241 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3243 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3245 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3249 // Reciprocal approximations. Note that these typically require refinement
3250 // in order to obtain suitable precision.
3251 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3252 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3253 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3254 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3256 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3259 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3260 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3261 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3262 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3264 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3268 let AddedComplexity = 1 in {
3269 def : Pat<(f32 (fsqrt FR32:$src)),
3270 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3271 def : Pat<(f32 (fsqrt (load addr:$src))),
3272 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3273 Requires<[HasAVX, OptForSize]>;
3274 def : Pat<(f64 (fsqrt FR64:$src)),
3275 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3276 def : Pat<(f64 (fsqrt (load addr:$src))),
3277 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3278 Requires<[HasAVX, OptForSize]>;
3280 def : Pat<(f32 (X86frsqrt FR32:$src)),
3281 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3282 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3283 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3284 Requires<[HasAVX, OptForSize]>;
3286 def : Pat<(f32 (X86frcp FR32:$src)),
3287 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3288 def : Pat<(f32 (X86frcp (load addr:$src))),
3289 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3290 Requires<[HasAVX, OptForSize]>;
3293 let Predicates = [HasAVX], AddedComplexity = 1 in {
3294 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3295 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3296 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3297 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3299 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3300 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3302 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3303 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3304 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3305 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3307 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3308 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3310 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3311 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3312 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3313 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3315 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3316 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3318 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3319 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3320 (VRCPSSr (f32 (IMPLICIT_DEF)),
3321 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3323 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3324 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3328 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3330 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3331 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3332 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3334 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3335 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3337 // Reciprocal approximations. Note that these typically require refinement
3338 // in order to obtain suitable precision.
3339 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3341 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3342 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3344 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3346 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3347 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3349 // There is no f64 version of the reciprocal approximation instructions.
3351 //===----------------------------------------------------------------------===//
3352 // SSE 1 & 2 - Non-temporal stores
3353 //===----------------------------------------------------------------------===//
3355 let AddedComplexity = 400 in { // Prefer non-temporal versions
3356 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3357 (ins f128mem:$dst, VR128:$src),
3358 "movntps\t{$src, $dst|$dst, $src}",
3359 [(alignednontemporalstore (v4f32 VR128:$src),
3361 IIC_SSE_MOVNT>, VEX;
3362 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3363 (ins f128mem:$dst, VR128:$src),
3364 "movntpd\t{$src, $dst|$dst, $src}",
3365 [(alignednontemporalstore (v2f64 VR128:$src),
3367 IIC_SSE_MOVNT>, VEX;
3369 let ExeDomain = SSEPackedInt in
3370 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3371 (ins f128mem:$dst, VR128:$src),
3372 "movntdq\t{$src, $dst|$dst, $src}",
3373 [(alignednontemporalstore (v2i64 VR128:$src),
3375 IIC_SSE_MOVNT>, VEX;
3377 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3378 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3380 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3381 (ins f256mem:$dst, VR256:$src),
3382 "movntps\t{$src, $dst|$dst, $src}",
3383 [(alignednontemporalstore (v8f32 VR256:$src),
3385 IIC_SSE_MOVNT>, VEX;
3386 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3387 (ins f256mem:$dst, VR256:$src),
3388 "movntpd\t{$src, $dst|$dst, $src}",
3389 [(alignednontemporalstore (v4f64 VR256:$src),
3391 IIC_SSE_MOVNT>, VEX;
3392 let ExeDomain = SSEPackedInt in
3393 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3394 (ins f256mem:$dst, VR256:$src),
3395 "movntdq\t{$src, $dst|$dst, $src}",
3396 [(alignednontemporalstore (v4i64 VR256:$src),
3398 IIC_SSE_MOVNT>, VEX;
3401 let AddedComplexity = 400 in { // Prefer non-temporal versions
3402 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3403 "movntps\t{$src, $dst|$dst, $src}",
3404 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3406 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3407 "movntpd\t{$src, $dst|$dst, $src}",
3408 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3411 let ExeDomain = SSEPackedInt in
3412 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3413 "movntdq\t{$src, $dst|$dst, $src}",
3414 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3417 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3418 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3420 // There is no AVX form for instructions below this point
3421 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3422 "movnti{l}\t{$src, $dst|$dst, $src}",
3423 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3425 TB, Requires<[HasSSE2]>;
3426 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3427 "movnti{q}\t{$src, $dst|$dst, $src}",
3428 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3430 TB, Requires<[HasSSE2]>;
3433 //===----------------------------------------------------------------------===//
3434 // SSE 1 & 2 - Prefetch and memory fence
3435 //===----------------------------------------------------------------------===//
3437 // Prefetch intrinsic.
3438 let Predicates = [HasSSE1] in {
3439 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3440 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3441 IIC_SSE_PREFETCH>, TB;
3442 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3443 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3444 IIC_SSE_PREFETCH>, TB;
3445 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3446 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3447 IIC_SSE_PREFETCH>, TB;
3448 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3449 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3450 IIC_SSE_PREFETCH>, TB;
3454 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3455 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3456 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3458 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3459 // was introduced with SSE2, it's backward compatible.
3460 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3462 // Load, store, and memory fence
3463 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3464 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3465 TB, Requires<[HasSSE1]>;
3466 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3467 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3468 TB, Requires<[HasSSE2]>;
3469 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3470 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3471 TB, Requires<[HasSSE2]>;
3473 def : Pat<(X86SFence), (SFENCE)>;
3474 def : Pat<(X86LFence), (LFENCE)>;
3475 def : Pat<(X86MFence), (MFENCE)>;
3477 //===----------------------------------------------------------------------===//
3478 // SSE 1 & 2 - Load/Store XCSR register
3479 //===----------------------------------------------------------------------===//
3481 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3482 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3483 IIC_SSE_LDMXCSR>, VEX;
3484 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3485 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3486 IIC_SSE_STMXCSR>, VEX;
3488 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3489 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3491 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3492 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3495 //===---------------------------------------------------------------------===//
3496 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3497 //===---------------------------------------------------------------------===//
3499 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3501 let neverHasSideEffects = 1 in {
3502 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3503 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3505 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3506 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3509 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3510 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3512 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3513 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3517 let isCodeGenOnly = 1 in {
3518 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3519 "movdqa\t{$src, $dst|$dst, $src}", [],
3522 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3523 "movdqa\t{$src, $dst|$dst, $src}", [],
3526 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3527 "movdqu\t{$src, $dst|$dst, $src}", [],
3530 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3531 "movdqu\t{$src, $dst|$dst, $src}", [],
3536 let canFoldAsLoad = 1, mayLoad = 1 in {
3537 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3538 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3540 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3541 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3543 let Predicates = [HasAVX] in {
3544 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3545 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3547 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3548 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3553 let mayStore = 1 in {
3554 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3555 (ins i128mem:$dst, VR128:$src),
3556 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3558 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3559 (ins i256mem:$dst, VR256:$src),
3560 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3562 let Predicates = [HasAVX] in {
3563 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3564 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3566 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3567 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3572 let neverHasSideEffects = 1 in
3573 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3574 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3576 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3577 "movdqu\t{$src, $dst|$dst, $src}",
3578 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3581 let isCodeGenOnly = 1 in {
3582 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3583 "movdqa\t{$src, $dst|$dst, $src}", [],
3586 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3587 "movdqu\t{$src, $dst|$dst, $src}",
3588 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3591 let canFoldAsLoad = 1, mayLoad = 1 in {
3592 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3593 "movdqa\t{$src, $dst|$dst, $src}",
3594 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3596 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3597 "movdqu\t{$src, $dst|$dst, $src}",
3598 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3600 XS, Requires<[HasSSE2]>;
3603 let mayStore = 1 in {
3604 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3605 "movdqa\t{$src, $dst|$dst, $src}",
3606 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3608 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3609 "movdqu\t{$src, $dst|$dst, $src}",
3610 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3612 XS, Requires<[HasSSE2]>;
3615 // Intrinsic forms of MOVDQU load and store
3616 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3617 "vmovdqu\t{$src, $dst|$dst, $src}",
3618 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3620 XS, VEX, Requires<[HasAVX]>;
3622 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3623 "movdqu\t{$src, $dst|$dst, $src}",
3624 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3626 XS, Requires<[HasSSE2]>;
3628 } // ExeDomain = SSEPackedInt
3630 let Predicates = [HasAVX] in {
3631 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3632 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3635 //===---------------------------------------------------------------------===//
3636 // SSE2 - Packed Integer Arithmetic Instructions
3637 //===---------------------------------------------------------------------===//
3639 def SSE_PMADD : OpndItins<
3640 IIC_SSE_PMADD, IIC_SSE_PMADD
3643 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3645 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3646 RegisterClass RC, PatFrag memop_frag,
3647 X86MemOperand x86memop,
3649 bit IsCommutable = 0,
3651 let isCommutable = IsCommutable in
3652 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3653 (ins RC:$src1, RC:$src2),
3655 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3656 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3657 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3658 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3659 (ins RC:$src1, x86memop:$src2),
3661 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3662 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3663 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3667 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3668 string OpcodeStr, SDNode OpNode,
3669 SDNode OpNode2, RegisterClass RC,
3670 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3671 ShiftOpndItins itins,
3673 // src2 is always 128-bit
3674 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3675 (ins RC:$src1, VR128:$src2),
3677 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3678 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3679 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3681 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3682 (ins RC:$src1, i128mem:$src2),
3684 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3685 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3686 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3687 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3688 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3689 (ins RC:$src1, i32i8imm:$src2),
3691 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3692 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3693 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3696 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3697 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3698 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3699 PatFrag memop_frag, X86MemOperand x86memop,
3701 bit IsCommutable = 0, bit Is2Addr = 1> {
3702 let isCommutable = IsCommutable in
3703 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3704 (ins RC:$src1, RC:$src2),
3706 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3707 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3708 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3709 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3710 (ins RC:$src1, x86memop:$src2),
3712 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3713 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3714 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3715 (bitconvert (memop_frag addr:$src2)))))]>;
3717 } // ExeDomain = SSEPackedInt
3719 // 128-bit Integer Arithmetic
3721 let Predicates = [HasAVX] in {
3722 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3723 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3725 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3726 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3727 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3728 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3729 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3730 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3731 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3732 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3733 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3734 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3735 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3736 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3737 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3738 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3739 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3740 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3741 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3742 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3746 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3747 VR128, memopv2i64, i128mem,
3748 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3749 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3750 VR128, memopv2i64, i128mem,
3751 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3752 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3753 VR128, memopv2i64, i128mem,
3754 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3755 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3756 VR128, memopv2i64, i128mem,
3757 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3758 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3759 VR128, memopv2i64, i128mem,
3760 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3761 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3762 VR128, memopv2i64, i128mem,
3763 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3764 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3765 VR128, memopv2i64, i128mem,
3766 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3767 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3768 VR128, memopv2i64, i128mem,
3769 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3770 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3771 VR128, memopv2i64, i128mem,
3772 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3773 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3774 VR128, memopv2i64, i128mem,
3775 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3776 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3777 VR128, memopv2i64, i128mem,
3778 SSE_PMADD, 1, 0>, VEX_4V;
3779 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3780 VR128, memopv2i64, i128mem,
3781 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3782 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3783 VR128, memopv2i64, i128mem,
3784 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3785 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3786 VR128, memopv2i64, i128mem,
3787 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3788 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3789 VR128, memopv2i64, i128mem,
3790 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3791 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3792 VR128, memopv2i64, i128mem,
3793 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3794 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3795 VR128, memopv2i64, i128mem,
3796 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3797 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3798 VR128, memopv2i64, i128mem,
3799 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3802 let Predicates = [HasAVX2] in {
3803 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3804 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3805 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3806 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3807 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3808 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3809 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3810 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3811 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3812 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3813 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3814 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3815 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3816 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3817 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3818 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3819 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3820 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3821 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3822 VR256, memopv4i64, i256mem,
3823 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3826 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3827 VR256, memopv4i64, i256mem,
3828 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3829 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3830 VR256, memopv4i64, i256mem,
3831 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3832 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3833 VR256, memopv4i64, i256mem,
3834 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3835 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3836 VR256, memopv4i64, i256mem,
3837 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3838 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3839 VR256, memopv4i64, i256mem,
3840 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3841 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3842 VR256, memopv4i64, i256mem,
3843 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3844 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3845 VR256, memopv4i64, i256mem,
3846 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3847 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3848 VR256, memopv4i64, i256mem,
3849 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3850 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3851 VR256, memopv4i64, i256mem,
3852 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3853 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3854 VR256, memopv4i64, i256mem,
3855 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3856 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3857 VR256, memopv4i64, i256mem,
3858 SSE_PMADD, 1, 0>, VEX_4V;
3859 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3860 VR256, memopv4i64, i256mem,
3861 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3862 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3863 VR256, memopv4i64, i256mem,
3864 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3865 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3866 VR256, memopv4i64, i256mem,
3867 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3868 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3869 VR256, memopv4i64, i256mem,
3870 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3871 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3872 VR256, memopv4i64, i256mem,
3873 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3874 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3875 VR256, memopv4i64, i256mem,
3876 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3877 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3878 VR256, memopv4i64, i256mem,
3879 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3882 let Constraints = "$src1 = $dst" in {
3883 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3884 i128mem, SSE_INTALU_ITINS_P, 1>;
3885 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3886 i128mem, SSE_INTALU_ITINS_P, 1>;
3887 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3888 i128mem, SSE_INTALU_ITINS_P, 1>;
3889 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3890 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3891 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3892 i128mem, SSE_INTMUL_ITINS_P, 1>;
3893 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3894 i128mem, SSE_INTALU_ITINS_P>;
3895 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3896 i128mem, SSE_INTALU_ITINS_P>;
3897 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3898 i128mem, SSE_INTALU_ITINS_P>;
3899 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3900 i128mem, SSE_INTALUQ_ITINS_P>;
3901 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3902 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3905 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3906 VR128, memopv2i64, i128mem,
3907 SSE_INTALU_ITINS_P>;
3908 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3909 VR128, memopv2i64, i128mem,
3910 SSE_INTALU_ITINS_P>;
3911 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3912 VR128, memopv2i64, i128mem,
3913 SSE_INTALU_ITINS_P>;
3914 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3915 VR128, memopv2i64, i128mem,
3916 SSE_INTALU_ITINS_P>;
3917 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3918 VR128, memopv2i64, i128mem,
3919 SSE_INTALU_ITINS_P, 1>;
3920 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3921 VR128, memopv2i64, i128mem,
3922 SSE_INTALU_ITINS_P, 1>;
3923 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3924 VR128, memopv2i64, i128mem,
3925 SSE_INTALU_ITINS_P, 1>;
3926 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3927 VR128, memopv2i64, i128mem,
3928 SSE_INTALU_ITINS_P, 1>;
3929 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3930 VR128, memopv2i64, i128mem,
3931 SSE_INTMUL_ITINS_P, 1>;
3932 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3933 VR128, memopv2i64, i128mem,
3934 SSE_INTMUL_ITINS_P, 1>;
3935 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3936 VR128, memopv2i64, i128mem,
3938 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3939 VR128, memopv2i64, i128mem,
3940 SSE_INTALU_ITINS_P, 1>;
3941 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3942 VR128, memopv2i64, i128mem,
3943 SSE_INTALU_ITINS_P, 1>;
3944 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3945 VR128, memopv2i64, i128mem,
3946 SSE_INTALU_ITINS_P, 1>;
3947 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3948 VR128, memopv2i64, i128mem,
3949 SSE_INTALU_ITINS_P, 1>;
3950 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3951 VR128, memopv2i64, i128mem,
3952 SSE_INTALU_ITINS_P, 1>;
3953 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3954 VR128, memopv2i64, i128mem,
3955 SSE_INTALU_ITINS_P, 1>;
3956 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3957 VR128, memopv2i64, i128mem,
3958 SSE_INTALU_ITINS_P, 1>;
3960 } // Constraints = "$src1 = $dst"
3962 //===---------------------------------------------------------------------===//
3963 // SSE2 - Packed Integer Logical Instructions
3964 //===---------------------------------------------------------------------===//
3966 let Predicates = [HasAVX] in {
3967 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3968 VR128, v8i16, v8i16, bc_v8i16,
3969 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3970 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3971 VR128, v4i32, v4i32, bc_v4i32,
3972 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3973 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3974 VR128, v2i64, v2i64, bc_v2i64,
3975 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3977 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3978 VR128, v8i16, v8i16, bc_v8i16,
3979 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3980 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3981 VR128, v4i32, v4i32, bc_v4i32,
3982 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3983 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3984 VR128, v2i64, v2i64, bc_v2i64,
3985 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3987 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3988 VR128, v8i16, v8i16, bc_v8i16,
3989 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3990 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3991 VR128, v4i32, v4i32, bc_v4i32,
3992 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3994 let ExeDomain = SSEPackedInt in {
3995 // 128-bit logical shifts.
3996 def VPSLLDQri : PDIi8<0x73, MRM7r,
3997 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3998 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4000 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
4002 def VPSRLDQri : PDIi8<0x73, MRM3r,
4003 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4004 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4006 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
4008 // PSRADQri doesn't exist in SSE[1-3].
4010 } // Predicates = [HasAVX]
4012 let Predicates = [HasAVX2] in {
4013 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
4014 VR256, v16i16, v8i16, bc_v8i16,
4015 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4016 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
4017 VR256, v8i32, v4i32, bc_v4i32,
4018 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4019 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
4020 VR256, v4i64, v2i64, bc_v2i64,
4021 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4023 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
4024 VR256, v16i16, v8i16, bc_v8i16,
4025 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4026 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
4027 VR256, v8i32, v4i32, bc_v4i32,
4028 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4029 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
4030 VR256, v4i64, v2i64, bc_v2i64,
4031 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4033 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4034 VR256, v16i16, v8i16, bc_v8i16,
4035 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4036 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4037 VR256, v8i32, v4i32, bc_v4i32,
4038 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4040 let ExeDomain = SSEPackedInt in {
4041 // 256-bit logical shifts.
4042 def VPSLLDQYri : PDIi8<0x73, MRM7r,
4043 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4044 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4046 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
4048 def VPSRLDQYri : PDIi8<0x73, MRM3r,
4049 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4050 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4052 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
4054 // PSRADQYri doesn't exist in SSE[1-3].
4056 } // Predicates = [HasAVX2]
4058 let Constraints = "$src1 = $dst" in {
4059 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
4060 VR128, v8i16, v8i16, bc_v8i16,
4061 SSE_INTSHIFT_ITINS_P>;
4062 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
4063 VR128, v4i32, v4i32, bc_v4i32,
4064 SSE_INTSHIFT_ITINS_P>;
4065 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
4066 VR128, v2i64, v2i64, bc_v2i64,
4067 SSE_INTSHIFT_ITINS_P>;
4069 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
4070 VR128, v8i16, v8i16, bc_v8i16,
4071 SSE_INTSHIFT_ITINS_P>;
4072 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
4073 VR128, v4i32, v4i32, bc_v4i32,
4074 SSE_INTSHIFT_ITINS_P>;
4075 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
4076 VR128, v2i64, v2i64, bc_v2i64,
4077 SSE_INTSHIFT_ITINS_P>;
4079 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
4080 VR128, v8i16, v8i16, bc_v8i16,
4081 SSE_INTSHIFT_ITINS_P>;
4082 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
4083 VR128, v4i32, v4i32, bc_v4i32,
4084 SSE_INTSHIFT_ITINS_P>;
4086 let ExeDomain = SSEPackedInt in {
4087 // 128-bit logical shifts.
4088 def PSLLDQri : PDIi8<0x73, MRM7r,
4089 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4090 "pslldq\t{$src2, $dst|$dst, $src2}",
4092 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
4093 def PSRLDQri : PDIi8<0x73, MRM3r,
4094 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4095 "psrldq\t{$src2, $dst|$dst, $src2}",
4097 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
4098 // PSRADQri doesn't exist in SSE[1-3].
4100 } // Constraints = "$src1 = $dst"
4102 let Predicates = [HasAVX] in {
4103 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4104 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4105 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4106 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4107 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4108 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4110 // Shift up / down and insert zero's.
4111 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4112 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4113 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4114 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4117 let Predicates = [HasAVX2] in {
4118 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4119 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4120 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4121 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4124 let Predicates = [HasSSE2] in {
4125 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4126 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4127 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4128 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4129 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4130 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4132 // Shift up / down and insert zero's.
4133 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4134 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4135 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4136 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4139 //===---------------------------------------------------------------------===//
4140 // SSE2 - Packed Integer Comparison Instructions
4141 //===---------------------------------------------------------------------===//
4143 let Predicates = [HasAVX] in {
4144 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4145 VR128, memopv2i64, i128mem,
4146 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4147 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4148 VR128, memopv2i64, i128mem,
4149 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4150 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4151 VR128, memopv2i64, i128mem,
4152 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4153 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4154 VR128, memopv2i64, i128mem,
4155 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4156 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4157 VR128, memopv2i64, i128mem,
4158 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4159 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4160 VR128, memopv2i64, i128mem,
4161 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4164 let Predicates = [HasAVX2] in {
4165 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4166 VR256, memopv4i64, i256mem,
4167 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4168 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4169 VR256, memopv4i64, i256mem,
4170 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4171 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4172 VR256, memopv4i64, i256mem,
4173 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4174 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4175 VR256, memopv4i64, i256mem,
4176 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4177 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4178 VR256, memopv4i64, i256mem,
4179 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4180 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4181 VR256, memopv4i64, i256mem,
4182 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4185 let Constraints = "$src1 = $dst" in {
4186 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4187 VR128, memopv2i64, i128mem,
4188 SSE_INTALU_ITINS_P, 1>;
4189 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4190 VR128, memopv2i64, i128mem,
4191 SSE_INTALU_ITINS_P, 1>;
4192 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4193 VR128, memopv2i64, i128mem,
4194 SSE_INTALU_ITINS_P, 1>;
4195 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4196 VR128, memopv2i64, i128mem,
4197 SSE_INTALU_ITINS_P>;
4198 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4199 VR128, memopv2i64, i128mem,
4200 SSE_INTALU_ITINS_P>;
4201 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4202 VR128, memopv2i64, i128mem,
4203 SSE_INTALU_ITINS_P>;
4204 } // Constraints = "$src1 = $dst"
4206 //===---------------------------------------------------------------------===//
4207 // SSE2 - Packed Integer Pack Instructions
4208 //===---------------------------------------------------------------------===//
4210 let Predicates = [HasAVX] in {
4211 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4212 VR128, memopv2i64, i128mem,
4213 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4214 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4215 VR128, memopv2i64, i128mem,
4216 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4217 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4218 VR128, memopv2i64, i128mem,
4219 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4222 let Predicates = [HasAVX2] in {
4223 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4224 VR256, memopv4i64, i256mem,
4225 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4226 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4227 VR256, memopv4i64, i256mem,
4228 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4229 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4230 VR256, memopv4i64, i256mem,
4231 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4234 let Constraints = "$src1 = $dst" in {
4235 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4236 VR128, memopv2i64, i128mem,
4237 SSE_INTALU_ITINS_P>;
4238 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4239 VR128, memopv2i64, i128mem,
4240 SSE_INTALU_ITINS_P>;
4241 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4242 VR128, memopv2i64, i128mem,
4243 SSE_INTALU_ITINS_P>;
4244 } // Constraints = "$src1 = $dst"
4246 //===---------------------------------------------------------------------===//
4247 // SSE2 - Packed Integer Shuffle Instructions
4248 //===---------------------------------------------------------------------===//
4250 let ExeDomain = SSEPackedInt in {
4251 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4252 def ri : Ii8<0x70, MRMSrcReg,
4253 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4254 !strconcat(OpcodeStr,
4255 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4256 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4258 def mi : Ii8<0x70, MRMSrcMem,
4259 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4260 !strconcat(OpcodeStr,
4261 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4263 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4268 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4269 def Yri : Ii8<0x70, MRMSrcReg,
4270 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4271 !strconcat(OpcodeStr,
4272 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4273 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4274 def Ymi : Ii8<0x70, MRMSrcMem,
4275 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4276 !strconcat(OpcodeStr,
4277 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4279 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4280 (i8 imm:$src2))))]>;
4282 } // ExeDomain = SSEPackedInt
4284 let Predicates = [HasAVX] in {
4285 let AddedComplexity = 5 in
4286 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4288 // SSE2 with ImmT == Imm8 and XS prefix.
4289 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4291 // SSE2 with ImmT == Imm8 and XD prefix.
4292 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4294 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4295 (VPSHUFDmi addr:$src1, imm:$imm)>;
4296 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4297 (VPSHUFDri VR128:$src1, imm:$imm)>;
4300 let Predicates = [HasAVX2] in {
4301 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4302 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4303 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4306 let Predicates = [HasSSE2] in {
4307 let AddedComplexity = 5 in
4308 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4310 // SSE2 with ImmT == Imm8 and XS prefix.
4311 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4313 // SSE2 with ImmT == Imm8 and XD prefix.
4314 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4316 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4317 (PSHUFDmi addr:$src1, imm:$imm)>;
4318 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4319 (PSHUFDri VR128:$src1, imm:$imm)>;
4322 //===---------------------------------------------------------------------===//
4323 // SSE2 - Packed Integer Unpack Instructions
4324 //===---------------------------------------------------------------------===//
4326 let ExeDomain = SSEPackedInt in {
4327 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4328 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4329 def rr : PDI<opc, MRMSrcReg,
4330 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4332 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4333 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4334 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4336 def rm : PDI<opc, MRMSrcMem,
4337 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4339 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4340 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4341 [(set VR128:$dst, (OpNode VR128:$src1,
4342 (bc_frag (memopv2i64
4347 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4348 SDNode OpNode, PatFrag bc_frag> {
4349 def Yrr : PDI<opc, MRMSrcReg,
4350 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4351 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4352 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4353 def Yrm : PDI<opc, MRMSrcMem,
4354 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4355 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4356 [(set VR256:$dst, (OpNode VR256:$src1,
4357 (bc_frag (memopv4i64 addr:$src2))))]>;
4360 let Predicates = [HasAVX] in {
4361 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4362 bc_v16i8, 0>, VEX_4V;
4363 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4364 bc_v8i16, 0>, VEX_4V;
4365 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4366 bc_v4i32, 0>, VEX_4V;
4367 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4368 bc_v2i64, 0>, VEX_4V;
4370 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4371 bc_v16i8, 0>, VEX_4V;
4372 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4373 bc_v8i16, 0>, VEX_4V;
4374 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4375 bc_v4i32, 0>, VEX_4V;
4376 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4377 bc_v2i64, 0>, VEX_4V;
4380 let Predicates = [HasAVX2] in {
4381 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4383 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4385 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4387 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4390 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4392 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4394 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4396 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4400 let Constraints = "$src1 = $dst" in {
4401 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4403 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4405 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4407 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4410 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4412 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4414 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4416 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4419 } // ExeDomain = SSEPackedInt
4421 // Patterns for using AVX1 instructions with integer vectors
4422 // Here to give AVX2 priority
4423 let Predicates = [HasAVX] in {
4424 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4425 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4426 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4427 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4428 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4429 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4430 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4431 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4433 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4434 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4435 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4436 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4437 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4438 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4439 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4440 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4443 //===---------------------------------------------------------------------===//
4444 // SSE2 - Packed Integer Extract and Insert
4445 //===---------------------------------------------------------------------===//
4447 let ExeDomain = SSEPackedInt in {
4448 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4449 def rri : Ii8<0xC4, MRMSrcReg,
4450 (outs VR128:$dst), (ins VR128:$src1,
4451 GR32:$src2, i32i8imm:$src3),
4453 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4454 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4456 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4457 def rmi : Ii8<0xC4, MRMSrcMem,
4458 (outs VR128:$dst), (ins VR128:$src1,
4459 i16mem:$src2, i32i8imm:$src3),
4461 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4462 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4464 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4465 imm:$src3))], IIC_SSE_PINSRW>;
4469 let Predicates = [HasAVX] in
4470 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4471 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4472 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4473 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4474 imm:$src2))]>, TB, OpSize, VEX;
4475 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4476 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4477 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4478 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4479 imm:$src2))], IIC_SSE_PEXTRW>;
4482 let Predicates = [HasAVX] in {
4483 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4484 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4485 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4486 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4487 []>, TB, OpSize, VEX_4V;
4490 let Constraints = "$src1 = $dst" in
4491 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4493 } // ExeDomain = SSEPackedInt
4495 //===---------------------------------------------------------------------===//
4496 // SSE2 - Packed Mask Creation
4497 //===---------------------------------------------------------------------===//
4499 let ExeDomain = SSEPackedInt in {
4501 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4502 "pmovmskb\t{$src, $dst|$dst, $src}",
4503 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4504 IIC_SSE_MOVMSK>, VEX;
4505 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4506 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4508 let Predicates = [HasAVX2] in {
4509 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4510 "pmovmskb\t{$src, $dst|$dst, $src}",
4511 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4512 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4513 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4516 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4517 "pmovmskb\t{$src, $dst|$dst, $src}",
4518 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4521 } // ExeDomain = SSEPackedInt
4523 //===---------------------------------------------------------------------===//
4524 // SSE2 - Conditional Store
4525 //===---------------------------------------------------------------------===//
4527 let ExeDomain = SSEPackedInt in {
4530 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4531 (ins VR128:$src, VR128:$mask),
4532 "maskmovdqu\t{$mask, $src|$src, $mask}",
4533 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4534 IIC_SSE_MASKMOV>, VEX;
4536 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4537 (ins VR128:$src, VR128:$mask),
4538 "maskmovdqu\t{$mask, $src|$src, $mask}",
4539 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4540 IIC_SSE_MASKMOV>, VEX;
4543 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4544 "maskmovdqu\t{$mask, $src|$src, $mask}",
4545 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4548 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4549 "maskmovdqu\t{$mask, $src|$src, $mask}",
4550 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4553 } // ExeDomain = SSEPackedInt
4555 //===---------------------------------------------------------------------===//
4556 // SSE2 - Move Doubleword
4557 //===---------------------------------------------------------------------===//
4559 //===---------------------------------------------------------------------===//
4560 // Move Int Doubleword to Packed Double Int
4562 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4563 "movd\t{$src, $dst|$dst, $src}",
4565 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4567 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4568 "movd\t{$src, $dst|$dst, $src}",
4570 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4573 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4574 "mov{d|q}\t{$src, $dst|$dst, $src}",
4576 (v2i64 (scalar_to_vector GR64:$src)))],
4577 IIC_SSE_MOVDQ>, VEX;
4578 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4579 "mov{d|q}\t{$src, $dst|$dst, $src}",
4580 [(set FR64:$dst, (bitconvert GR64:$src))],
4581 IIC_SSE_MOVDQ>, VEX;
4583 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4584 "movd\t{$src, $dst|$dst, $src}",
4586 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4587 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4588 "movd\t{$src, $dst|$dst, $src}",
4590 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4592 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4593 "mov{d|q}\t{$src, $dst|$dst, $src}",
4595 (v2i64 (scalar_to_vector GR64:$src)))],
4597 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4598 "mov{d|q}\t{$src, $dst|$dst, $src}",
4599 [(set FR64:$dst, (bitconvert GR64:$src))],
4602 //===---------------------------------------------------------------------===//
4603 // Move Int Doubleword to Single Scalar
4605 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4606 "movd\t{$src, $dst|$dst, $src}",
4607 [(set FR32:$dst, (bitconvert GR32:$src))],
4608 IIC_SSE_MOVDQ>, VEX;
4610 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4611 "movd\t{$src, $dst|$dst, $src}",
4612 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4615 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4616 "movd\t{$src, $dst|$dst, $src}",
4617 [(set FR32:$dst, (bitconvert GR32:$src))],
4620 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4621 "movd\t{$src, $dst|$dst, $src}",
4622 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4625 //===---------------------------------------------------------------------===//
4626 // Move Packed Doubleword Int to Packed Double Int
4628 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4629 "movd\t{$src, $dst|$dst, $src}",
4630 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4631 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4632 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4633 (ins i32mem:$dst, VR128:$src),
4634 "movd\t{$src, $dst|$dst, $src}",
4635 [(store (i32 (vector_extract (v4i32 VR128:$src),
4636 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4638 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4639 "movd\t{$src, $dst|$dst, $src}",
4640 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4641 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4642 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4643 "movd\t{$src, $dst|$dst, $src}",
4644 [(store (i32 (vector_extract (v4i32 VR128:$src),
4645 (iPTR 0))), addr:$dst)],
4648 //===---------------------------------------------------------------------===//
4649 // Move Packed Doubleword Int first element to Doubleword Int
4651 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4652 "mov{d|q}\t{$src, $dst|$dst, $src}",
4653 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4656 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4658 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4659 "mov{d|q}\t{$src, $dst|$dst, $src}",
4660 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4664 //===---------------------------------------------------------------------===//
4665 // Bitcast FR64 <-> GR64
4667 let Predicates = [HasAVX] in
4668 def VMOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4669 "vmovq\t{$src, $dst|$dst, $src}",
4670 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4672 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4673 "mov{d|q}\t{$src, $dst|$dst, $src}",
4674 [(set GR64:$dst, (bitconvert FR64:$src))],
4675 IIC_SSE_MOVDQ>, VEX;
4676 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4677 "movq\t{$src, $dst|$dst, $src}",
4678 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4679 IIC_SSE_MOVDQ>, VEX;
4681 def MOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4682 "movq\t{$src, $dst|$dst, $src}",
4683 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4685 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4686 "mov{d|q}\t{$src, $dst|$dst, $src}",
4687 [(set GR64:$dst, (bitconvert FR64:$src))],
4689 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4690 "movq\t{$src, $dst|$dst, $src}",
4691 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4694 //===---------------------------------------------------------------------===//
4695 // Move Scalar Single to Double Int
4697 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4698 "movd\t{$src, $dst|$dst, $src}",
4699 [(set GR32:$dst, (bitconvert FR32:$src))],
4700 IIC_SSE_MOVD_ToGP>, VEX;
4701 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4702 "movd\t{$src, $dst|$dst, $src}",
4703 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4704 IIC_SSE_MOVDQ>, VEX;
4705 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4706 "movd\t{$src, $dst|$dst, $src}",
4707 [(set GR32:$dst, (bitconvert FR32:$src))],
4709 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4710 "movd\t{$src, $dst|$dst, $src}",
4711 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4714 //===---------------------------------------------------------------------===//
4715 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4717 let AddedComplexity = 15 in {
4718 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4719 "movd\t{$src, $dst|$dst, $src}",
4720 [(set VR128:$dst, (v4i32 (X86vzmovl
4721 (v4i32 (scalar_to_vector GR32:$src)))))],
4722 IIC_SSE_MOVDQ>, VEX;
4723 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4724 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4725 [(set VR128:$dst, (v2i64 (X86vzmovl
4726 (v2i64 (scalar_to_vector GR64:$src)))))],
4730 let AddedComplexity = 15 in {
4731 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4732 "movd\t{$src, $dst|$dst, $src}",
4733 [(set VR128:$dst, (v4i32 (X86vzmovl
4734 (v4i32 (scalar_to_vector GR32:$src)))))],
4736 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4737 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4738 [(set VR128:$dst, (v2i64 (X86vzmovl
4739 (v2i64 (scalar_to_vector GR64:$src)))))],
4743 let AddedComplexity = 20 in {
4744 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4745 "movd\t{$src, $dst|$dst, $src}",
4747 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4748 (loadi32 addr:$src))))))],
4749 IIC_SSE_MOVDQ>, VEX;
4750 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4751 "movd\t{$src, $dst|$dst, $src}",
4753 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4754 (loadi32 addr:$src))))))],
4758 let Predicates = [HasAVX] in {
4759 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4760 let AddedComplexity = 20 in {
4761 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4762 (VMOVZDI2PDIrm addr:$src)>;
4763 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4764 (VMOVZDI2PDIrm addr:$src)>;
4766 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4767 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4768 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4769 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4770 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4771 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4772 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4775 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4776 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4777 (MOVZDI2PDIrm addr:$src)>;
4778 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4779 (MOVZDI2PDIrm addr:$src)>;
4782 // These are the correct encodings of the instructions so that we know how to
4783 // read correct assembly, even though we continue to emit the wrong ones for
4784 // compatibility with Darwin's buggy assembler.
4785 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4786 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4787 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4788 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4789 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4790 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4791 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4792 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4793 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4794 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4795 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4796 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4798 //===---------------------------------------------------------------------===//
4799 // SSE2 - Move Quadword
4800 //===---------------------------------------------------------------------===//
4802 //===---------------------------------------------------------------------===//
4803 // Move Quadword Int to Packed Quadword Int
4805 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4806 "vmovq\t{$src, $dst|$dst, $src}",
4808 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4809 VEX, Requires<[HasAVX]>;
4810 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4811 "movq\t{$src, $dst|$dst, $src}",
4813 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4815 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4817 //===---------------------------------------------------------------------===//
4818 // Move Packed Quadword Int to Quadword Int
4820 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4821 "movq\t{$src, $dst|$dst, $src}",
4822 [(store (i64 (vector_extract (v2i64 VR128:$src),
4823 (iPTR 0))), addr:$dst)],
4824 IIC_SSE_MOVDQ>, VEX;
4825 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4826 "movq\t{$src, $dst|$dst, $src}",
4827 [(store (i64 (vector_extract (v2i64 VR128:$src),
4828 (iPTR 0))), addr:$dst)],
4831 //===---------------------------------------------------------------------===//
4832 // Store / copy lower 64-bits of a XMM register.
4834 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4835 "movq\t{$src, $dst|$dst, $src}",
4836 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4837 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4838 "movq\t{$src, $dst|$dst, $src}",
4839 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4842 let AddedComplexity = 20 in
4843 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4844 "vmovq\t{$src, $dst|$dst, $src}",
4846 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4847 (loadi64 addr:$src))))))],
4849 XS, VEX, Requires<[HasAVX]>;
4851 let AddedComplexity = 20 in
4852 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4853 "movq\t{$src, $dst|$dst, $src}",
4855 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4856 (loadi64 addr:$src))))))],
4858 XS, Requires<[HasSSE2]>;
4860 let Predicates = [HasAVX], AddedComplexity = 20 in {
4861 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4862 (VMOVZQI2PQIrm addr:$src)>;
4863 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4864 (VMOVZQI2PQIrm addr:$src)>;
4865 def : Pat<(v2i64 (X86vzload addr:$src)),
4866 (VMOVZQI2PQIrm addr:$src)>;
4869 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4870 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4871 (MOVZQI2PQIrm addr:$src)>;
4872 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4873 (MOVZQI2PQIrm addr:$src)>;
4874 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4877 let Predicates = [HasAVX] in {
4878 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4879 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4880 def : Pat<(v4i64 (X86vzload addr:$src)),
4881 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4884 //===---------------------------------------------------------------------===//
4885 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4886 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4888 let AddedComplexity = 15 in
4889 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4890 "vmovq\t{$src, $dst|$dst, $src}",
4891 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4893 XS, VEX, Requires<[HasAVX]>;
4894 let AddedComplexity = 15 in
4895 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4896 "movq\t{$src, $dst|$dst, $src}",
4897 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4899 XS, Requires<[HasSSE2]>;
4901 let AddedComplexity = 20 in
4902 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4903 "vmovq\t{$src, $dst|$dst, $src}",
4904 [(set VR128:$dst, (v2i64 (X86vzmovl
4905 (loadv2i64 addr:$src))))],
4907 XS, VEX, Requires<[HasAVX]>;
4908 let AddedComplexity = 20 in {
4909 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4910 "movq\t{$src, $dst|$dst, $src}",
4911 [(set VR128:$dst, (v2i64 (X86vzmovl
4912 (loadv2i64 addr:$src))))],
4914 XS, Requires<[HasSSE2]>;
4917 let AddedComplexity = 20 in {
4918 let Predicates = [HasAVX] in {
4919 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4920 (VMOVZPQILo2PQIrm addr:$src)>;
4921 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4922 (VMOVZPQILo2PQIrr VR128:$src)>;
4924 let Predicates = [HasSSE2] in {
4925 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4926 (MOVZPQILo2PQIrm addr:$src)>;
4927 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4928 (MOVZPQILo2PQIrr VR128:$src)>;
4932 // Instructions to match in the assembler
4933 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4934 "movq\t{$src, $dst|$dst, $src}", [],
4935 IIC_SSE_MOVDQ>, VEX, VEX_W;
4936 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4937 "movq\t{$src, $dst|$dst, $src}", [],
4938 IIC_SSE_MOVDQ>, VEX, VEX_W;
4939 // Recognize "movd" with GR64 destination, but encode as a "movq"
4940 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4941 "movd\t{$src, $dst|$dst, $src}", [],
4942 IIC_SSE_MOVDQ>, VEX, VEX_W;
4944 // Instructions for the disassembler
4945 // xr = XMM register
4948 let Predicates = [HasAVX] in
4949 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4950 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4951 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4952 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4954 //===---------------------------------------------------------------------===//
4955 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4956 //===---------------------------------------------------------------------===//
4957 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4958 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4959 X86MemOperand x86memop> {
4960 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4961 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4962 [(set RC:$dst, (vt (OpNode RC:$src)))],
4964 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4965 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4966 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4970 let Predicates = [HasAVX] in {
4971 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4972 v4f32, VR128, memopv4f32, f128mem>, VEX;
4973 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4974 v4f32, VR128, memopv4f32, f128mem>, VEX;
4975 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4976 v8f32, VR256, memopv8f32, f256mem>, VEX;
4977 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4978 v8f32, VR256, memopv8f32, f256mem>, VEX;
4980 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4981 memopv4f32, f128mem>;
4982 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4983 memopv4f32, f128mem>;
4985 let Predicates = [HasAVX] in {
4986 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4987 (VMOVSHDUPrr VR128:$src)>;
4988 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4989 (VMOVSHDUPrm addr:$src)>;
4990 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4991 (VMOVSLDUPrr VR128:$src)>;
4992 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4993 (VMOVSLDUPrm addr:$src)>;
4994 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4995 (VMOVSHDUPYrr VR256:$src)>;
4996 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4997 (VMOVSHDUPYrm addr:$src)>;
4998 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4999 (VMOVSLDUPYrr VR256:$src)>;
5000 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
5001 (VMOVSLDUPYrm addr:$src)>;
5004 let Predicates = [HasSSE3] in {
5005 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5006 (MOVSHDUPrr VR128:$src)>;
5007 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
5008 (MOVSHDUPrm addr:$src)>;
5009 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5010 (MOVSLDUPrr VR128:$src)>;
5011 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
5012 (MOVSLDUPrm addr:$src)>;
5015 //===---------------------------------------------------------------------===//
5016 // SSE3 - Replicate Double FP - MOVDDUP
5017 //===---------------------------------------------------------------------===//
5019 multiclass sse3_replicate_dfp<string OpcodeStr> {
5020 let neverHasSideEffects = 1 in
5021 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5022 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5023 [], IIC_SSE_MOV_LH>;
5024 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
5025 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5028 (scalar_to_vector (loadf64 addr:$src)))))],
5032 // FIXME: Merge with above classe when there're patterns for the ymm version
5033 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
5034 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
5035 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5036 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
5037 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
5038 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5041 (scalar_to_vector (loadf64 addr:$src)))))]>;
5044 let Predicates = [HasAVX] in {
5045 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
5046 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
5049 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
5051 let Predicates = [HasAVX] in {
5052 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5053 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5054 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5055 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5056 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5057 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5058 def : Pat<(X86Movddup (bc_v2f64
5059 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5060 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5063 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5064 (VMOVDDUPYrm addr:$src)>;
5065 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5066 (VMOVDDUPYrm addr:$src)>;
5067 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5068 (VMOVDDUPYrm addr:$src)>;
5069 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5070 (VMOVDDUPYrr VR256:$src)>;
5073 let Predicates = [HasSSE3] in {
5074 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5075 (MOVDDUPrm addr:$src)>;
5076 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5077 (MOVDDUPrm addr:$src)>;
5078 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5079 (MOVDDUPrm addr:$src)>;
5080 def : Pat<(X86Movddup (bc_v2f64
5081 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5082 (MOVDDUPrm addr:$src)>;
5085 //===---------------------------------------------------------------------===//
5086 // SSE3 - Move Unaligned Integer
5087 //===---------------------------------------------------------------------===//
5089 let Predicates = [HasAVX] in {
5090 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5091 "vlddqu\t{$src, $dst|$dst, $src}",
5092 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5093 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5094 "vlddqu\t{$src, $dst|$dst, $src}",
5095 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5097 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5098 "lddqu\t{$src, $dst|$dst, $src}",
5099 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5102 //===---------------------------------------------------------------------===//
5103 // SSE3 - Arithmetic
5104 //===---------------------------------------------------------------------===//
5106 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5107 X86MemOperand x86memop, OpndItins itins,
5109 def rr : I<0xD0, MRMSrcReg,
5110 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5112 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5113 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5114 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
5115 def rm : I<0xD0, MRMSrcMem,
5116 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5118 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5119 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5120 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5123 let Predicates = [HasAVX] in {
5124 let ExeDomain = SSEPackedSingle in {
5125 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5126 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5127 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5128 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5130 let ExeDomain = SSEPackedDouble in {
5131 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5132 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5133 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5134 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5137 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5138 let ExeDomain = SSEPackedSingle in
5139 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5140 f128mem, SSE_ALU_F32P>, TB, XD;
5141 let ExeDomain = SSEPackedDouble in
5142 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5143 f128mem, SSE_ALU_F64P>, TB, OpSize;
5146 //===---------------------------------------------------------------------===//
5147 // SSE3 Instructions
5148 //===---------------------------------------------------------------------===//
5151 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5152 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5153 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5157 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5159 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5161 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5162 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5163 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5164 IIC_SSE_HADDSUB_RM>;
5166 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5167 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5168 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5170 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5171 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5172 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5174 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5176 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5177 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5178 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5179 IIC_SSE_HADDSUB_RM>;
5182 let Predicates = [HasAVX] in {
5183 let ExeDomain = SSEPackedSingle in {
5184 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5185 X86fhadd, 0>, VEX_4V;
5186 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5187 X86fhsub, 0>, VEX_4V;
5188 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5189 X86fhadd, 0>, VEX_4V;
5190 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5191 X86fhsub, 0>, VEX_4V;
5193 let ExeDomain = SSEPackedDouble in {
5194 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5195 X86fhadd, 0>, VEX_4V;
5196 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5197 X86fhsub, 0>, VEX_4V;
5198 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5199 X86fhadd, 0>, VEX_4V;
5200 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5201 X86fhsub, 0>, VEX_4V;
5205 let Constraints = "$src1 = $dst" in {
5206 let ExeDomain = SSEPackedSingle in {
5207 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5208 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5210 let ExeDomain = SSEPackedDouble in {
5211 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5212 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5216 //===---------------------------------------------------------------------===//
5217 // SSSE3 - Packed Absolute Instructions
5218 //===---------------------------------------------------------------------===//
5221 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5222 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5223 Intrinsic IntId128> {
5224 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5226 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5227 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5230 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5232 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5235 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5239 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5240 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5241 Intrinsic IntId256> {
5242 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5244 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5245 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5248 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5250 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5253 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5256 let Predicates = [HasAVX] in {
5257 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5258 int_x86_ssse3_pabs_b_128>, VEX;
5259 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5260 int_x86_ssse3_pabs_w_128>, VEX;
5261 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5262 int_x86_ssse3_pabs_d_128>, VEX;
5265 let Predicates = [HasAVX2] in {
5266 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5267 int_x86_avx2_pabs_b>, VEX;
5268 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5269 int_x86_avx2_pabs_w>, VEX;
5270 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5271 int_x86_avx2_pabs_d>, VEX;
5274 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5275 int_x86_ssse3_pabs_b_128>;
5276 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5277 int_x86_ssse3_pabs_w_128>;
5278 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5279 int_x86_ssse3_pabs_d_128>;
5281 //===---------------------------------------------------------------------===//
5282 // SSSE3 - Packed Binary Operator Instructions
5283 //===---------------------------------------------------------------------===//
5285 def SSE_PHADDSUBD : OpndItins<
5286 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5288 def SSE_PHADDSUBSW : OpndItins<
5289 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5291 def SSE_PHADDSUBW : OpndItins<
5292 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5294 def SSE_PSHUFB : OpndItins<
5295 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5297 def SSE_PSIGN : OpndItins<
5298 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5300 def SSE_PMULHRSW : OpndItins<
5301 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5304 /// SS3I_binop_rm - Simple SSSE3 bin op
5305 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5306 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5307 X86MemOperand x86memop, OpndItins itins,
5309 let isCommutable = 1 in
5310 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5311 (ins RC:$src1, RC:$src2),
5313 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5314 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5315 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5317 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5318 (ins RC:$src1, x86memop:$src2),
5320 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5323 (OpVT (OpNode RC:$src1,
5324 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5327 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5328 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5329 Intrinsic IntId128, OpndItins itins,
5331 let isCommutable = 1 in
5332 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5333 (ins VR128:$src1, VR128:$src2),
5335 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5337 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5339 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5340 (ins VR128:$src1, i128mem:$src2),
5342 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5345 (IntId128 VR128:$src1,
5346 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5349 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5350 Intrinsic IntId256> {
5351 let isCommutable = 1 in
5352 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5353 (ins VR256:$src1, VR256:$src2),
5354 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5355 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5357 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5358 (ins VR256:$src1, i256mem:$src2),
5359 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5361 (IntId256 VR256:$src1,
5362 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5365 let ImmT = NoImm, Predicates = [HasAVX] in {
5366 let isCommutable = 0 in {
5367 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5368 memopv2i64, i128mem,
5369 SSE_PHADDSUBW, 0>, VEX_4V;
5370 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5371 memopv2i64, i128mem,
5372 SSE_PHADDSUBD, 0>, VEX_4V;
5373 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5374 memopv2i64, i128mem,
5375 SSE_PHADDSUBW, 0>, VEX_4V;
5376 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5377 memopv2i64, i128mem,
5378 SSE_PHADDSUBD, 0>, VEX_4V;
5379 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5380 memopv2i64, i128mem,
5381 SSE_PSIGN, 0>, VEX_4V;
5382 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5383 memopv2i64, i128mem,
5384 SSE_PSIGN, 0>, VEX_4V;
5385 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5386 memopv2i64, i128mem,
5387 SSE_PSIGN, 0>, VEX_4V;
5388 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5389 memopv2i64, i128mem,
5390 SSE_PSHUFB, 0>, VEX_4V;
5391 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5392 int_x86_ssse3_phadd_sw_128,
5393 SSE_PHADDSUBSW, 0>, VEX_4V;
5394 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5395 int_x86_ssse3_phsub_sw_128,
5396 SSE_PHADDSUBSW, 0>, VEX_4V;
5397 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5398 int_x86_ssse3_pmadd_ub_sw_128,
5399 SSE_PMADD, 0>, VEX_4V;
5401 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5402 int_x86_ssse3_pmul_hr_sw_128,
5403 SSE_PMULHRSW, 0>, VEX_4V;
5406 let ImmT = NoImm, Predicates = [HasAVX2] in {
5407 let isCommutable = 0 in {
5408 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5409 memopv4i64, i256mem,
5410 SSE_PHADDSUBW, 0>, VEX_4V;
5411 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5412 memopv4i64, i256mem,
5413 SSE_PHADDSUBW, 0>, VEX_4V;
5414 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5415 memopv4i64, i256mem,
5416 SSE_PHADDSUBW, 0>, VEX_4V;
5417 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5418 memopv4i64, i256mem,
5419 SSE_PHADDSUBW, 0>, VEX_4V;
5420 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5421 memopv4i64, i256mem,
5422 SSE_PHADDSUBW, 0>, VEX_4V;
5423 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5424 memopv4i64, i256mem,
5425 SSE_PHADDSUBW, 0>, VEX_4V;
5426 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5427 memopv4i64, i256mem,
5428 SSE_PHADDSUBW, 0>, VEX_4V;
5429 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5430 memopv4i64, i256mem,
5431 SSE_PHADDSUBW, 0>, VEX_4V;
5432 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5433 int_x86_avx2_phadd_sw>, VEX_4V;
5434 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5435 int_x86_avx2_phsub_sw>, VEX_4V;
5436 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5437 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5439 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5440 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5443 // None of these have i8 immediate fields.
5444 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5445 let isCommutable = 0 in {
5446 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5447 memopv2i64, i128mem, SSE_PHADDSUBW>;
5448 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5449 memopv2i64, i128mem, SSE_PHADDSUBD>;
5450 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5451 memopv2i64, i128mem, SSE_PHADDSUBW>;
5452 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5453 memopv2i64, i128mem, SSE_PHADDSUBD>;
5454 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5455 memopv2i64, i128mem, SSE_PSIGN>;
5456 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5457 memopv2i64, i128mem, SSE_PSIGN>;
5458 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5459 memopv2i64, i128mem, SSE_PSIGN>;
5460 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5461 memopv2i64, i128mem, SSE_PSHUFB>;
5462 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5463 int_x86_ssse3_phadd_sw_128,
5465 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5466 int_x86_ssse3_phsub_sw_128,
5468 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5469 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5471 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5472 int_x86_ssse3_pmul_hr_sw_128,
5476 //===---------------------------------------------------------------------===//
5477 // SSSE3 - Packed Align Instruction Patterns
5478 //===---------------------------------------------------------------------===//
5480 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5481 let neverHasSideEffects = 1 in {
5482 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5483 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5485 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5487 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5488 [], IIC_SSE_PALIGNR>, OpSize;
5490 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5491 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5493 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5495 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5496 [], IIC_SSE_PALIGNR>, OpSize;
5500 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5501 let neverHasSideEffects = 1 in {
5502 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5503 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5505 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5508 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5509 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5511 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5516 let Predicates = [HasAVX] in
5517 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5518 let Predicates = [HasAVX2] in
5519 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5520 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5521 defm PALIGN : ssse3_palign<"palignr">;
5523 let Predicates = [HasAVX2] in {
5524 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5525 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5526 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5527 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5528 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5529 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5530 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5531 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5534 let Predicates = [HasAVX] in {
5535 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5536 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5537 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5538 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5539 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5540 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5541 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5542 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5545 let Predicates = [HasSSSE3] in {
5546 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5547 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5548 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5549 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5550 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5551 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5552 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5553 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5556 //===---------------------------------------------------------------------===//
5557 // SSSE3 - Thread synchronization
5558 //===---------------------------------------------------------------------===//
5560 let usesCustomInserter = 1 in {
5561 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5562 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5563 Requires<[HasSSE3]>;
5564 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5565 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5566 Requires<[HasSSE3]>;
5569 let Uses = [EAX, ECX, EDX] in
5570 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5571 TB, Requires<[HasSSE3]>;
5572 let Uses = [ECX, EAX] in
5573 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5574 TB, Requires<[HasSSE3]>;
5576 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5577 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5579 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5580 Requires<[In32BitMode]>;
5581 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5582 Requires<[In64BitMode]>;
5584 //===----------------------------------------------------------------------===//
5585 // SSE4.1 - Packed Move with Sign/Zero Extend
5586 //===----------------------------------------------------------------------===//
5588 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5589 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5590 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5591 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5593 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5594 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5596 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5600 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5602 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5603 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5604 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5606 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5607 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5608 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5611 let Predicates = [HasAVX] in {
5612 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5614 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5616 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5618 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5620 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5622 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5626 let Predicates = [HasAVX2] in {
5627 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5628 int_x86_avx2_pmovsxbw>, VEX;
5629 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5630 int_x86_avx2_pmovsxwd>, VEX;
5631 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5632 int_x86_avx2_pmovsxdq>, VEX;
5633 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5634 int_x86_avx2_pmovzxbw>, VEX;
5635 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5636 int_x86_avx2_pmovzxwd>, VEX;
5637 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5638 int_x86_avx2_pmovzxdq>, VEX;
5641 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5642 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5643 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5644 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5645 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5646 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5648 let Predicates = [HasAVX] in {
5649 // Common patterns involving scalar load.
5650 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5651 (VPMOVSXBWrm addr:$src)>;
5652 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5653 (VPMOVSXBWrm addr:$src)>;
5655 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5656 (VPMOVSXWDrm addr:$src)>;
5657 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5658 (VPMOVSXWDrm addr:$src)>;
5660 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5661 (VPMOVSXDQrm addr:$src)>;
5662 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5663 (VPMOVSXDQrm addr:$src)>;
5665 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5666 (VPMOVZXBWrm addr:$src)>;
5667 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5668 (VPMOVZXBWrm addr:$src)>;
5670 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5671 (VPMOVZXWDrm addr:$src)>;
5672 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5673 (VPMOVZXWDrm addr:$src)>;
5675 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5676 (VPMOVZXDQrm addr:$src)>;
5677 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5678 (VPMOVZXDQrm addr:$src)>;
5681 let Predicates = [HasSSE41] in {
5682 // Common patterns involving scalar load.
5683 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5684 (PMOVSXBWrm addr:$src)>;
5685 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5686 (PMOVSXBWrm addr:$src)>;
5688 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5689 (PMOVSXWDrm addr:$src)>;
5690 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5691 (PMOVSXWDrm addr:$src)>;
5693 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5694 (PMOVSXDQrm addr:$src)>;
5695 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5696 (PMOVSXDQrm addr:$src)>;
5698 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5699 (PMOVZXBWrm addr:$src)>;
5700 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5701 (PMOVZXBWrm addr:$src)>;
5703 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5704 (PMOVZXWDrm addr:$src)>;
5705 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5706 (PMOVZXWDrm addr:$src)>;
5708 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5709 (PMOVZXDQrm addr:$src)>;
5710 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5711 (PMOVZXDQrm addr:$src)>;
5714 let Predicates = [HasAVX2] in {
5715 let AddedComplexity = 15 in {
5716 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5717 (VPMOVZXDQYrr VR128:$src)>;
5718 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5719 (VPMOVZXWDYrr VR128:$src)>;
5722 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5723 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5726 let Predicates = [HasAVX] in {
5727 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5728 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5731 let Predicates = [HasSSE41] in {
5732 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5733 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5737 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5738 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5739 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5740 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5742 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5743 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5745 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5749 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5751 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5752 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5753 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5755 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5756 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5758 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5762 let Predicates = [HasAVX] in {
5763 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5765 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5767 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5769 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5773 let Predicates = [HasAVX2] in {
5774 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5775 int_x86_avx2_pmovsxbd>, VEX;
5776 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5777 int_x86_avx2_pmovsxwq>, VEX;
5778 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5779 int_x86_avx2_pmovzxbd>, VEX;
5780 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5781 int_x86_avx2_pmovzxwq>, VEX;
5784 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5785 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5786 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5787 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5789 let Predicates = [HasAVX] in {
5790 // Common patterns involving scalar load
5791 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5792 (VPMOVSXBDrm addr:$src)>;
5793 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5794 (VPMOVSXWQrm addr:$src)>;
5796 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5797 (VPMOVZXBDrm addr:$src)>;
5798 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5799 (VPMOVZXWQrm addr:$src)>;
5802 let Predicates = [HasSSE41] in {
5803 // Common patterns involving scalar load
5804 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5805 (PMOVSXBDrm addr:$src)>;
5806 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5807 (PMOVSXWQrm addr:$src)>;
5809 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5810 (PMOVZXBDrm addr:$src)>;
5811 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5812 (PMOVZXWQrm addr:$src)>;
5815 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5816 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5817 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5818 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5820 // Expecting a i16 load any extended to i32 value.
5821 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5822 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5823 [(set VR128:$dst, (IntId (bitconvert
5824 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5828 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5830 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5831 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5832 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5834 // Expecting a i16 load any extended to i32 value.
5835 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5836 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5837 [(set VR256:$dst, (IntId (bitconvert
5838 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5842 let Predicates = [HasAVX] in {
5843 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5845 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5848 let Predicates = [HasAVX2] in {
5849 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5850 int_x86_avx2_pmovsxbq>, VEX;
5851 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5852 int_x86_avx2_pmovzxbq>, VEX;
5854 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5855 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5857 let Predicates = [HasAVX] in {
5858 // Common patterns involving scalar load
5859 def : Pat<(int_x86_sse41_pmovsxbq
5860 (bitconvert (v4i32 (X86vzmovl
5861 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5862 (VPMOVSXBQrm addr:$src)>;
5864 def : Pat<(int_x86_sse41_pmovzxbq
5865 (bitconvert (v4i32 (X86vzmovl
5866 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5867 (VPMOVZXBQrm addr:$src)>;
5870 let Predicates = [HasSSE41] in {
5871 // Common patterns involving scalar load
5872 def : Pat<(int_x86_sse41_pmovsxbq
5873 (bitconvert (v4i32 (X86vzmovl
5874 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5875 (PMOVSXBQrm addr:$src)>;
5877 def : Pat<(int_x86_sse41_pmovzxbq
5878 (bitconvert (v4i32 (X86vzmovl
5879 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5880 (PMOVZXBQrm addr:$src)>;
5883 //===----------------------------------------------------------------------===//
5884 // SSE4.1 - Extract Instructions
5885 //===----------------------------------------------------------------------===//
5887 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5888 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5889 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5890 (ins VR128:$src1, i32i8imm:$src2),
5891 !strconcat(OpcodeStr,
5892 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5893 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5895 let neverHasSideEffects = 1, mayStore = 1 in
5896 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5897 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5898 !strconcat(OpcodeStr,
5899 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5902 // There's an AssertZext in the way of writing the store pattern
5903 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5906 let Predicates = [HasAVX] in {
5907 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5908 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5909 (ins VR128:$src1, i32i8imm:$src2),
5910 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5913 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5916 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5917 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5918 let neverHasSideEffects = 1, mayStore = 1 in
5919 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5920 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5921 !strconcat(OpcodeStr,
5922 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5925 // There's an AssertZext in the way of writing the store pattern
5926 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5929 let Predicates = [HasAVX] in
5930 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5932 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5935 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5936 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5937 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5938 (ins VR128:$src1, i32i8imm:$src2),
5939 !strconcat(OpcodeStr,
5940 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5942 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5943 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5944 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5945 !strconcat(OpcodeStr,
5946 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5947 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5948 addr:$dst)]>, OpSize;
5951 let Predicates = [HasAVX] in
5952 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5954 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5956 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5957 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5958 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5959 (ins VR128:$src1, i32i8imm:$src2),
5960 !strconcat(OpcodeStr,
5961 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5963 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5964 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5965 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5966 !strconcat(OpcodeStr,
5967 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5968 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5969 addr:$dst)]>, OpSize, REX_W;
5972 let Predicates = [HasAVX] in
5973 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5975 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5977 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5979 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5980 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5981 (ins VR128:$src1, i32i8imm:$src2),
5982 !strconcat(OpcodeStr,
5983 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5985 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5987 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5988 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5989 !strconcat(OpcodeStr,
5990 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5991 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5992 addr:$dst)]>, OpSize;
5995 let ExeDomain = SSEPackedSingle in {
5996 let Predicates = [HasAVX] in {
5997 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5998 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5999 (ins VR128:$src1, i32i8imm:$src2),
6000 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
6003 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
6006 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
6007 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6010 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6012 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
6015 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6016 Requires<[HasSSE41]>;
6018 //===----------------------------------------------------------------------===//
6019 // SSE4.1 - Insert Instructions
6020 //===----------------------------------------------------------------------===//
6022 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6023 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6024 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6026 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6028 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6030 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6031 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6032 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6034 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6036 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6038 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6039 imm:$src3))]>, OpSize;
6042 let Predicates = [HasAVX] in
6043 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6044 let Constraints = "$src1 = $dst" in
6045 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6047 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6048 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6049 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6051 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6053 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6055 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6057 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6058 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6060 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6062 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6064 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6065 imm:$src3)))]>, OpSize;
6068 let Predicates = [HasAVX] in
6069 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6070 let Constraints = "$src1 = $dst" in
6071 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6073 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6074 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6075 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6077 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6079 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6081 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6083 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6084 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6086 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6088 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6090 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6091 imm:$src3)))]>, OpSize;
6094 let Predicates = [HasAVX] in
6095 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6096 let Constraints = "$src1 = $dst" in
6097 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6099 // insertps has a few different modes, there's the first two here below which
6100 // are optimized inserts that won't zero arbitrary elements in the destination
6101 // vector. The next one matches the intrinsic and could zero arbitrary elements
6102 // in the target vector.
6103 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6104 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6105 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6107 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6109 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6111 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6113 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6114 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6116 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6118 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6120 (X86insrtps VR128:$src1,
6121 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6122 imm:$src3))]>, OpSize;
6125 let ExeDomain = SSEPackedSingle in {
6126 let Predicates = [HasAVX] in
6127 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6128 let Constraints = "$src1 = $dst" in
6129 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6132 //===----------------------------------------------------------------------===//
6133 // SSE4.1 - Round Instructions
6134 //===----------------------------------------------------------------------===//
6136 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6137 X86MemOperand x86memop, RegisterClass RC,
6138 PatFrag mem_frag32, PatFrag mem_frag64,
6139 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6140 let ExeDomain = SSEPackedSingle in {
6141 // Intrinsic operation, reg.
6142 // Vector intrinsic operation, reg
6143 def PSr : SS4AIi8<opcps, MRMSrcReg,
6144 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6145 !strconcat(OpcodeStr,
6146 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6147 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6150 // Vector intrinsic operation, mem
6151 def PSm : SS4AIi8<opcps, MRMSrcMem,
6152 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6153 !strconcat(OpcodeStr,
6154 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6156 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6158 } // ExeDomain = SSEPackedSingle
6160 let ExeDomain = SSEPackedDouble in {
6161 // Vector intrinsic operation, reg
6162 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6163 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6164 !strconcat(OpcodeStr,
6165 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6166 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6169 // Vector intrinsic operation, mem
6170 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6171 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6172 !strconcat(OpcodeStr,
6173 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6175 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6177 } // ExeDomain = SSEPackedDouble
6180 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6183 Intrinsic F64Int, bit Is2Addr = 1> {
6184 let ExeDomain = GenericDomain in {
6186 def SSr : SS4AIi8<opcss, MRMSrcReg,
6187 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6189 !strconcat(OpcodeStr,
6190 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6191 !strconcat(OpcodeStr,
6192 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6195 // Intrinsic operation, reg.
6196 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6197 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6199 !strconcat(OpcodeStr,
6200 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6201 !strconcat(OpcodeStr,
6202 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6203 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6206 // Intrinsic operation, mem.
6207 def SSm : SS4AIi8<opcss, MRMSrcMem,
6208 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6210 !strconcat(OpcodeStr,
6211 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6212 !strconcat(OpcodeStr,
6213 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6215 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6219 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6220 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6222 !strconcat(OpcodeStr,
6223 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6224 !strconcat(OpcodeStr,
6225 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6228 // Intrinsic operation, reg.
6229 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6230 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6232 !strconcat(OpcodeStr,
6233 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6234 !strconcat(OpcodeStr,
6235 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6236 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6239 // Intrinsic operation, mem.
6240 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6241 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6243 !strconcat(OpcodeStr,
6244 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6245 !strconcat(OpcodeStr,
6246 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6248 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6250 } // ExeDomain = GenericDomain
6253 // FP round - roundss, roundps, roundsd, roundpd
6254 let Predicates = [HasAVX] in {
6256 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6257 memopv4f32, memopv2f64,
6258 int_x86_sse41_round_ps,
6259 int_x86_sse41_round_pd>, VEX;
6260 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6261 memopv8f32, memopv4f64,
6262 int_x86_avx_round_ps_256,
6263 int_x86_avx_round_pd_256>, VEX;
6264 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6265 int_x86_sse41_round_ss,
6266 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6268 def : Pat<(ffloor FR32:$src),
6269 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6270 def : Pat<(f64 (ffloor FR64:$src)),
6271 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6272 def : Pat<(f32 (fnearbyint FR32:$src)),
6273 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6274 def : Pat<(f64 (fnearbyint FR64:$src)),
6275 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6276 def : Pat<(f32 (fceil FR32:$src)),
6277 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6278 def : Pat<(f64 (fceil FR64:$src)),
6279 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6280 def : Pat<(f32 (frint FR32:$src)),
6281 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6282 def : Pat<(f64 (frint FR64:$src)),
6283 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6284 def : Pat<(f32 (ftrunc FR32:$src)),
6285 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6286 def : Pat<(f64 (ftrunc FR64:$src)),
6287 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6290 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6291 memopv4f32, memopv2f64,
6292 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6293 let Constraints = "$src1 = $dst" in
6294 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6295 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6297 def : Pat<(ffloor FR32:$src),
6298 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6299 def : Pat<(f64 (ffloor FR64:$src)),
6300 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6301 def : Pat<(f32 (fnearbyint FR32:$src)),
6302 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6303 def : Pat<(f64 (fnearbyint FR64:$src)),
6304 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6305 def : Pat<(f32 (fceil FR32:$src)),
6306 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6307 def : Pat<(f64 (fceil FR64:$src)),
6308 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6309 def : Pat<(f32 (frint FR32:$src)),
6310 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6311 def : Pat<(f64 (frint FR64:$src)),
6312 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6313 def : Pat<(f32 (ftrunc FR32:$src)),
6314 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6315 def : Pat<(f64 (ftrunc FR64:$src)),
6316 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6318 //===----------------------------------------------------------------------===//
6319 // SSE4.1 - Packed Bit Test
6320 //===----------------------------------------------------------------------===//
6322 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6323 // the intel intrinsic that corresponds to this.
6324 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6325 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6326 "vptest\t{$src2, $src1|$src1, $src2}",
6327 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6329 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6330 "vptest\t{$src2, $src1|$src1, $src2}",
6331 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6334 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6335 "vptest\t{$src2, $src1|$src1, $src2}",
6336 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6338 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6339 "vptest\t{$src2, $src1|$src1, $src2}",
6340 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6344 let Defs = [EFLAGS] in {
6345 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6346 "ptest\t{$src2, $src1|$src1, $src2}",
6347 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6349 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6350 "ptest\t{$src2, $src1|$src1, $src2}",
6351 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6355 // The bit test instructions below are AVX only
6356 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6357 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6358 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6359 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6360 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6361 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6362 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6363 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6367 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6368 let ExeDomain = SSEPackedSingle in {
6369 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6370 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6372 let ExeDomain = SSEPackedDouble in {
6373 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6374 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6378 //===----------------------------------------------------------------------===//
6379 // SSE4.1 - Misc Instructions
6380 //===----------------------------------------------------------------------===//
6382 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6383 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6384 "popcnt{w}\t{$src, $dst|$dst, $src}",
6385 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6387 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6388 "popcnt{w}\t{$src, $dst|$dst, $src}",
6389 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6390 (implicit EFLAGS)]>, OpSize, XS;
6392 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6393 "popcnt{l}\t{$src, $dst|$dst, $src}",
6394 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6396 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6397 "popcnt{l}\t{$src, $dst|$dst, $src}",
6398 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6399 (implicit EFLAGS)]>, XS;
6401 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6402 "popcnt{q}\t{$src, $dst|$dst, $src}",
6403 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6405 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6406 "popcnt{q}\t{$src, $dst|$dst, $src}",
6407 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6408 (implicit EFLAGS)]>, XS;
6413 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6414 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6415 Intrinsic IntId128> {
6416 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6418 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6419 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6420 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6422 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6425 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6428 let Predicates = [HasAVX] in
6429 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6430 int_x86_sse41_phminposuw>, VEX;
6431 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6432 int_x86_sse41_phminposuw>;
6434 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6435 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6436 Intrinsic IntId128, bit Is2Addr = 1> {
6437 let isCommutable = 1 in
6438 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6439 (ins VR128:$src1, VR128:$src2),
6441 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6443 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6444 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6445 (ins VR128:$src1, i128mem:$src2),
6447 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6450 (IntId128 VR128:$src1,
6451 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6454 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6455 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6456 Intrinsic IntId256> {
6457 let isCommutable = 1 in
6458 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6459 (ins VR256:$src1, VR256:$src2),
6460 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6461 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6462 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6463 (ins VR256:$src1, i256mem:$src2),
6464 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6466 (IntId256 VR256:$src1,
6467 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6470 let Predicates = [HasAVX] in {
6471 let isCommutable = 0 in
6472 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6474 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6476 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6478 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6480 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6482 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6484 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6486 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6488 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6490 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6494 let Predicates = [HasAVX2] in {
6495 let isCommutable = 0 in
6496 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6497 int_x86_avx2_packusdw>, VEX_4V;
6498 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6499 int_x86_avx2_pmins_b>, VEX_4V;
6500 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6501 int_x86_avx2_pmins_d>, VEX_4V;
6502 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6503 int_x86_avx2_pminu_d>, VEX_4V;
6504 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6505 int_x86_avx2_pminu_w>, VEX_4V;
6506 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6507 int_x86_avx2_pmaxs_b>, VEX_4V;
6508 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6509 int_x86_avx2_pmaxs_d>, VEX_4V;
6510 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6511 int_x86_avx2_pmaxu_d>, VEX_4V;
6512 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6513 int_x86_avx2_pmaxu_w>, VEX_4V;
6514 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6515 int_x86_avx2_pmul_dq>, VEX_4V;
6518 let Constraints = "$src1 = $dst" in {
6519 let isCommutable = 0 in
6520 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6521 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6522 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6523 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6524 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6525 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6526 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6527 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6528 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6529 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6532 /// SS48I_binop_rm - Simple SSE41 binary operator.
6533 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6534 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6535 X86MemOperand x86memop, bit Is2Addr = 1> {
6536 let isCommutable = 1 in
6537 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6538 (ins RC:$src1, RC:$src2),
6540 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6542 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6543 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6544 (ins RC:$src1, x86memop:$src2),
6546 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6547 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6549 (OpVT (OpNode RC:$src1,
6550 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6553 let Predicates = [HasAVX] in {
6554 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6555 memopv2i64, i128mem, 0>, VEX_4V;
6556 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6557 memopv2i64, i128mem, 0>, VEX_4V;
6559 let Predicates = [HasAVX2] in {
6560 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6561 memopv4i64, i256mem, 0>, VEX_4V;
6562 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6563 memopv4i64, i256mem, 0>, VEX_4V;
6566 let Constraints = "$src1 = $dst" in {
6567 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6568 memopv2i64, i128mem>;
6569 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6570 memopv2i64, i128mem>;
6573 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6574 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6575 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6576 X86MemOperand x86memop, bit Is2Addr = 1> {
6577 let isCommutable = 1 in
6578 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6579 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6581 !strconcat(OpcodeStr,
6582 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6583 !strconcat(OpcodeStr,
6584 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6585 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6587 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6588 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6590 !strconcat(OpcodeStr,
6591 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6592 !strconcat(OpcodeStr,
6593 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6596 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6600 let Predicates = [HasAVX] in {
6601 let isCommutable = 0 in {
6602 let ExeDomain = SSEPackedSingle in {
6603 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6604 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6605 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6606 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6608 let ExeDomain = SSEPackedDouble in {
6609 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6610 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6611 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6612 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6614 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6615 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6616 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6617 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6619 let ExeDomain = SSEPackedSingle in
6620 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6621 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6622 let ExeDomain = SSEPackedDouble in
6623 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6624 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6625 let ExeDomain = SSEPackedSingle in
6626 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6627 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6630 let Predicates = [HasAVX2] in {
6631 let isCommutable = 0 in {
6632 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6633 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6634 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6635 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6639 let Constraints = "$src1 = $dst" in {
6640 let isCommutable = 0 in {
6641 let ExeDomain = SSEPackedSingle in
6642 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6643 VR128, memopv4f32, i128mem>;
6644 let ExeDomain = SSEPackedDouble in
6645 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6646 VR128, memopv2f64, i128mem>;
6647 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6648 VR128, memopv2i64, i128mem>;
6649 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6650 VR128, memopv2i64, i128mem>;
6652 let ExeDomain = SSEPackedSingle in
6653 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6654 VR128, memopv4f32, i128mem>;
6655 let ExeDomain = SSEPackedDouble in
6656 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6657 VR128, memopv2f64, i128mem>;
6660 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6661 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6662 RegisterClass RC, X86MemOperand x86memop,
6663 PatFrag mem_frag, Intrinsic IntId> {
6664 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6665 (ins RC:$src1, RC:$src2, RC:$src3),
6666 !strconcat(OpcodeStr,
6667 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6668 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6669 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6671 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6672 (ins RC:$src1, x86memop:$src2, RC:$src3),
6673 !strconcat(OpcodeStr,
6674 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6676 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6678 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6681 let Predicates = [HasAVX] in {
6682 let ExeDomain = SSEPackedDouble in {
6683 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6684 memopv2f64, int_x86_sse41_blendvpd>;
6685 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6686 memopv4f64, int_x86_avx_blendv_pd_256>;
6687 } // ExeDomain = SSEPackedDouble
6688 let ExeDomain = SSEPackedSingle in {
6689 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6690 memopv4f32, int_x86_sse41_blendvps>;
6691 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6692 memopv8f32, int_x86_avx_blendv_ps_256>;
6693 } // ExeDomain = SSEPackedSingle
6694 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6695 memopv2i64, int_x86_sse41_pblendvb>;
6698 let Predicates = [HasAVX2] in {
6699 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6700 memopv4i64, int_x86_avx2_pblendvb>;
6703 let Predicates = [HasAVX] in {
6704 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6705 (v16i8 VR128:$src2))),
6706 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6707 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6708 (v4i32 VR128:$src2))),
6709 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6710 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6711 (v4f32 VR128:$src2))),
6712 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6713 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6714 (v2i64 VR128:$src2))),
6715 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6716 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6717 (v2f64 VR128:$src2))),
6718 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6719 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6720 (v8i32 VR256:$src2))),
6721 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6722 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6723 (v8f32 VR256:$src2))),
6724 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6725 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6726 (v4i64 VR256:$src2))),
6727 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6728 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6729 (v4f64 VR256:$src2))),
6730 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6732 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6734 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6735 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6737 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6739 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6741 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6742 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6744 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6745 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6747 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6750 let Predicates = [HasAVX2] in {
6751 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6752 (v32i8 VR256:$src2))),
6753 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6754 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6756 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6759 /// SS41I_ternary_int - SSE 4.1 ternary operator
6760 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6761 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6763 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6764 (ins VR128:$src1, VR128:$src2),
6765 !strconcat(OpcodeStr,
6766 "\t{$src2, $dst|$dst, $src2}"),
6767 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6770 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6771 (ins VR128:$src1, i128mem:$src2),
6772 !strconcat(OpcodeStr,
6773 "\t{$src2, $dst|$dst, $src2}"),
6776 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6780 let ExeDomain = SSEPackedDouble in
6781 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6782 int_x86_sse41_blendvpd>;
6783 let ExeDomain = SSEPackedSingle in
6784 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6785 int_x86_sse41_blendvps>;
6786 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6787 int_x86_sse41_pblendvb>;
6789 let Predicates = [HasSSE41] in {
6790 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6791 (v16i8 VR128:$src2))),
6792 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6793 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6794 (v4i32 VR128:$src2))),
6795 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6796 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6797 (v4f32 VR128:$src2))),
6798 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6799 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6800 (v2i64 VR128:$src2))),
6801 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6802 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6803 (v2f64 VR128:$src2))),
6804 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6806 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6808 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6809 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6811 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6812 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6814 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6818 let Predicates = [HasAVX] in
6819 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6820 "vmovntdqa\t{$src, $dst|$dst, $src}",
6821 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6823 let Predicates = [HasAVX2] in
6824 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6825 "vmovntdqa\t{$src, $dst|$dst, $src}",
6826 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6828 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6829 "movntdqa\t{$src, $dst|$dst, $src}",
6830 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6833 //===----------------------------------------------------------------------===//
6834 // SSE4.2 - Compare Instructions
6835 //===----------------------------------------------------------------------===//
6837 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6838 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6839 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6840 X86MemOperand x86memop, bit Is2Addr = 1> {
6841 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6842 (ins RC:$src1, RC:$src2),
6844 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6845 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6846 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6848 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6849 (ins RC:$src1, x86memop:$src2),
6851 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6852 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6854 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6857 let Predicates = [HasAVX] in
6858 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6859 memopv2i64, i128mem, 0>, VEX_4V;
6861 let Predicates = [HasAVX2] in
6862 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6863 memopv4i64, i256mem, 0>, VEX_4V;
6865 let Constraints = "$src1 = $dst" in
6866 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6867 memopv2i64, i128mem>;
6869 //===----------------------------------------------------------------------===//
6870 // SSE4.2 - String/text Processing Instructions
6871 //===----------------------------------------------------------------------===//
6873 // Packed Compare Implicit Length Strings, Return Mask
6874 multiclass pseudo_pcmpistrm<string asm> {
6875 def REG : PseudoI<(outs VR128:$dst),
6876 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6877 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6879 def MEM : PseudoI<(outs VR128:$dst),
6880 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6881 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6882 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6885 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6886 let AddedComplexity = 1 in
6887 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6888 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6891 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6892 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6893 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6894 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6896 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6897 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6898 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6901 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6902 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6903 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6904 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6906 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6907 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6908 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6911 // Packed Compare Explicit Length Strings, Return Mask
6912 multiclass pseudo_pcmpestrm<string asm> {
6913 def REG : PseudoI<(outs VR128:$dst),
6914 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6915 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6916 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6917 def MEM : PseudoI<(outs VR128:$dst),
6918 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6919 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6920 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6923 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6924 let AddedComplexity = 1 in
6925 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6926 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6929 let Predicates = [HasAVX],
6930 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6931 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6932 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6933 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6935 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6936 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6937 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6940 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6941 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6942 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6943 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6945 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6946 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6947 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6950 // Packed Compare Implicit Length Strings, Return Index
6951 let Defs = [ECX, EFLAGS] in {
6952 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6953 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6954 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6955 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6956 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6957 (implicit EFLAGS)]>, OpSize;
6958 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6959 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6960 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6961 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6962 (implicit EFLAGS)]>, OpSize;
6966 let Predicates = [HasAVX] in {
6967 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6969 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6971 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6973 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6975 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6977 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6981 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6982 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6983 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6984 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6985 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6986 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6988 // Packed Compare Explicit Length Strings, Return Index
6989 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6990 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6991 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6992 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6993 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6994 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6995 (implicit EFLAGS)]>, OpSize;
6996 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6997 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6998 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7000 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
7001 (implicit EFLAGS)]>, OpSize;
7005 let Predicates = [HasAVX] in {
7006 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
7008 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
7010 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
7012 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
7014 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
7016 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
7020 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
7021 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
7022 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
7023 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
7024 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
7025 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
7027 //===----------------------------------------------------------------------===//
7028 // SSE4.2 - CRC Instructions
7029 //===----------------------------------------------------------------------===//
7031 // No CRC instructions have AVX equivalents
7033 // crc intrinsic instruction
7034 // This set of instructions are only rm, the only difference is the size
7036 let Constraints = "$src1 = $dst" in {
7037 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7038 (ins GR32:$src1, i8mem:$src2),
7039 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7041 (int_x86_sse42_crc32_32_8 GR32:$src1,
7042 (load addr:$src2)))]>;
7043 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7044 (ins GR32:$src1, GR8:$src2),
7045 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7047 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7048 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7049 (ins GR32:$src1, i16mem:$src2),
7050 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7052 (int_x86_sse42_crc32_32_16 GR32:$src1,
7053 (load addr:$src2)))]>,
7055 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7056 (ins GR32:$src1, GR16:$src2),
7057 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7059 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7061 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7062 (ins GR32:$src1, i32mem:$src2),
7063 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7065 (int_x86_sse42_crc32_32_32 GR32:$src1,
7066 (load addr:$src2)))]>;
7067 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7068 (ins GR32:$src1, GR32:$src2),
7069 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7071 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7072 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7073 (ins GR64:$src1, i8mem:$src2),
7074 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7076 (int_x86_sse42_crc32_64_8 GR64:$src1,
7077 (load addr:$src2)))]>,
7079 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7080 (ins GR64:$src1, GR8:$src2),
7081 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7083 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7085 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7086 (ins GR64:$src1, i64mem:$src2),
7087 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7089 (int_x86_sse42_crc32_64_64 GR64:$src1,
7090 (load addr:$src2)))]>,
7092 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7093 (ins GR64:$src1, GR64:$src2),
7094 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7096 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7100 //===----------------------------------------------------------------------===//
7101 // AES-NI Instructions
7102 //===----------------------------------------------------------------------===//
7104 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7105 Intrinsic IntId128, bit Is2Addr = 1> {
7106 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7107 (ins VR128:$src1, VR128:$src2),
7109 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7110 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7111 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7113 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7114 (ins VR128:$src1, i128mem:$src2),
7116 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7117 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7119 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7122 // Perform One Round of an AES Encryption/Decryption Flow
7123 let Predicates = [HasAVX, HasAES] in {
7124 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7125 int_x86_aesni_aesenc, 0>, VEX_4V;
7126 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7127 int_x86_aesni_aesenclast, 0>, VEX_4V;
7128 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7129 int_x86_aesni_aesdec, 0>, VEX_4V;
7130 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7131 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7134 let Constraints = "$src1 = $dst" in {
7135 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7136 int_x86_aesni_aesenc>;
7137 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7138 int_x86_aesni_aesenclast>;
7139 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7140 int_x86_aesni_aesdec>;
7141 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7142 int_x86_aesni_aesdeclast>;
7145 // Perform the AES InvMixColumn Transformation
7146 let Predicates = [HasAVX, HasAES] in {
7147 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7149 "vaesimc\t{$src1, $dst|$dst, $src1}",
7151 (int_x86_aesni_aesimc VR128:$src1))]>,
7153 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7154 (ins i128mem:$src1),
7155 "vaesimc\t{$src1, $dst|$dst, $src1}",
7156 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7159 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7161 "aesimc\t{$src1, $dst|$dst, $src1}",
7163 (int_x86_aesni_aesimc VR128:$src1))]>,
7165 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7166 (ins i128mem:$src1),
7167 "aesimc\t{$src1, $dst|$dst, $src1}",
7168 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7171 // AES Round Key Generation Assist
7172 let Predicates = [HasAVX, HasAES] in {
7173 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7174 (ins VR128:$src1, i8imm:$src2),
7175 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7177 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7179 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7180 (ins i128mem:$src1, i8imm:$src2),
7181 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7183 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7186 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7187 (ins VR128:$src1, i8imm:$src2),
7188 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7190 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7192 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7193 (ins i128mem:$src1, i8imm:$src2),
7194 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7196 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7199 //===----------------------------------------------------------------------===//
7200 // PCLMUL Instructions
7201 //===----------------------------------------------------------------------===//
7203 // AVX carry-less Multiplication instructions
7204 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7205 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7206 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7208 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7210 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7211 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7212 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7213 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7214 (memopv2i64 addr:$src2), imm:$src3))]>;
7216 // Carry-less Multiplication instructions
7217 let Constraints = "$src1 = $dst" in {
7218 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7219 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7220 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7222 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7224 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7225 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7226 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7227 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7228 (memopv2i64 addr:$src2), imm:$src3))]>;
7229 } // Constraints = "$src1 = $dst"
7232 multiclass pclmul_alias<string asm, int immop> {
7233 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7234 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7236 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7237 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7239 def : InstAlias<!strconcat("vpclmul", asm,
7240 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7241 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7243 def : InstAlias<!strconcat("vpclmul", asm,
7244 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7245 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7247 defm : pclmul_alias<"hqhq", 0x11>;
7248 defm : pclmul_alias<"hqlq", 0x01>;
7249 defm : pclmul_alias<"lqhq", 0x10>;
7250 defm : pclmul_alias<"lqlq", 0x00>;
7252 //===----------------------------------------------------------------------===//
7253 // SSE4A Instructions
7254 //===----------------------------------------------------------------------===//
7256 let Predicates = [HasSSE4A] in {
7258 let Constraints = "$src = $dst" in {
7259 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7260 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7261 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7262 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7263 imm:$idx))]>, TB, OpSize;
7264 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7265 (ins VR128:$src, VR128:$mask),
7266 "extrq\t{$mask, $src|$src, $mask}",
7267 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7268 VR128:$mask))]>, TB, OpSize;
7270 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7271 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7272 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7273 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7274 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7275 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7276 (ins VR128:$src, VR128:$mask),
7277 "insertq\t{$mask, $src|$src, $mask}",
7278 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7279 VR128:$mask))]>, XD;
7282 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7283 "movntss\t{$src, $dst|$dst, $src}",
7284 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7286 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7287 "movntsd\t{$src, $dst|$dst, $src}",
7288 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7291 //===----------------------------------------------------------------------===//
7293 //===----------------------------------------------------------------------===//
7295 //===----------------------------------------------------------------------===//
7296 // VBROADCAST - Load from memory and broadcast to all elements of the
7297 // destination operand
7299 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7300 X86MemOperand x86memop, Intrinsic Int> :
7301 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7302 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7303 [(set RC:$dst, (Int addr:$src))]>, VEX;
7305 // AVX2 adds register forms
7306 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7308 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7309 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7310 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7312 let ExeDomain = SSEPackedSingle in {
7313 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7314 int_x86_avx_vbroadcast_ss>;
7315 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7316 int_x86_avx_vbroadcast_ss_256>;
7318 let ExeDomain = SSEPackedDouble in
7319 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7320 int_x86_avx_vbroadcast_sd_256>;
7321 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7322 int_x86_avx_vbroadcastf128_pd_256>;
7324 let ExeDomain = SSEPackedSingle in {
7325 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7326 int_x86_avx2_vbroadcast_ss_ps>;
7327 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7328 int_x86_avx2_vbroadcast_ss_ps_256>;
7330 let ExeDomain = SSEPackedDouble in
7331 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7332 int_x86_avx2_vbroadcast_sd_pd_256>;
7334 let Predicates = [HasAVX2] in
7335 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7336 int_x86_avx2_vbroadcasti128>;
7338 let Predicates = [HasAVX] in
7339 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7340 (VBROADCASTF128 addr:$src)>;
7343 //===----------------------------------------------------------------------===//
7344 // VINSERTF128 - Insert packed floating-point values
7346 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7347 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7348 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7349 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7352 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7353 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7354 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7358 let Predicates = [HasAVX] in {
7359 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7361 (VINSERTF128rr VR256:$src1, VR128:$src2,
7362 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7363 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7365 (VINSERTF128rr VR256:$src1, VR128:$src2,
7366 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7367 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7369 (VINSERTF128rr VR256:$src1, VR128:$src2,
7370 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7371 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7373 (VINSERTF128rr VR256:$src1, VR128:$src2,
7374 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7375 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7377 (VINSERTF128rr VR256:$src1, VR128:$src2,
7378 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7379 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7381 (VINSERTF128rr VR256:$src1, VR128:$src2,
7382 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7384 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7386 (VINSERTF128rm VR256:$src1, addr:$src2,
7387 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7388 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7390 (VINSERTF128rm VR256:$src1, addr:$src2,
7391 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7392 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7394 (VINSERTF128rm VR256:$src1, addr:$src2,
7395 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7398 //===----------------------------------------------------------------------===//
7399 // VEXTRACTF128 - Extract packed floating-point values
7401 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7402 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7403 (ins VR256:$src1, i8imm:$src2),
7404 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7407 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7408 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7409 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7413 // Extract and store.
7414 let Predicates = [HasAVX] in {
7415 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7416 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7417 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7418 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7419 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7420 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7422 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7423 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7424 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7425 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7426 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7427 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7431 let Predicates = [HasAVX] in {
7432 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7433 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7434 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7435 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7436 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7437 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7439 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7440 (v4f32 (VEXTRACTF128rr
7441 (v8f32 VR256:$src1),
7442 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7443 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7444 (v2f64 (VEXTRACTF128rr
7445 (v4f64 VR256:$src1),
7446 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7447 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7448 (v2i64 (VEXTRACTF128rr
7449 (v4i64 VR256:$src1),
7450 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7451 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7452 (v4i32 (VEXTRACTF128rr
7453 (v8i32 VR256:$src1),
7454 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7455 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7456 (v8i16 (VEXTRACTF128rr
7457 (v16i16 VR256:$src1),
7458 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7459 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7460 (v16i8 (VEXTRACTF128rr
7461 (v32i8 VR256:$src1),
7462 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7465 //===----------------------------------------------------------------------===//
7466 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7468 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7469 Intrinsic IntLd, Intrinsic IntLd256,
7470 Intrinsic IntSt, Intrinsic IntSt256> {
7471 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7472 (ins VR128:$src1, f128mem:$src2),
7473 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7474 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7476 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7477 (ins VR256:$src1, f256mem:$src2),
7478 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7479 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7481 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7482 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7483 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7484 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7485 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7486 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7487 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7488 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7491 let ExeDomain = SSEPackedSingle in
7492 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7493 int_x86_avx_maskload_ps,
7494 int_x86_avx_maskload_ps_256,
7495 int_x86_avx_maskstore_ps,
7496 int_x86_avx_maskstore_ps_256>;
7497 let ExeDomain = SSEPackedDouble in
7498 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7499 int_x86_avx_maskload_pd,
7500 int_x86_avx_maskload_pd_256,
7501 int_x86_avx_maskstore_pd,
7502 int_x86_avx_maskstore_pd_256>;
7504 //===----------------------------------------------------------------------===//
7505 // VPERMIL - Permute Single and Double Floating-Point Values
7507 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7508 RegisterClass RC, X86MemOperand x86memop_f,
7509 X86MemOperand x86memop_i, PatFrag i_frag,
7510 Intrinsic IntVar, ValueType vt> {
7511 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7512 (ins RC:$src1, RC:$src2),
7513 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7514 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7515 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7516 (ins RC:$src1, x86memop_i:$src2),
7517 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7518 [(set RC:$dst, (IntVar RC:$src1,
7519 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7521 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7522 (ins RC:$src1, i8imm:$src2),
7523 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7524 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7525 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7526 (ins x86memop_f:$src1, i8imm:$src2),
7527 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7529 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7532 let ExeDomain = SSEPackedSingle in {
7533 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7534 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7535 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7536 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7538 let ExeDomain = SSEPackedDouble in {
7539 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7540 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7541 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7542 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7545 let Predicates = [HasAVX] in {
7546 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7547 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7548 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7549 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7550 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7552 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7553 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7554 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7556 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7557 (VPERMILPDri VR128:$src1, imm:$imm)>;
7558 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7559 (VPERMILPDmi addr:$src1, imm:$imm)>;
7562 //===----------------------------------------------------------------------===//
7563 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7565 let ExeDomain = SSEPackedSingle in {
7566 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7567 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7568 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7569 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7570 (i8 imm:$src3))))]>, VEX_4V;
7571 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7572 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7573 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7574 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7575 (i8 imm:$src3)))]>, VEX_4V;
7578 let Predicates = [HasAVX] in {
7579 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7580 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7581 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7582 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7583 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7584 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7585 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7586 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7587 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7588 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7590 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7591 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7592 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7593 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7594 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7595 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7596 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7597 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7598 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7599 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7600 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7601 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7602 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7603 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7604 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7605 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7606 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7607 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7610 //===----------------------------------------------------------------------===//
7611 // VZERO - Zero YMM registers
7613 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7614 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7615 // Zero All YMM registers
7616 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7617 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7619 // Zero Upper bits of YMM registers
7620 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7621 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7624 //===----------------------------------------------------------------------===//
7625 // Half precision conversion instructions
7626 //===----------------------------------------------------------------------===//
7627 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7628 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7629 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7630 [(set RC:$dst, (Int VR128:$src))]>,
7632 let neverHasSideEffects = 1, mayLoad = 1 in
7633 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7634 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7637 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7638 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7639 (ins RC:$src1, i32i8imm:$src2),
7640 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7641 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7643 let neverHasSideEffects = 1, mayStore = 1 in
7644 def mr : Ii8<0x1D, MRMDestMem, (outs),
7645 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7646 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7650 let Predicates = [HasAVX, HasF16C] in {
7651 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7652 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7653 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7654 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7657 //===----------------------------------------------------------------------===//
7658 // AVX2 Instructions
7659 //===----------------------------------------------------------------------===//
7661 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7662 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7663 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7664 X86MemOperand x86memop> {
7665 let isCommutable = 1 in
7666 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7667 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7668 !strconcat(OpcodeStr,
7669 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7670 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7672 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7673 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7674 !strconcat(OpcodeStr,
7675 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7678 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7682 let isCommutable = 0 in {
7683 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7684 VR128, memopv2i64, i128mem>;
7685 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7686 VR256, memopv4i64, i256mem>;
7689 //===----------------------------------------------------------------------===//
7690 // VPBROADCAST - Load from memory and broadcast to all elements of the
7691 // destination operand
7693 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7694 X86MemOperand x86memop, PatFrag ld_frag,
7695 Intrinsic Int128, Intrinsic Int256> {
7696 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7698 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7699 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7700 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7702 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7703 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7704 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7705 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7706 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7707 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7709 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7712 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7713 int_x86_avx2_pbroadcastb_128,
7714 int_x86_avx2_pbroadcastb_256>;
7715 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7716 int_x86_avx2_pbroadcastw_128,
7717 int_x86_avx2_pbroadcastw_256>;
7718 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7719 int_x86_avx2_pbroadcastd_128,
7720 int_x86_avx2_pbroadcastd_256>;
7721 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7722 int_x86_avx2_pbroadcastq_128,
7723 int_x86_avx2_pbroadcastq_256>;
7725 let Predicates = [HasAVX2] in {
7726 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7727 (VPBROADCASTBrm addr:$src)>;
7728 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7729 (VPBROADCASTBYrm addr:$src)>;
7730 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7731 (VPBROADCASTWrm addr:$src)>;
7732 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7733 (VPBROADCASTWYrm addr:$src)>;
7734 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7735 (VPBROADCASTDrm addr:$src)>;
7736 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7737 (VPBROADCASTDYrm addr:$src)>;
7738 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7739 (VPBROADCASTQrm addr:$src)>;
7740 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7741 (VPBROADCASTQYrm addr:$src)>;
7743 // Provide fallback in case the load node that is used in the patterns above
7744 // is used by additional users, which prevents the pattern selection.
7745 let AddedComplexity = 20 in {
7746 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7748 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7749 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7751 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7752 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7754 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd))>;
7756 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7758 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7759 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7761 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7762 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7764 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd))>;
7768 // AVX1 broadcast patterns
7769 let Predicates = [HasAVX] in {
7770 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7771 (VBROADCASTSSYrm addr:$src)>;
7772 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7773 (VBROADCASTSDrm addr:$src)>;
7774 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7775 (VBROADCASTSSYrm addr:$src)>;
7776 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7777 (VBROADCASTSDrm addr:$src)>;
7778 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7779 (VBROADCASTSSrm addr:$src)>;
7780 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7781 (VBROADCASTSSrm addr:$src)>;
7783 // Provide fallback in case the load node that is used in the patterns above
7784 // is used by additional users, which prevents the pattern selection.
7785 let AddedComplexity = 20 in {
7786 // 128bit broadcasts:
7787 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7789 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0)>;
7790 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7791 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7793 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0),
7796 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss),
7798 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7799 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7801 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd), 0),
7804 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
7807 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7809 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0)>;
7810 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7811 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7813 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0),
7816 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss),
7818 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7819 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7821 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd), 0),
7824 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
7829 //===----------------------------------------------------------------------===//
7830 // VPERM - Permute instructions
7833 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7835 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7836 (ins VR256:$src1, VR256:$src2),
7837 !strconcat(OpcodeStr,
7838 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7840 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7841 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7842 (ins VR256:$src1, i256mem:$src2),
7843 !strconcat(OpcodeStr,
7844 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7846 (OpVT (X86VPermv VR256:$src1,
7847 (bitconvert (mem_frag addr:$src2)))))]>,
7851 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7852 let ExeDomain = SSEPackedSingle in
7853 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7855 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7857 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7858 (ins VR256:$src1, i8imm:$src2),
7859 !strconcat(OpcodeStr,
7860 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7862 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7863 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7864 (ins i256mem:$src1, i8imm:$src2),
7865 !strconcat(OpcodeStr,
7866 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7868 (OpVT (X86VPermi (mem_frag addr:$src1),
7869 (i8 imm:$src2))))]>, VEX;
7872 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7873 let ExeDomain = SSEPackedDouble in
7874 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7876 //===----------------------------------------------------------------------===//
7877 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7879 let AddedComplexity = 1 in {
7880 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7881 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7882 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7883 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7884 (i8 imm:$src3))))]>, VEX_4V;
7885 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7886 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7887 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7888 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7889 (i8 imm:$src3)))]>, VEX_4V;
7892 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7893 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7894 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7895 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7896 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7897 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7898 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7900 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7902 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7903 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7904 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7905 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7906 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7908 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7912 //===----------------------------------------------------------------------===//
7913 // VINSERTI128 - Insert packed integer values
7915 let neverHasSideEffects = 1 in {
7916 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7917 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7918 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7921 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7922 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7923 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7927 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7928 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7930 (VINSERTI128rr VR256:$src1, VR128:$src2,
7931 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7932 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7934 (VINSERTI128rr VR256:$src1, VR128:$src2,
7935 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7936 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7938 (VINSERTI128rr VR256:$src1, VR128:$src2,
7939 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7940 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7942 (VINSERTI128rr VR256:$src1, VR128:$src2,
7943 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7946 //===----------------------------------------------------------------------===//
7947 // VEXTRACTI128 - Extract packed integer values
7949 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7950 (ins VR256:$src1, i8imm:$src2),
7951 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7953 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7955 let neverHasSideEffects = 1, mayStore = 1 in
7956 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7957 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7958 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7960 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7961 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7962 (v2i64 (VEXTRACTI128rr
7963 (v4i64 VR256:$src1),
7964 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7965 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7966 (v4i32 (VEXTRACTI128rr
7967 (v8i32 VR256:$src1),
7968 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7969 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7970 (v8i16 (VEXTRACTI128rr
7971 (v16i16 VR256:$src1),
7972 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7973 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7974 (v16i8 (VEXTRACTI128rr
7975 (v32i8 VR256:$src1),
7976 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7979 //===----------------------------------------------------------------------===//
7980 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7982 multiclass avx2_pmovmask<string OpcodeStr,
7983 Intrinsic IntLd128, Intrinsic IntLd256,
7984 Intrinsic IntSt128, Intrinsic IntSt256> {
7985 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7986 (ins VR128:$src1, i128mem:$src2),
7987 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7988 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7989 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7990 (ins VR256:$src1, i256mem:$src2),
7991 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7992 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7993 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7994 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7995 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7996 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7997 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7998 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7999 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8000 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
8003 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8004 int_x86_avx2_maskload_d,
8005 int_x86_avx2_maskload_d_256,
8006 int_x86_avx2_maskstore_d,
8007 int_x86_avx2_maskstore_d_256>;
8008 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8009 int_x86_avx2_maskload_q,
8010 int_x86_avx2_maskload_q_256,
8011 int_x86_avx2_maskstore_q,
8012 int_x86_avx2_maskstore_q_256>, VEX_W;
8015 //===----------------------------------------------------------------------===//
8016 // Variable Bit Shifts
8018 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8019 ValueType vt128, ValueType vt256> {
8020 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8021 (ins VR128:$src1, VR128:$src2),
8022 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8024 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8026 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8027 (ins VR128:$src1, i128mem:$src2),
8028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8030 (vt128 (OpNode VR128:$src1,
8031 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8033 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8034 (ins VR256:$src1, VR256:$src2),
8035 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8037 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8039 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8040 (ins VR256:$src1, i256mem:$src2),
8041 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8043 (vt256 (OpNode VR256:$src1,
8044 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8048 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8049 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8050 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8051 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8052 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;