1 //===-- X86InstrSSE.td - SSE Instruction Set ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
21 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
27 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
28 InstrItinClass arg_ri> {
29 InstrItinClass rr = arg_rr;
30 InstrItinClass rm = arg_rm;
31 InstrItinClass ri = arg_ri;
36 def SSE_ALU_F32S : OpndItins<
37 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
40 def SSE_ALU_F64S : OpndItins<
41 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
44 def SSE_ALU_ITINS_S : SizeItins<
45 SSE_ALU_F32S, SSE_ALU_F64S
48 def SSE_MUL_F32S : OpndItins<
49 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
52 def SSE_MUL_F64S : OpndItins<
53 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
56 def SSE_MUL_ITINS_S : SizeItins<
57 SSE_MUL_F32S, SSE_MUL_F64S
60 def SSE_DIV_F32S : OpndItins<
61 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
64 def SSE_DIV_F64S : OpndItins<
65 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
68 def SSE_DIV_ITINS_S : SizeItins<
69 SSE_DIV_F32S, SSE_DIV_F64S
73 def SSE_ALU_F32P : OpndItins<
74 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
77 def SSE_ALU_F64P : OpndItins<
78 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
81 def SSE_ALU_ITINS_P : SizeItins<
82 SSE_ALU_F32P, SSE_ALU_F64P
85 def SSE_MUL_F32P : OpndItins<
86 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
89 def SSE_MUL_F64P : OpndItins<
90 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
93 def SSE_MUL_ITINS_P : SizeItins<
94 SSE_MUL_F32P, SSE_MUL_F64P
97 def SSE_DIV_F32P : OpndItins<
98 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
101 def SSE_DIV_F64P : OpndItins<
102 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
105 def SSE_DIV_ITINS_P : SizeItins<
106 SSE_DIV_F32P, SSE_DIV_F64P
109 def SSE_BIT_ITINS_P : OpndItins<
110 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
113 def SSE_INTALU_ITINS_P : OpndItins<
114 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
117 def SSE_INTALUQ_ITINS_P : OpndItins<
118 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
121 def SSE_INTMUL_ITINS_P : OpndItins<
122 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
125 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
126 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
129 def SSE_MOVA_ITINS : OpndItins<
130 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
133 def SSE_MOVU_ITINS : OpndItins<
134 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
137 //===----------------------------------------------------------------------===//
138 // SSE 1 & 2 Instructions Classes
139 //===----------------------------------------------------------------------===//
141 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
142 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
143 RegisterClass RC, X86MemOperand x86memop,
146 let isCommutable = 1 in {
147 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
149 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
150 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
151 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
153 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
157 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
160 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
161 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
162 string asm, string SSEVer, string FPSizeStr,
163 Operand memopr, ComplexPattern mem_cpat,
166 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
168 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
169 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
170 [(set RC:$dst, (!cast<Intrinsic>(
171 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
172 RC:$src1, RC:$src2))], itins.rr>;
173 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
175 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
176 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
177 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
178 SSEVer, "_", OpcodeStr, FPSizeStr))
179 RC:$src1, mem_cpat:$src2))], itins.rm>;
182 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
183 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
184 RegisterClass RC, ValueType vt,
185 X86MemOperand x86memop, PatFrag mem_frag,
186 Domain d, OpndItins itins, bit Is2Addr = 1> {
187 let isCommutable = 1 in
188 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
190 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
191 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
192 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
194 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
196 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
197 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
198 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
202 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
203 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
204 string OpcodeStr, X86MemOperand x86memop,
205 list<dag> pat_rr, list<dag> pat_rm,
207 bit rr_hasSideEffects = 0> {
208 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
209 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
213 pat_rr, IIC_DEFAULT, d>;
214 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
218 pat_rm, IIC_DEFAULT, d>;
221 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
222 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
223 string asm, string SSEVer, string FPSizeStr,
224 X86MemOperand x86memop, PatFrag mem_frag,
225 Domain d, OpndItins itins, bit Is2Addr = 1> {
226 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
228 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
229 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
230 [(set RC:$dst, (!cast<Intrinsic>(
231 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
232 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
233 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
235 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
236 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
237 [(set RC:$dst, (!cast<Intrinsic>(
238 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
239 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
242 //===----------------------------------------------------------------------===//
243 // Non-instruction patterns
244 //===----------------------------------------------------------------------===//
246 // A vector extract of the first f32/f64 position is a subregister copy
247 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
248 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
249 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
250 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
252 // A 128-bit subvector extract from the first 256-bit vector position
253 // is a subregister copy that needs no instruction.
254 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
255 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
256 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
257 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
259 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
260 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
261 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
262 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
264 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
265 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
266 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
267 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
269 // A 128-bit subvector insert to the first 256-bit vector position
270 // is a subregister copy that needs no instruction.
271 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
272 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
273 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
274 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
275 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
276 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
277 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
278 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
279 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
280 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
281 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
282 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
284 // Implicitly promote a 32-bit scalar to a vector.
285 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
286 (COPY_TO_REGCLASS FR32:$src, VR128)>;
287 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
288 (COPY_TO_REGCLASS FR32:$src, VR128)>;
289 // Implicitly promote a 64-bit scalar to a vector.
290 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
291 (COPY_TO_REGCLASS FR64:$src, VR128)>;
292 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
293 (COPY_TO_REGCLASS FR64:$src, VR128)>;
295 // Bitcasts between 128-bit vector types. Return the original type since
296 // no instruction is needed for the conversion
297 let Predicates = [HasSSE2] in {
298 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
299 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
300 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
301 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
302 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
303 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
304 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
305 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
306 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
307 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
308 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
309 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
310 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
311 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
312 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
313 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
314 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
315 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
316 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
317 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
318 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
319 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
320 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
321 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
322 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
323 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
324 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
325 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
326 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
327 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
330 // Bitcasts between 256-bit vector types. Return the original type since
331 // no instruction is needed for the conversion
332 let Predicates = [HasAVX] in {
333 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
334 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
335 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
336 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
337 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
338 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
339 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
340 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
341 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
342 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
343 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
344 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
345 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
346 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
347 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
348 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
349 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
350 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
351 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
352 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
353 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
354 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
355 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
356 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
357 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
358 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
359 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
360 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
361 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
362 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
365 // Alias instructions that map fld0 to pxor for sse.
366 // This is expanded by ExpandPostRAPseudos.
367 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
369 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
370 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
371 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
372 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
375 //===----------------------------------------------------------------------===//
376 // AVX & SSE - Zero/One Vectors
377 //===----------------------------------------------------------------------===//
379 // Alias instruction that maps zero vector to pxor / xorp* for sse.
380 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
381 // swizzled by ExecutionDepsFix to pxor.
382 // We set canFoldAsLoad because this can be converted to a constant-pool
383 // load of an all-zeros value if folding it would be beneficial.
384 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
385 isPseudo = 1, neverHasSideEffects = 1 in {
386 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
389 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
390 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
391 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
392 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
393 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
394 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
397 // The same as done above but for AVX. The 256-bit ISA does not support PI,
398 // and doesn't need it because on sandy bridge the register is set to zero
399 // at the rename stage without using any execution unit, so SET0PSY
400 // and SET0PDY can be used for vector int instructions without penalty
401 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
402 // JIT implementatioan, it does not expand the instructions below like
403 // X86MCInstLower does.
404 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
405 isCodeGenOnly = 1 in {
406 let Predicates = [HasAVX] in {
407 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
408 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
409 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
410 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
412 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
413 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
417 let Predicates = [HasAVX2], AddedComplexity = 5 in {
418 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
419 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
420 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
421 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
424 // AVX has no support for 256-bit integer instructions, but since the 128-bit
425 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
426 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
427 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
428 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
430 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
431 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
432 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
434 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
435 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
436 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
438 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
439 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
440 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
442 // We set canFoldAsLoad because this can be converted to a constant-pool
443 // load of an all-ones value if folding it would be beneficial.
444 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
445 // JIT implementation, it does not expand the instructions below like
446 // X86MCInstLower does.
447 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
448 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
449 let Predicates = [HasAVX] in
450 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
451 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
452 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
453 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
454 let Predicates = [HasAVX2] in
455 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
456 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
460 //===----------------------------------------------------------------------===//
461 // SSE 1 & 2 - Move FP Scalar Instructions
463 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
464 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
465 // is used instead. Register-to-register movss/movsd is not modeled as an
466 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
467 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
468 //===----------------------------------------------------------------------===//
470 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
471 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
472 [(set VR128:$dst, (vt (OpNode VR128:$src1,
473 (scalar_to_vector RC:$src2))))],
476 // Loading from memory automatically zeroing upper bits.
477 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
478 PatFrag mem_pat, string OpcodeStr> :
479 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
481 [(set RC:$dst, (mem_pat addr:$src))],
485 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
486 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
488 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
489 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
492 // For the disassembler
493 let isCodeGenOnly = 1 in {
494 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
495 (ins VR128:$src1, FR32:$src2),
496 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
499 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
500 (ins VR128:$src1, FR64:$src2),
501 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
506 let canFoldAsLoad = 1, isReMaterializable = 1 in {
507 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
509 let AddedComplexity = 20 in
510 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
514 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
515 "movss\t{$src, $dst|$dst, $src}",
516 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
518 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
519 "movsd\t{$src, $dst|$dst, $src}",
520 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
524 let Constraints = "$src1 = $dst" in {
525 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
526 "movss\t{$src2, $dst|$dst, $src2}">, XS;
527 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
528 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
530 // For the disassembler
531 let isCodeGenOnly = 1 in {
532 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
533 (ins VR128:$src1, FR32:$src2),
534 "movss\t{$src2, $dst|$dst, $src2}", [],
535 IIC_SSE_MOV_S_RR>, XS;
536 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
537 (ins VR128:$src1, FR64:$src2),
538 "movsd\t{$src2, $dst|$dst, $src2}", [],
539 IIC_SSE_MOV_S_RR>, XD;
543 let canFoldAsLoad = 1, isReMaterializable = 1 in {
544 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
546 let AddedComplexity = 20 in
547 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
550 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
551 "movss\t{$src, $dst|$dst, $src}",
552 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
553 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
554 "movsd\t{$src, $dst|$dst, $src}",
555 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
558 let Predicates = [HasAVX] in {
559 let AddedComplexity = 15 in {
560 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
561 // MOVS{S,D} to the lower bits.
562 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
563 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
564 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
565 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
566 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
567 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
568 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
569 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
571 // Move low f32 and clear high bits.
572 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
573 (SUBREG_TO_REG (i32 0),
574 (VMOVSSrr (v4f32 (V_SET0)),
575 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
576 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
577 (SUBREG_TO_REG (i32 0),
578 (VMOVSSrr (v4i32 (V_SET0)),
579 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
582 let AddedComplexity = 20 in {
583 // MOVSSrm zeros the high parts of the register; represent this
584 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
585 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
586 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
587 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
588 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
589 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
590 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
592 // MOVSDrm zeros the high parts of the register; represent this
593 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
594 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
595 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
596 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
597 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
598 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
599 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
600 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
601 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
602 def : Pat<(v2f64 (X86vzload addr:$src)),
603 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
605 // Represent the same patterns above but in the form they appear for
607 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
608 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
609 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
610 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
611 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
612 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
613 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
614 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
615 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
617 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
618 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
619 (SUBREG_TO_REG (i32 0),
620 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
622 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
623 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
624 (SUBREG_TO_REG (i64 0),
625 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
627 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
628 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
629 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
631 // Move low f64 and clear high bits.
632 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
633 (SUBREG_TO_REG (i32 0),
634 (VMOVSDrr (v2f64 (V_SET0)),
635 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
637 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
638 (SUBREG_TO_REG (i32 0),
639 (VMOVSDrr (v2i64 (V_SET0)),
640 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
642 // Extract and store.
643 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
645 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
646 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
648 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
650 // Shuffle with VMOVSS
651 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
652 (VMOVSSrr (v4i32 VR128:$src1),
653 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
654 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
655 (VMOVSSrr (v4f32 VR128:$src1),
656 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
659 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
660 (SUBREG_TO_REG (i32 0),
661 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
662 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
664 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
665 (SUBREG_TO_REG (i32 0),
666 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
667 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
670 // Shuffle with VMOVSD
671 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
672 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
673 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
674 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
675 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
676 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
677 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
678 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
681 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
682 (SUBREG_TO_REG (i32 0),
683 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
684 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
686 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
687 (SUBREG_TO_REG (i32 0),
688 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
689 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
693 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
694 // is during lowering, where it's not possible to recognize the fold cause
695 // it has two uses through a bitcast. One use disappears at isel time and the
696 // fold opportunity reappears.
697 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
698 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
699 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
700 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
701 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
702 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
703 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
704 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
707 let Predicates = [HasSSE1] in {
708 let AddedComplexity = 15 in {
709 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
710 // MOVSS to the lower bits.
711 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
712 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
713 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
714 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
715 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
716 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
719 let AddedComplexity = 20 in {
720 // MOVSSrm already zeros the high parts of the register.
721 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
722 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
723 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
724 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
725 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
726 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
729 // Extract and store.
730 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
732 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
734 // Shuffle with MOVSS
735 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
736 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
737 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
738 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
741 let Predicates = [HasSSE2] in {
742 let AddedComplexity = 15 in {
743 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
744 // MOVSD to the lower bits.
745 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
746 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
749 let AddedComplexity = 20 in {
750 // MOVSDrm already zeros the high parts of the register.
751 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
752 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
753 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
754 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
755 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
756 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
757 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
758 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
759 def : Pat<(v2f64 (X86vzload addr:$src)),
760 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
763 // Extract and store.
764 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
766 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
768 // Shuffle with MOVSD
769 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
770 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
771 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
772 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
773 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
774 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
775 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
776 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
778 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
779 // is during lowering, where it's not possible to recognize the fold cause
780 // it has two uses through a bitcast. One use disappears at isel time and the
781 // fold opportunity reappears.
782 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
783 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
784 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
785 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
786 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
787 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
788 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
789 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
792 //===----------------------------------------------------------------------===//
793 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
794 //===----------------------------------------------------------------------===//
796 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
797 X86MemOperand x86memop, PatFrag ld_frag,
798 string asm, Domain d,
800 bit IsReMaterializable = 1> {
801 let neverHasSideEffects = 1 in
802 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
803 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
804 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
805 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
806 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
807 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
810 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
811 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
813 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
814 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
816 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
817 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
819 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
820 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
823 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
824 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
826 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
827 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
829 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
830 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
832 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
833 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
835 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
836 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
838 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
839 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
841 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
842 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
844 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
845 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
848 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
849 "movaps\t{$src, $dst|$dst, $src}",
850 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
851 IIC_SSE_MOVA_P_MR>, VEX;
852 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
853 "movapd\t{$src, $dst|$dst, $src}",
854 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
855 IIC_SSE_MOVA_P_MR>, VEX;
856 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
857 "movups\t{$src, $dst|$dst, $src}",
858 [(store (v4f32 VR128:$src), addr:$dst)],
859 IIC_SSE_MOVU_P_MR>, VEX;
860 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
861 "movupd\t{$src, $dst|$dst, $src}",
862 [(store (v2f64 VR128:$src), addr:$dst)],
863 IIC_SSE_MOVU_P_MR>, VEX;
864 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
865 "movaps\t{$src, $dst|$dst, $src}",
866 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
867 IIC_SSE_MOVA_P_MR>, VEX;
868 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
869 "movapd\t{$src, $dst|$dst, $src}",
870 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
871 IIC_SSE_MOVA_P_MR>, VEX;
872 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
873 "movups\t{$src, $dst|$dst, $src}",
874 [(store (v8f32 VR256:$src), addr:$dst)],
875 IIC_SSE_MOVU_P_MR>, VEX;
876 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
877 "movupd\t{$src, $dst|$dst, $src}",
878 [(store (v4f64 VR256:$src), addr:$dst)],
879 IIC_SSE_MOVU_P_MR>, VEX;
882 let isCodeGenOnly = 1 in {
883 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
885 "movaps\t{$src, $dst|$dst, $src}", [],
886 IIC_SSE_MOVA_P_RR>, VEX;
887 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
889 "movapd\t{$src, $dst|$dst, $src}", [],
890 IIC_SSE_MOVA_P_RR>, VEX;
891 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
893 "movups\t{$src, $dst|$dst, $src}", [],
894 IIC_SSE_MOVU_P_RR>, VEX;
895 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
897 "movupd\t{$src, $dst|$dst, $src}", [],
898 IIC_SSE_MOVU_P_RR>, VEX;
899 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
901 "movaps\t{$src, $dst|$dst, $src}", [],
902 IIC_SSE_MOVA_P_RR>, VEX;
903 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
905 "movapd\t{$src, $dst|$dst, $src}", [],
906 IIC_SSE_MOVA_P_RR>, VEX;
907 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
909 "movups\t{$src, $dst|$dst, $src}", [],
910 IIC_SSE_MOVU_P_RR>, VEX;
911 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
913 "movupd\t{$src, $dst|$dst, $src}", [],
914 IIC_SSE_MOVU_P_RR>, VEX;
917 let Predicates = [HasAVX] in {
918 def : Pat<(v8i32 (X86vzmovl
919 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
920 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
921 def : Pat<(v4i64 (X86vzmovl
922 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
923 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
924 def : Pat<(v8f32 (X86vzmovl
925 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
926 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
927 def : Pat<(v4f64 (X86vzmovl
928 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
929 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
933 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
934 (VMOVUPSYmr addr:$dst, VR256:$src)>;
935 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
936 (VMOVUPDYmr addr:$dst, VR256:$src)>;
938 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
939 "movaps\t{$src, $dst|$dst, $src}",
940 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
942 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
943 "movapd\t{$src, $dst|$dst, $src}",
944 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
946 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
947 "movups\t{$src, $dst|$dst, $src}",
948 [(store (v4f32 VR128:$src), addr:$dst)],
950 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
951 "movupd\t{$src, $dst|$dst, $src}",
952 [(store (v2f64 VR128:$src), addr:$dst)],
956 let isCodeGenOnly = 1 in {
957 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
958 "movaps\t{$src, $dst|$dst, $src}", [],
960 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
961 "movapd\t{$src, $dst|$dst, $src}", [],
963 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
964 "movups\t{$src, $dst|$dst, $src}", [],
966 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
967 "movupd\t{$src, $dst|$dst, $src}", [],
971 let Predicates = [HasAVX] in {
972 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
973 (VMOVUPSmr addr:$dst, VR128:$src)>;
974 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
975 (VMOVUPDmr addr:$dst, VR128:$src)>;
978 let Predicates = [HasSSE1] in
979 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
980 (MOVUPSmr addr:$dst, VR128:$src)>;
981 let Predicates = [HasSSE2] in
982 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
983 (MOVUPDmr addr:$dst, VR128:$src)>;
985 // Use vmovaps/vmovups for AVX integer load/store.
986 let Predicates = [HasAVX] in {
987 // 128-bit load/store
988 def : Pat<(alignedloadv2i64 addr:$src),
989 (VMOVAPSrm addr:$src)>;
990 def : Pat<(loadv2i64 addr:$src),
991 (VMOVUPSrm addr:$src)>;
993 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
994 (VMOVAPSmr addr:$dst, VR128:$src)>;
995 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
996 (VMOVAPSmr addr:$dst, VR128:$src)>;
997 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
998 (VMOVAPSmr addr:$dst, VR128:$src)>;
999 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1000 (VMOVAPSmr addr:$dst, VR128:$src)>;
1001 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1002 (VMOVUPSmr addr:$dst, VR128:$src)>;
1003 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1004 (VMOVUPSmr addr:$dst, VR128:$src)>;
1005 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1006 (VMOVUPSmr addr:$dst, VR128:$src)>;
1007 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1008 (VMOVUPSmr addr:$dst, VR128:$src)>;
1010 // 256-bit load/store
1011 def : Pat<(alignedloadv4i64 addr:$src),
1012 (VMOVAPSYrm addr:$src)>;
1013 def : Pat<(loadv4i64 addr:$src),
1014 (VMOVUPSYrm addr:$src)>;
1015 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1016 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1017 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1018 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1019 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1020 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1021 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1022 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1023 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1024 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1025 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1026 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1027 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1028 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1029 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1030 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1033 // Use movaps / movups for SSE integer load / store (one byte shorter).
1034 // The instructions selected below are then converted to MOVDQA/MOVDQU
1035 // during the SSE domain pass.
1036 let Predicates = [HasSSE1] in {
1037 def : Pat<(alignedloadv2i64 addr:$src),
1038 (MOVAPSrm addr:$src)>;
1039 def : Pat<(loadv2i64 addr:$src),
1040 (MOVUPSrm addr:$src)>;
1042 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1043 (MOVAPSmr addr:$dst, VR128:$src)>;
1044 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1045 (MOVAPSmr addr:$dst, VR128:$src)>;
1046 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1047 (MOVAPSmr addr:$dst, VR128:$src)>;
1048 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1049 (MOVAPSmr addr:$dst, VR128:$src)>;
1050 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1051 (MOVUPSmr addr:$dst, VR128:$src)>;
1052 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1053 (MOVUPSmr addr:$dst, VR128:$src)>;
1054 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1055 (MOVUPSmr addr:$dst, VR128:$src)>;
1056 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1057 (MOVUPSmr addr:$dst, VR128:$src)>;
1060 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1061 // bits are disregarded. FIXME: Set encoding to pseudo!
1062 let neverHasSideEffects = 1 in {
1063 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1064 "movaps\t{$src, $dst|$dst, $src}", [],
1065 IIC_SSE_MOVA_P_RR>, VEX;
1066 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1067 "movapd\t{$src, $dst|$dst, $src}", [],
1068 IIC_SSE_MOVA_P_RR>, VEX;
1069 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1070 "movaps\t{$src, $dst|$dst, $src}", [],
1072 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1073 "movapd\t{$src, $dst|$dst, $src}", [],
1077 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1078 // bits are disregarded. FIXME: Set encoding to pseudo!
1079 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1080 let isCodeGenOnly = 1 in {
1081 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1082 "movaps\t{$src, $dst|$dst, $src}",
1083 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1084 IIC_SSE_MOVA_P_RM>, VEX;
1085 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1086 "movapd\t{$src, $dst|$dst, $src}",
1087 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1088 IIC_SSE_MOVA_P_RM>, VEX;
1090 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1091 "movaps\t{$src, $dst|$dst, $src}",
1092 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1094 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1095 "movapd\t{$src, $dst|$dst, $src}",
1096 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1100 //===----------------------------------------------------------------------===//
1101 // SSE 1 & 2 - Move Low packed FP Instructions
1102 //===----------------------------------------------------------------------===//
1104 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1105 SDNode psnode, SDNode pdnode, string base_opc,
1106 string asm_opr, InstrItinClass itin> {
1107 def PSrm : PI<opc, MRMSrcMem,
1108 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1109 !strconcat(base_opc, "s", asm_opr),
1112 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1113 itin, SSEPackedSingle>, TB;
1115 def PDrm : PI<opc, MRMSrcMem,
1116 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1117 !strconcat(base_opc, "d", asm_opr),
1118 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1119 (scalar_to_vector (loadf64 addr:$src2)))))],
1120 itin, SSEPackedDouble>, TB, OpSize;
1123 let AddedComplexity = 20 in {
1124 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1125 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1126 IIC_SSE_MOV_LH>, VEX_4V;
1128 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1129 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1130 "\t{$src2, $dst|$dst, $src2}",
1134 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1135 "movlps\t{$src, $dst|$dst, $src}",
1136 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1137 (iPTR 0))), addr:$dst)],
1138 IIC_SSE_MOV_LH>, VEX;
1139 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1140 "movlpd\t{$src, $dst|$dst, $src}",
1141 [(store (f64 (vector_extract (v2f64 VR128:$src),
1142 (iPTR 0))), addr:$dst)],
1143 IIC_SSE_MOV_LH>, VEX;
1144 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1145 "movlps\t{$src, $dst|$dst, $src}",
1146 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1147 (iPTR 0))), addr:$dst)],
1149 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1150 "movlpd\t{$src, $dst|$dst, $src}",
1151 [(store (f64 (vector_extract (v2f64 VR128:$src),
1152 (iPTR 0))), addr:$dst)],
1155 let Predicates = [HasAVX] in {
1156 // Shuffle with VMOVLPS
1157 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1158 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1159 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1160 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1162 // Shuffle with VMOVLPD
1163 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1164 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1165 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1166 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1169 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1171 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1172 def : Pat<(store (v4i32 (X86Movlps
1173 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1174 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1175 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1177 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1178 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1180 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1183 let Predicates = [HasSSE1] in {
1184 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1185 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1186 (iPTR 0))), addr:$src1),
1187 (MOVLPSmr addr:$src1, VR128:$src2)>;
1189 // Shuffle with MOVLPS
1190 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1191 (MOVLPSrm VR128:$src1, addr:$src2)>;
1192 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1193 (MOVLPSrm VR128:$src1, addr:$src2)>;
1194 def : Pat<(X86Movlps VR128:$src1,
1195 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1196 (MOVLPSrm VR128:$src1, addr:$src2)>;
1199 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1201 (MOVLPSmr addr:$src1, VR128:$src2)>;
1202 def : Pat<(store (v4i32 (X86Movlps
1203 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1205 (MOVLPSmr addr:$src1, VR128:$src2)>;
1208 let Predicates = [HasSSE2] in {
1209 // Shuffle with MOVLPD
1210 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1211 (MOVLPDrm VR128:$src1, addr:$src2)>;
1212 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1213 (MOVLPDrm VR128:$src1, addr:$src2)>;
1216 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1218 (MOVLPDmr addr:$src1, VR128:$src2)>;
1219 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1221 (MOVLPDmr addr:$src1, VR128:$src2)>;
1224 //===----------------------------------------------------------------------===//
1225 // SSE 1 & 2 - Move Hi packed FP Instructions
1226 //===----------------------------------------------------------------------===//
1228 let AddedComplexity = 20 in {
1229 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1230 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 IIC_SSE_MOV_LH>, VEX_4V;
1233 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1234 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1235 "\t{$src2, $dst|$dst, $src2}",
1239 // v2f64 extract element 1 is always custom lowered to unpack high to low
1240 // and extract element 0 so the non-store version isn't too horrible.
1241 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1242 "movhps\t{$src, $dst|$dst, $src}",
1243 [(store (f64 (vector_extract
1244 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1245 (bc_v2f64 (v4f32 VR128:$src))),
1246 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1247 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1248 "movhpd\t{$src, $dst|$dst, $src}",
1249 [(store (f64 (vector_extract
1250 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1251 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1252 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1253 "movhps\t{$src, $dst|$dst, $src}",
1254 [(store (f64 (vector_extract
1255 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1256 (bc_v2f64 (v4f32 VR128:$src))),
1257 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1258 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1259 "movhpd\t{$src, $dst|$dst, $src}",
1260 [(store (f64 (vector_extract
1261 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1262 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1264 let Predicates = [HasAVX] in {
1266 def : Pat<(X86Movlhps VR128:$src1,
1267 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1268 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1269 def : Pat<(X86Movlhps VR128:$src1,
1270 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1271 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1273 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1274 // is during lowering, where it's not possible to recognize the load fold
1275 // cause it has two uses through a bitcast. One use disappears at isel time
1276 // and the fold opportunity reappears.
1277 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1278 (scalar_to_vector (loadf64 addr:$src2)))),
1279 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1282 let Predicates = [HasSSE1] in {
1284 def : Pat<(X86Movlhps VR128:$src1,
1285 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1286 (MOVHPSrm VR128:$src1, addr:$src2)>;
1287 def : Pat<(X86Movlhps VR128:$src1,
1288 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1289 (MOVHPSrm VR128:$src1, addr:$src2)>;
1292 let Predicates = [HasSSE2] in {
1293 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1294 // is during lowering, where it's not possible to recognize the load fold
1295 // cause it has two uses through a bitcast. One use disappears at isel time
1296 // and the fold opportunity reappears.
1297 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1298 (scalar_to_vector (loadf64 addr:$src2)))),
1299 (MOVHPDrm VR128:$src1, addr:$src2)>;
1302 //===----------------------------------------------------------------------===//
1303 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1304 //===----------------------------------------------------------------------===//
1306 let AddedComplexity = 20 in {
1307 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1308 (ins VR128:$src1, VR128:$src2),
1309 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1311 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1314 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1315 (ins VR128:$src1, VR128:$src2),
1316 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1318 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1322 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1323 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1324 (ins VR128:$src1, VR128:$src2),
1325 "movlhps\t{$src2, $dst|$dst, $src2}",
1327 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1329 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1330 (ins VR128:$src1, VR128:$src2),
1331 "movhlps\t{$src2, $dst|$dst, $src2}",
1333 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1337 let Predicates = [HasAVX] in {
1339 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1340 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1341 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1342 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1345 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1349 let Predicates = [HasSSE1] in {
1351 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1352 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1353 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1354 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1357 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1358 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1361 //===----------------------------------------------------------------------===//
1362 // SSE 1 & 2 - Conversion Instructions
1363 //===----------------------------------------------------------------------===//
1365 def SSE_CVT_PD : OpndItins<
1366 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1369 def SSE_CVT_PS : OpndItins<
1370 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1373 def SSE_CVT_Scalar : OpndItins<
1374 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1377 def SSE_CVT_SS2SI_32 : OpndItins<
1378 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1381 def SSE_CVT_SS2SI_64 : OpndItins<
1382 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1385 def SSE_CVT_SD2SI : OpndItins<
1386 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1389 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1390 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1391 string asm, OpndItins itins> {
1392 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1393 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1395 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1396 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1400 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1401 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1402 string asm, Domain d, OpndItins itins> {
1403 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1404 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1406 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1407 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1411 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1412 X86MemOperand x86memop, string asm> {
1413 let neverHasSideEffects = 1 in {
1414 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1415 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1417 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1418 (ins DstRC:$src1, x86memop:$src),
1419 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1420 } // neverHasSideEffects = 1
1423 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1424 "cvttss2si\t{$src, $dst|$dst, $src}",
1427 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1428 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1430 XS, VEX, VEX_W, VEX_LIG;
1431 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1432 "cvttsd2si\t{$src, $dst|$dst, $src}",
1435 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1436 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1438 XD, VEX, VEX_W, VEX_LIG;
1440 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1441 // register, but the same isn't true when only using memory operands,
1442 // provide other assembly "l" and "q" forms to address this explicitly
1443 // where appropriate to do so.
1444 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1445 XS, VEX_4V, VEX_LIG;
1446 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1447 XS, VEX_4V, VEX_W, VEX_LIG;
1448 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1449 XD, VEX_4V, VEX_LIG;
1450 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1451 XD, VEX_4V, VEX_W, VEX_LIG;
1453 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1454 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1455 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1456 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1458 let Predicates = [HasAVX], AddedComplexity = 1 in {
1459 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1460 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1461 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1462 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1463 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1464 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1466 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1468 def : Pat<(f32 (sint_to_fp GR32:$src)),
1469 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1470 def : Pat<(f32 (sint_to_fp GR64:$src)),
1471 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1472 def : Pat<(f64 (sint_to_fp GR32:$src)),
1473 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1474 def : Pat<(f64 (sint_to_fp GR64:$src)),
1475 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1478 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1479 "cvttss2si\t{$src, $dst|$dst, $src}",
1480 SSE_CVT_SS2SI_32>, XS;
1481 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1482 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1483 SSE_CVT_SS2SI_64>, XS, REX_W;
1484 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1485 "cvttsd2si\t{$src, $dst|$dst, $src}",
1487 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1488 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1489 SSE_CVT_SD2SI>, XD, REX_W;
1490 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1491 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1492 SSE_CVT_Scalar>, XS;
1493 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1494 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1495 SSE_CVT_Scalar>, XS, REX_W;
1496 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1497 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1498 SSE_CVT_Scalar>, XD;
1499 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1500 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1501 SSE_CVT_Scalar>, XD, REX_W;
1503 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1504 // and/or XMM operand(s).
1506 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1507 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1508 string asm, OpndItins itins> {
1509 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1510 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1511 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1512 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1513 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1514 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1517 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1518 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1519 PatFrag ld_frag, string asm, OpndItins itins,
1521 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1523 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1524 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1525 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1527 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1528 (ins DstRC:$src1, x86memop:$src2),
1530 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1531 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1532 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1536 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1537 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
1538 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1539 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1540 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
1541 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1543 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1544 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1545 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1546 sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1549 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1550 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1551 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1552 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1553 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1554 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1556 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1557 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1558 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1559 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1560 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1561 SSE_CVT_Scalar, 0>, XD,
1564 let Constraints = "$src1 = $dst" in {
1565 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1566 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1567 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1568 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1569 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1570 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1571 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1572 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1573 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1574 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1575 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1576 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1581 // Aliases for intrinsics
1582 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1583 ssmem, sse_load_f32, "cvttss2si",
1584 SSE_CVT_SS2SI_32>, XS, VEX;
1585 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1586 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1587 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1589 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1590 sdmem, sse_load_f64, "cvttsd2si",
1591 SSE_CVT_SD2SI>, XD, VEX;
1592 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1593 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1594 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1596 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1597 ssmem, sse_load_f32, "cvttss2si",
1598 SSE_CVT_SS2SI_32>, XS;
1599 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1600 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1601 "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
1602 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1603 sdmem, sse_load_f64, "cvttsd2si",
1605 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1606 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1607 "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1609 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1610 ssmem, sse_load_f32, "cvtss2si{l}",
1611 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1612 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1613 ssmem, sse_load_f32, "cvtss2si{q}",
1614 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1616 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1617 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1618 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1619 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1621 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1622 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1623 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1627 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1628 ssmem, sse_load_f32, "cvtss2si{l}",
1629 SSE_CVT_SS2SI_32>, XS;
1630 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1631 ssmem, sse_load_f32, "cvtss2si{q}",
1632 SSE_CVT_SS2SI_64>, XS, REX_W;
1634 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1635 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1636 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1637 SSEPackedSingle, SSE_CVT_PS>, TB,
1638 Requires<[HasSSE2]>;
1643 // Convert scalar double to scalar single
1644 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1645 (ins FR64:$src1, FR64:$src2),
1646 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1647 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1649 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1650 (ins FR64:$src1, f64mem:$src2),
1651 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1652 [], IIC_SSE_CVT_Scalar_RM>,
1653 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1655 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1658 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1659 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1660 [(set FR32:$dst, (fround FR64:$src))],
1661 IIC_SSE_CVT_Scalar_RR>;
1662 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1663 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1664 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1665 IIC_SSE_CVT_Scalar_RM>,
1667 Requires<[HasSSE2, OptForSize]>;
1669 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1670 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1673 let Constraints = "$src1 = $dst" in
1674 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1675 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1676 SSE_CVT_Scalar>, XS;
1678 // Convert scalar single to scalar double
1679 // SSE2 instructions with XS prefix
1680 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1681 (ins FR32:$src1, FR32:$src2),
1682 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1683 [], IIC_SSE_CVT_Scalar_RR>,
1684 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1686 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1687 (ins FR32:$src1, f32mem:$src2),
1688 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1689 [], IIC_SSE_CVT_Scalar_RM>,
1690 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1692 let Predicates = [HasAVX] in {
1693 def : Pat<(f64 (fextend FR32:$src)),
1694 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1695 def : Pat<(fextend (loadf32 addr:$src)),
1696 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1697 def : Pat<(extloadf32 addr:$src),
1698 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1701 def : Pat<(extloadf32 addr:$src),
1702 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1703 Requires<[HasAVX, OptForSpeed]>;
1705 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1706 "cvtss2sd\t{$src, $dst|$dst, $src}",
1707 [(set FR64:$dst, (fextend FR32:$src))],
1708 IIC_SSE_CVT_Scalar_RR>, XS,
1709 Requires<[HasSSE2]>;
1710 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1711 "cvtss2sd\t{$src, $dst|$dst, $src}",
1712 [(set FR64:$dst, (extloadf32 addr:$src))],
1713 IIC_SSE_CVT_Scalar_RM>, XS,
1714 Requires<[HasSSE2, OptForSize]>;
1716 // extload f32 -> f64. This matches load+fextend because we have a hack in
1717 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1719 // Since these loads aren't folded into the fextend, we have to match it
1721 def : Pat<(fextend (loadf32 addr:$src)),
1722 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1723 def : Pat<(extloadf32 addr:$src),
1724 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1726 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1727 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1728 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1729 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1731 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V,
1733 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1734 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1735 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1736 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1737 (load addr:$src2)))],
1738 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V,
1740 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1741 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1742 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1743 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1744 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1746 IIC_SSE_CVT_Scalar_RR>, XS,
1747 Requires<[HasSSE2]>;
1748 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1749 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1750 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1751 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1752 (load addr:$src2)))],
1753 IIC_SSE_CVT_Scalar_RM>, XS,
1754 Requires<[HasSSE2]>;
1757 // Convert packed single/double fp to doubleword
1758 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1759 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1760 IIC_SSE_CVT_PS_RR>, VEX;
1761 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1762 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1763 IIC_SSE_CVT_PS_RM>, VEX;
1764 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1765 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1766 IIC_SSE_CVT_PS_RR>, VEX;
1767 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1768 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1769 IIC_SSE_CVT_PS_RM>, VEX;
1770 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1771 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1773 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1774 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1777 let Predicates = [HasAVX] in {
1778 def : Pat<(int_x86_sse2_cvtps2dq VR128:$src),
1779 (VCVTPS2DQrr VR128:$src)>;
1780 def : Pat<(int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)),
1781 (VCVTPS2DQrm addr:$src)>;
1784 let Predicates = [HasSSE2] in {
1785 def : Pat<(int_x86_sse2_cvtps2dq VR128:$src),
1786 (CVTPS2DQrr VR128:$src)>;
1787 def : Pat<(int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)),
1788 (CVTPS2DQrm addr:$src)>;
1791 // Convert Packed Double FP to Packed DW Integers
1792 let Predicates = [HasAVX] in {
1793 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1794 // register, but the same isn't true when using memory operands instead.
1795 // Provide other assembly rr and rm forms to address this explicitly.
1796 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1797 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1800 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1801 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1802 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1803 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1806 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1807 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX;
1808 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1809 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1810 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1811 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1814 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1815 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1817 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1818 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1821 let Predicates = [HasAVX] in {
1822 def : Pat<(int_x86_sse2_cvtpd2dq VR128:$src),
1823 (VCVTPD2DQrr VR128:$src)>;
1824 def : Pat<(int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)),
1825 (VCVTPD2DQXrm addr:$src)>;
1828 let Predicates = [HasSSE2] in {
1829 def : Pat<(int_x86_sse2_cvtpd2dq VR128:$src),
1830 (CVTPD2DQrr VR128:$src)>;
1831 def : Pat<(int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)),
1832 (CVTPD2DQrm addr:$src)>;
1835 // Convert with truncation packed single/double fp to doubleword
1836 // SSE2 packed instructions with XS prefix
1837 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1838 "cvttps2dq\t{$src, $dst|$dst, $src}",
1840 (int_x86_sse2_cvttps2dq VR128:$src))],
1841 IIC_SSE_CVT_PS_RR>, VEX;
1842 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1843 "cvttps2dq\t{$src, $dst|$dst, $src}",
1844 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1845 (memopv4f32 addr:$src)))],
1846 IIC_SSE_CVT_PS_RM>, VEX;
1847 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1848 "cvttps2dq\t{$src, $dst|$dst, $src}",
1850 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1851 IIC_SSE_CVT_PS_RR>, VEX;
1852 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1853 "cvttps2dq\t{$src, $dst|$dst, $src}",
1854 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1855 (memopv8f32 addr:$src)))],
1856 IIC_SSE_CVT_PS_RM>, VEX;
1858 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1859 "cvttps2dq\t{$src, $dst|$dst, $src}",
1861 (int_x86_sse2_cvttps2dq VR128:$src))],
1863 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1864 "cvttps2dq\t{$src, $dst|$dst, $src}",
1866 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1869 let Predicates = [HasAVX] in {
1870 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1871 (VCVTDQ2PSrr VR128:$src)>;
1872 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1873 (VCVTDQ2PSrm addr:$src)>;
1875 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1876 (VCVTDQ2PSrr VR128:$src)>;
1877 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1878 (VCVTDQ2PSrm addr:$src)>;
1880 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1881 (VCVTTPS2DQrr VR128:$src)>;
1882 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1883 (VCVTTPS2DQrm addr:$src)>;
1885 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1886 (VCVTDQ2PSYrr VR256:$src)>;
1887 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1888 (VCVTDQ2PSYrm addr:$src)>;
1890 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1891 (VCVTTPS2DQYrr VR256:$src)>;
1892 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1893 (VCVTTPS2DQYrm addr:$src)>;
1896 let Predicates = [HasSSE2] in {
1897 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1898 (CVTDQ2PSrr VR128:$src)>;
1899 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1900 (CVTDQ2PSrm addr:$src)>;
1902 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1903 (CVTDQ2PSrr VR128:$src)>;
1904 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1905 (CVTDQ2PSrm addr:$src)>;
1907 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1908 (CVTTPS2DQrr VR128:$src)>;
1909 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1910 (CVTTPS2DQrm addr:$src)>;
1913 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1914 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1916 (int_x86_sse2_cvttpd2dq VR128:$src))],
1917 IIC_SSE_CVT_PD_RR>, VEX;
1919 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1920 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1921 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1923 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1924 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1925 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1926 (memopv2f64 addr:$src)))],
1929 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1930 // register, but the same isn't true when using memory operands instead.
1931 // Provide other assembly rr and rm forms to address this explicitly.
1934 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1935 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1936 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1937 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1938 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1939 (memopv2f64 addr:$src)))],
1940 IIC_SSE_CVT_PD_RM>, VEX;
1943 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1944 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
1945 IIC_SSE_CVT_PD_RR>, VEX;
1946 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1947 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
1948 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1949 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1950 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1952 let Predicates = [HasAVX] in {
1953 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1954 (VCVTTPD2DQYrr VR256:$src)>;
1955 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1956 (VCVTTPD2DQYrm addr:$src)>;
1957 } // Predicates = [HasAVX]
1959 // Convert packed single to packed double
1960 let Predicates = [HasAVX] in {
1961 // SSE2 instructions without OpSize prefix
1962 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1963 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1964 IIC_SSE_CVT_PD_RR>, TB, VEX;
1965 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1966 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1967 IIC_SSE_CVT_PD_RM>, TB, VEX;
1968 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1969 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1970 IIC_SSE_CVT_PD_RR>, TB, VEX;
1971 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1972 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
1973 IIC_SSE_CVT_PD_RM>, TB, VEX;
1976 let Predicates = [HasSSE2] in {
1977 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1978 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
1979 IIC_SSE_CVT_PD_RR>, TB;
1980 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1981 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
1982 IIC_SSE_CVT_PD_RM>, TB;
1985 let Predicates = [HasAVX] in {
1986 def : Pat<(int_x86_sse2_cvtps2pd VR128:$src),
1987 (VCVTPS2PDrr VR128:$src)>;
1990 let Predicates = [HasSSE2] in {
1991 def : Pat<(int_x86_sse2_cvtps2pd VR128:$src),
1992 (CVTPS2PDrr VR128:$src)>;
1995 // Convert Packed DW Integers to Packed Double FP
1996 let Predicates = [HasAVX] in {
1997 def VCVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1998 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1999 def VCVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2000 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2001 def VCVTDQ2PDYrm : SSDI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2002 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2003 def VCVTDQ2PDYrr : SSDI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2004 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2007 def CVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2008 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2010 def CVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2011 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2014 // 128 bit register conversion intrinsics
2015 let Predicates = [HasAVX] in
2016 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2017 (VCVTDQ2PDrr VR128:$src)>;
2019 let Predicates = [HasSSE2] in
2020 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2021 (CVTDQ2PDrr VR128:$src)>;
2023 // AVX 256-bit register conversion intrinsics
2024 let Predicates = [HasAVX] in {
2025 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
2026 (VCVTDQ2PDYrr VR128:$src)>;
2027 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
2028 (VCVTDQ2PDYrm addr:$src)>;
2030 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
2031 (VCVTPD2DQYrr VR256:$src)>;
2032 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
2033 (VCVTPD2DQYrm addr:$src)>;
2035 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2036 (VCVTDQ2PDYrr VR128:$src)>;
2037 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2038 (VCVTDQ2PDYrm addr:$src)>;
2039 } // Predicates = [HasAVX]
2041 // Convert packed double to packed single
2042 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2043 // register, but the same isn't true when using memory operands instead.
2044 // Provide other assembly rr and rm forms to address this explicitly.
2045 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2046 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2047 IIC_SSE_CVT_PD_RR>, VEX;
2050 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2051 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2052 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2053 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2054 IIC_SSE_CVT_PD_RM>, VEX;
2057 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2058 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2059 IIC_SSE_CVT_PD_RR>, VEX;
2060 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2061 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2062 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2063 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2064 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2066 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2067 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2069 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2070 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2074 let Predicates = [HasAVX] in {
2075 def : Pat<(int_x86_sse2_cvtpd2ps VR128:$src),
2076 (VCVTPD2PSrr VR128:$src)>;
2077 def : Pat<(int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)),
2078 (VCVTPD2PSXrm addr:$src)>;
2081 let Predicates = [HasSSE2] in {
2082 def : Pat<(int_x86_sse2_cvtpd2ps VR128:$src),
2083 (CVTPD2PSrr VR128:$src)>;
2084 def : Pat<(int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)),
2085 (CVTPD2PSrm addr:$src)>;
2088 // AVX 256-bit register conversion intrinsics
2089 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2090 // whenever possible to avoid declaring two versions of each one.
2091 let Predicates = [HasAVX] in {
2092 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2093 (VCVTDQ2PSYrr VR256:$src)>;
2094 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2095 (VCVTDQ2PSYrm addr:$src)>;
2097 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2098 (VCVTPD2PSYrr VR256:$src)>;
2099 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2100 (VCVTPD2PSYrm addr:$src)>;
2102 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2103 (VCVTPS2DQYrr VR256:$src)>;
2104 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2105 (VCVTPS2DQYrm addr:$src)>;
2107 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2108 (VCVTPS2PDYrr VR128:$src)>;
2109 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2110 (VCVTPS2PDYrm addr:$src)>;
2112 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2113 (VCVTTPD2DQYrr VR256:$src)>;
2114 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2115 (VCVTTPD2DQYrm addr:$src)>;
2117 // Match fround and fextend for 128/256-bit conversions
2118 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2119 (VCVTPD2PSYrr VR256:$src)>;
2120 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2121 (VCVTPD2PSYrm addr:$src)>;
2123 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2124 (VCVTPS2PDYrr VR128:$src)>;
2125 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2126 (VCVTPS2PDYrm addr:$src)>;
2129 //===----------------------------------------------------------------------===//
2130 // SSE 1 & 2 - Compare Instructions
2131 //===----------------------------------------------------------------------===//
2133 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2134 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2135 Operand CC, SDNode OpNode, ValueType VT,
2136 PatFrag ld_frag, string asm, string asm_alt,
2138 def rr : SIi8<0xC2, MRMSrcReg,
2139 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2140 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2142 def rm : SIi8<0xC2, MRMSrcMem,
2143 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2144 [(set RC:$dst, (OpNode (VT RC:$src1),
2145 (ld_frag addr:$src2), imm:$cc))],
2148 // Accept explicit immediate argument form instead of comparison code.
2149 let neverHasSideEffects = 1 in {
2150 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2151 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2152 IIC_SSE_ALU_F32S_RR>;
2154 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2155 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2156 IIC_SSE_ALU_F32S_RM>;
2160 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2161 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2162 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2164 XS, VEX_4V, VEX_LIG;
2165 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2166 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2167 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2168 SSE_ALU_F32S>, // same latency as 32 bit compare
2169 XD, VEX_4V, VEX_LIG;
2171 let Constraints = "$src1 = $dst" in {
2172 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2173 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2174 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2176 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2177 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2178 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2179 SSE_ALU_F32S>, // same latency as 32 bit compare
2183 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2184 Intrinsic Int, string asm, OpndItins itins> {
2185 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2186 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2187 [(set VR128:$dst, (Int VR128:$src1,
2188 VR128:$src, imm:$cc))],
2190 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2191 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2192 [(set VR128:$dst, (Int VR128:$src1,
2193 (load addr:$src), imm:$cc))],
2197 // Aliases to match intrinsics which expect XMM operand(s).
2198 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2199 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2202 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2203 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2204 SSE_ALU_F32S>, // same latency as f32
2206 let Constraints = "$src1 = $dst" in {
2207 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2208 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2210 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2211 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2212 SSE_ALU_F32S>, // same latency as f32
2217 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2218 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2219 ValueType vt, X86MemOperand x86memop,
2220 PatFrag ld_frag, string OpcodeStr, Domain d> {
2221 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2222 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2223 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2224 IIC_SSE_COMIS_RR, d>;
2225 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2226 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2227 [(set EFLAGS, (OpNode (vt RC:$src1),
2228 (ld_frag addr:$src2)))],
2229 IIC_SSE_COMIS_RM, d>;
2232 let Defs = [EFLAGS] in {
2233 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2234 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2235 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2236 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2238 let Pattern = []<dag> in {
2239 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2240 "comiss", SSEPackedSingle>, TB, VEX,
2242 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2243 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2247 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2248 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2249 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2250 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2252 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2253 load, "comiss", SSEPackedSingle>, TB, VEX;
2254 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2255 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2256 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2257 "ucomiss", SSEPackedSingle>, TB;
2258 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2259 "ucomisd", SSEPackedDouble>, TB, OpSize;
2261 let Pattern = []<dag> in {
2262 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2263 "comiss", SSEPackedSingle>, TB;
2264 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2265 "comisd", SSEPackedDouble>, TB, OpSize;
2268 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2269 load, "ucomiss", SSEPackedSingle>, TB;
2270 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2271 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2273 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2274 "comiss", SSEPackedSingle>, TB;
2275 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2276 "comisd", SSEPackedDouble>, TB, OpSize;
2277 } // Defs = [EFLAGS]
2279 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2280 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2281 Operand CC, Intrinsic Int, string asm,
2282 string asm_alt, Domain d> {
2283 def rri : PIi8<0xC2, MRMSrcReg,
2284 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2285 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2286 IIC_SSE_CMPP_RR, d>;
2287 def rmi : PIi8<0xC2, MRMSrcMem,
2288 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2289 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2290 IIC_SSE_CMPP_RM, d>;
2292 // Accept explicit immediate argument form instead of comparison code.
2293 let neverHasSideEffects = 1 in {
2294 def rri_alt : PIi8<0xC2, MRMSrcReg,
2295 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2296 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2297 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2298 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2299 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2303 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2304 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2305 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2306 SSEPackedSingle>, TB, VEX_4V;
2307 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2308 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2309 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2310 SSEPackedDouble>, TB, OpSize, VEX_4V;
2311 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2312 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2313 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2314 SSEPackedSingle>, TB, VEX_4V;
2315 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2316 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2317 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2318 SSEPackedDouble>, TB, OpSize, VEX_4V;
2319 let Constraints = "$src1 = $dst" in {
2320 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2321 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2322 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2323 SSEPackedSingle>, TB;
2324 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2325 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2326 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2327 SSEPackedDouble>, TB, OpSize;
2330 let Predicates = [HasAVX] in {
2331 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2332 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2333 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2334 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2335 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2336 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2337 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2338 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2340 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2341 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2342 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2343 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2344 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2345 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2346 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2347 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2350 let Predicates = [HasSSE1] in {
2351 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2352 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2353 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2354 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2357 let Predicates = [HasSSE2] in {
2358 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2359 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2360 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2361 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2364 //===----------------------------------------------------------------------===//
2365 // SSE 1 & 2 - Shuffle Instructions
2366 //===----------------------------------------------------------------------===//
2368 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2369 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2370 ValueType vt, string asm, PatFrag mem_frag,
2371 Domain d, bit IsConvertibleToThreeAddress = 0> {
2372 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2373 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2374 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2375 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2376 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2377 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2378 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2379 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2380 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2383 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2384 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2385 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2386 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2387 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2388 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2389 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2390 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2391 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2392 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2393 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2394 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2396 let Constraints = "$src1 = $dst" in {
2397 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2398 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2399 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2401 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2402 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2403 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2407 let Predicates = [HasAVX] in {
2408 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2409 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2410 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2411 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2412 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2414 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2415 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2416 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2417 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2418 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2421 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2422 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2423 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2424 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2425 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2427 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2428 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2429 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2430 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2431 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2434 let Predicates = [HasSSE1] in {
2435 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2436 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2437 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2438 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2439 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2442 let Predicates = [HasSSE2] in {
2443 // Generic SHUFPD patterns
2444 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2445 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2446 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2447 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2448 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2451 //===----------------------------------------------------------------------===//
2452 // SSE 1 & 2 - Unpack Instructions
2453 //===----------------------------------------------------------------------===//
2455 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2456 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2457 PatFrag mem_frag, RegisterClass RC,
2458 X86MemOperand x86memop, string asm,
2460 def rr : PI<opc, MRMSrcReg,
2461 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2463 (vt (OpNode RC:$src1, RC:$src2)))],
2465 def rm : PI<opc, MRMSrcMem,
2466 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2468 (vt (OpNode RC:$src1,
2469 (mem_frag addr:$src2))))],
2473 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2474 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2475 SSEPackedSingle>, TB, VEX_4V;
2476 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2477 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2478 SSEPackedDouble>, TB, OpSize, VEX_4V;
2479 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2480 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2481 SSEPackedSingle>, TB, VEX_4V;
2482 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2483 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2484 SSEPackedDouble>, TB, OpSize, VEX_4V;
2486 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2487 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2488 SSEPackedSingle>, TB, VEX_4V;
2489 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2490 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2491 SSEPackedDouble>, TB, OpSize, VEX_4V;
2492 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2493 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2494 SSEPackedSingle>, TB, VEX_4V;
2495 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2496 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2497 SSEPackedDouble>, TB, OpSize, VEX_4V;
2499 let Constraints = "$src1 = $dst" in {
2500 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2501 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2502 SSEPackedSingle>, TB;
2503 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2504 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2505 SSEPackedDouble>, TB, OpSize;
2506 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2507 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2508 SSEPackedSingle>, TB;
2509 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2510 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2511 SSEPackedDouble>, TB, OpSize;
2512 } // Constraints = "$src1 = $dst"
2514 let Predicates = [HasAVX], AddedComplexity = 1 in {
2515 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2516 // problem is during lowering, where it's not possible to recognize the load
2517 // fold cause it has two uses through a bitcast. One use disappears at isel
2518 // time and the fold opportunity reappears.
2519 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2520 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2523 let Predicates = [HasSSE2] in {
2524 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2525 // problem is during lowering, where it's not possible to recognize the load
2526 // fold cause it has two uses through a bitcast. One use disappears at isel
2527 // time and the fold opportunity reappears.
2528 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2529 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2532 //===----------------------------------------------------------------------===//
2533 // SSE 1 & 2 - Extract Floating-Point Sign mask
2534 //===----------------------------------------------------------------------===//
2536 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2537 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2539 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2540 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2541 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2542 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2543 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2544 IIC_SSE_MOVMSK, d>, REX_W;
2547 let Predicates = [HasAVX] in {
2548 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2549 "movmskps", SSEPackedSingle>, TB, VEX;
2550 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2551 "movmskpd", SSEPackedDouble>, TB,
2553 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2554 "movmskps", SSEPackedSingle>, TB, VEX;
2555 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2556 "movmskpd", SSEPackedDouble>, TB,
2559 def : Pat<(i32 (X86fgetsign FR32:$src)),
2560 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2561 def : Pat<(i64 (X86fgetsign FR32:$src)),
2562 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2563 def : Pat<(i32 (X86fgetsign FR64:$src)),
2564 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2565 def : Pat<(i64 (X86fgetsign FR64:$src)),
2566 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2569 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2570 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2571 SSEPackedSingle>, TB, VEX;
2572 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2573 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2574 SSEPackedDouble>, TB,
2576 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2577 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2578 SSEPackedSingle>, TB, VEX;
2579 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2580 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2581 SSEPackedDouble>, TB,
2585 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2586 SSEPackedSingle>, TB;
2587 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2588 SSEPackedDouble>, TB, OpSize;
2590 def : Pat<(i32 (X86fgetsign FR32:$src)),
2591 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2592 Requires<[HasSSE1]>;
2593 def : Pat<(i64 (X86fgetsign FR32:$src)),
2594 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2595 Requires<[HasSSE1]>;
2596 def : Pat<(i32 (X86fgetsign FR64:$src)),
2597 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2598 Requires<[HasSSE2]>;
2599 def : Pat<(i64 (X86fgetsign FR64:$src)),
2600 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2601 Requires<[HasSSE2]>;
2603 //===---------------------------------------------------------------------===//
2604 // SSE2 - Packed Integer Logical Instructions
2605 //===---------------------------------------------------------------------===//
2607 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2609 /// PDI_binop_rm - Simple SSE2 binary operator.
2610 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2611 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2612 X86MemOperand x86memop,
2614 bit IsCommutable = 0,
2616 let isCommutable = IsCommutable in
2617 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2618 (ins RC:$src1, RC:$src2),
2620 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2621 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2622 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2623 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2624 (ins RC:$src1, x86memop:$src2),
2626 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2627 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2628 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2629 (bitconvert (memop_frag addr:$src2)))))],
2632 } // ExeDomain = SSEPackedInt
2634 // These are ordered here for pattern ordering requirements with the fp versions
2636 let Predicates = [HasAVX] in {
2637 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2638 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2639 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2640 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2641 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2642 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2643 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2644 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2647 let Constraints = "$src1 = $dst" in {
2648 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2649 i128mem, SSE_BIT_ITINS_P, 1>;
2650 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2651 i128mem, SSE_BIT_ITINS_P, 1>;
2652 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2653 i128mem, SSE_BIT_ITINS_P, 1>;
2654 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2655 i128mem, SSE_BIT_ITINS_P, 0>;
2656 } // Constraints = "$src1 = $dst"
2658 let Predicates = [HasAVX2] in {
2659 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2660 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2661 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2662 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2663 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2664 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2665 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2666 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2669 //===----------------------------------------------------------------------===//
2670 // SSE 1 & 2 - Logical Instructions
2671 //===----------------------------------------------------------------------===//
2673 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2675 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2676 SDNode OpNode, OpndItins itins> {
2677 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2678 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2681 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2682 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2685 let Constraints = "$src1 = $dst" in {
2686 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2687 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2690 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2691 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2696 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2697 let mayLoad = 0 in {
2698 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2700 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2702 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2706 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2707 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2710 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2712 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2714 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2715 // are all promoted to v2i64, and the patterns are covered by the int
2716 // version. This is needed in SSE only, because v2i64 isn't supported on
2717 // SSE1, but only on SSE2.
2718 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2719 !strconcat(OpcodeStr, "ps"), f128mem, [],
2720 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2721 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2723 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2724 !strconcat(OpcodeStr, "pd"), f128mem,
2725 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2726 (bc_v2i64 (v2f64 VR128:$src2))))],
2727 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2728 (memopv2i64 addr:$src2)))], 0>,
2730 let Constraints = "$src1 = $dst" in {
2731 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2732 !strconcat(OpcodeStr, "ps"), f128mem,
2733 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2734 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2735 (memopv2i64 addr:$src2)))]>, TB;
2737 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2738 !strconcat(OpcodeStr, "pd"), f128mem,
2739 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2740 (bc_v2i64 (v2f64 VR128:$src2))))],
2741 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2742 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2746 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2748 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2750 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2751 !strconcat(OpcodeStr, "ps"), f256mem,
2752 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2753 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2754 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2756 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2757 !strconcat(OpcodeStr, "pd"), f256mem,
2758 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2759 (bc_v4i64 (v4f64 VR256:$src2))))],
2760 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2761 (memopv4i64 addr:$src2)))], 0>,
2765 // AVX 256-bit packed logical ops forms
2766 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2767 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2768 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2769 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2771 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2772 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2773 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2774 let isCommutable = 0 in
2775 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2777 //===----------------------------------------------------------------------===//
2778 // SSE 1 & 2 - Arithmetic Instructions
2779 //===----------------------------------------------------------------------===//
2781 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2784 /// In addition, we also have a special variant of the scalar form here to
2785 /// represent the associated intrinsic operation. This form is unlike the
2786 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2787 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2789 /// These three forms can each be reg+reg or reg+mem.
2792 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2794 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2797 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2798 OpNode, FR32, f32mem,
2799 itins.s, Is2Addr>, XS;
2800 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2801 OpNode, FR64, f64mem,
2802 itins.d, Is2Addr>, XD;
2805 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2808 let mayLoad = 0 in {
2809 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2810 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2812 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2813 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2818 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2821 let mayLoad = 0 in {
2822 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2823 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2825 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2826 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2831 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2834 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2835 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2836 itins.s, Is2Addr>, XS;
2837 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2838 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2839 itins.d, Is2Addr>, XD;
2842 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2845 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2846 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2847 SSEPackedSingle, itins.s, Is2Addr>,
2850 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2851 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2852 SSEPackedDouble, itins.d, Is2Addr>,
2856 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2858 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2859 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2860 SSEPackedSingle, itins.s, 0>, TB;
2862 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2863 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2864 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2867 // Binary Arithmetic instructions
2868 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2869 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2871 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2872 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2874 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2875 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2877 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2878 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2881 let isCommutable = 0 in {
2882 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2883 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2885 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2886 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2887 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2888 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2890 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2891 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2893 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2894 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2896 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2897 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2898 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2899 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2901 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2902 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2904 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2905 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2906 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2907 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2911 let Constraints = "$src1 = $dst" in {
2912 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2913 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2914 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2915 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2916 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2917 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2919 let isCommutable = 0 in {
2920 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2921 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2922 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2923 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2924 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2925 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2926 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2927 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2928 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2929 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2930 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2931 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2932 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2933 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2938 /// In addition, we also have a special variant of the scalar form here to
2939 /// represent the associated intrinsic operation. This form is unlike the
2940 /// plain scalar form, in that it takes an entire vector (instead of a
2941 /// scalar) and leaves the top elements undefined.
2943 /// And, we have a special variant form for a full-vector intrinsic form.
2945 def SSE_SQRTP : OpndItins<
2946 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2949 def SSE_SQRTS : OpndItins<
2950 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2953 def SSE_RCPP : OpndItins<
2954 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2957 def SSE_RCPS : OpndItins<
2958 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2961 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2962 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2963 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2964 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2965 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2966 [(set FR32:$dst, (OpNode FR32:$src))]>;
2967 // For scalar unary operations, fold a load into the operation
2968 // only in OptForSize mode. It eliminates an instruction, but it also
2969 // eliminates a whole-register clobber (the load), so it introduces a
2970 // partial register update condition.
2971 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2972 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2973 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2974 Requires<[HasSSE1, OptForSize]>;
2975 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2976 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2977 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2978 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2979 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2980 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2983 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2984 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2985 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2986 !strconcat(OpcodeStr,
2987 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2989 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2990 !strconcat(OpcodeStr,
2991 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2992 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2993 (ins VR128:$src1, ssmem:$src2),
2994 !strconcat(OpcodeStr,
2995 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2998 /// sse1_fp_unop_p - SSE1 unops in packed form.
2999 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3001 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3002 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3003 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3004 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3005 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3006 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3009 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3010 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3012 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3013 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3014 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3016 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3017 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3018 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3022 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3023 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3024 Intrinsic V4F32Int, OpndItins itins> {
3025 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3026 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3027 [(set VR128:$dst, (V4F32Int VR128:$src))],
3029 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3030 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3031 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3035 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3036 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3037 Intrinsic V4F32Int, OpndItins itins> {
3038 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3039 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3040 [(set VR256:$dst, (V4F32Int VR256:$src))],
3042 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3043 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3044 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3048 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3049 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3050 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3051 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3052 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3053 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3054 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3055 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3056 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3057 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3058 Requires<[HasSSE2, OptForSize]>;
3059 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3060 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3061 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3062 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3063 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3064 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3067 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3068 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3069 let neverHasSideEffects = 1 in {
3070 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3071 !strconcat(OpcodeStr,
3072 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3074 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3075 !strconcat(OpcodeStr,
3076 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3078 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3079 (ins VR128:$src1, sdmem:$src2),
3080 !strconcat(OpcodeStr,
3081 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3084 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3085 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3086 SDNode OpNode, OpndItins itins> {
3087 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3088 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3089 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3090 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3091 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3092 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3095 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3096 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3098 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3099 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3100 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3102 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3103 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3104 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3108 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3109 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3110 Intrinsic V2F64Int, OpndItins itins> {
3111 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3112 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3113 [(set VR128:$dst, (V2F64Int VR128:$src))],
3115 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3116 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3117 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3121 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3122 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3123 Intrinsic V2F64Int, OpndItins itins> {
3124 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3125 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3126 [(set VR256:$dst, (V2F64Int VR256:$src))],
3128 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3129 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3130 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3134 let Predicates = [HasAVX] in {
3136 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3137 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3139 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3140 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3141 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3142 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3143 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3145 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3147 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3149 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3153 // Reciprocal approximations. Note that these typically require refinement
3154 // in order to obtain suitable precision.
3155 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3156 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3157 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3158 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3160 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3163 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3164 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3165 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3166 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3168 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3172 let AddedComplexity = 1 in {
3173 def : Pat<(f32 (fsqrt FR32:$src)),
3174 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3175 def : Pat<(f32 (fsqrt (load addr:$src))),
3176 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3177 Requires<[HasAVX, OptForSize]>;
3178 def : Pat<(f64 (fsqrt FR64:$src)),
3179 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3180 def : Pat<(f64 (fsqrt (load addr:$src))),
3181 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3182 Requires<[HasAVX, OptForSize]>;
3184 def : Pat<(f32 (X86frsqrt FR32:$src)),
3185 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3186 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3187 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3188 Requires<[HasAVX, OptForSize]>;
3190 def : Pat<(f32 (X86frcp FR32:$src)),
3191 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3192 def : Pat<(f32 (X86frcp (load addr:$src))),
3193 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3194 Requires<[HasAVX, OptForSize]>;
3197 let Predicates = [HasAVX], AddedComplexity = 1 in {
3198 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3199 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3200 (COPY_TO_REGCLASS VR128:$src, FR32)),
3202 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3203 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3205 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3206 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3207 (COPY_TO_REGCLASS VR128:$src, FR64)),
3209 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3210 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3212 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3213 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3214 (COPY_TO_REGCLASS VR128:$src, FR32)),
3216 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3217 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3219 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3220 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3221 (COPY_TO_REGCLASS VR128:$src, FR32)),
3223 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3224 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3228 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3230 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3231 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3232 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3234 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3235 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3237 // Reciprocal approximations. Note that these typically require refinement
3238 // in order to obtain suitable precision.
3239 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3241 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3242 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3244 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3246 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3247 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3249 // There is no f64 version of the reciprocal approximation instructions.
3251 //===----------------------------------------------------------------------===//
3252 // SSE 1 & 2 - Non-temporal stores
3253 //===----------------------------------------------------------------------===//
3255 let AddedComplexity = 400 in { // Prefer non-temporal versions
3256 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3257 (ins f128mem:$dst, VR128:$src),
3258 "movntps\t{$src, $dst|$dst, $src}",
3259 [(alignednontemporalstore (v4f32 VR128:$src),
3261 IIC_SSE_MOVNT>, VEX;
3262 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3263 (ins f128mem:$dst, VR128:$src),
3264 "movntpd\t{$src, $dst|$dst, $src}",
3265 [(alignednontemporalstore (v2f64 VR128:$src),
3267 IIC_SSE_MOVNT>, VEX;
3269 let ExeDomain = SSEPackedInt in
3270 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3271 (ins f128mem:$dst, VR128:$src),
3272 "movntdq\t{$src, $dst|$dst, $src}",
3273 [(alignednontemporalstore (v2i64 VR128:$src),
3275 IIC_SSE_MOVNT>, VEX;
3277 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3278 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3280 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3281 (ins f256mem:$dst, VR256:$src),
3282 "movntps\t{$src, $dst|$dst, $src}",
3283 [(alignednontemporalstore (v8f32 VR256:$src),
3285 IIC_SSE_MOVNT>, VEX;
3286 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3287 (ins f256mem:$dst, VR256:$src),
3288 "movntpd\t{$src, $dst|$dst, $src}",
3289 [(alignednontemporalstore (v4f64 VR256:$src),
3291 IIC_SSE_MOVNT>, VEX;
3292 let ExeDomain = SSEPackedInt in
3293 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3294 (ins f256mem:$dst, VR256:$src),
3295 "movntdq\t{$src, $dst|$dst, $src}",
3296 [(alignednontemporalstore (v4i64 VR256:$src),
3298 IIC_SSE_MOVNT>, VEX;
3301 let AddedComplexity = 400 in { // Prefer non-temporal versions
3302 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3303 "movntps\t{$src, $dst|$dst, $src}",
3304 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3306 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3307 "movntpd\t{$src, $dst|$dst, $src}",
3308 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3311 let ExeDomain = SSEPackedInt in
3312 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3313 "movntdq\t{$src, $dst|$dst, $src}",
3314 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3317 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3318 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3320 // There is no AVX form for instructions below this point
3321 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3322 "movnti{l}\t{$src, $dst|$dst, $src}",
3323 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3325 TB, Requires<[HasSSE2]>;
3326 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3327 "movnti{q}\t{$src, $dst|$dst, $src}",
3328 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3330 TB, Requires<[HasSSE2]>;
3333 //===----------------------------------------------------------------------===//
3334 // SSE 1 & 2 - Prefetch and memory fence
3335 //===----------------------------------------------------------------------===//
3337 // Prefetch intrinsic.
3338 let Predicates = [HasSSE1] in {
3339 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3340 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3341 IIC_SSE_PREFETCH>, TB;
3342 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3343 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3344 IIC_SSE_PREFETCH>, TB;
3345 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3346 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3347 IIC_SSE_PREFETCH>, TB;
3348 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3349 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3350 IIC_SSE_PREFETCH>, TB;
3354 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3355 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3356 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3358 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3359 // was introduced with SSE2, it's backward compatible.
3360 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3362 // Load, store, and memory fence
3363 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3364 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3365 TB, Requires<[HasSSE1]>;
3366 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3367 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3368 TB, Requires<[HasSSE2]>;
3369 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3370 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3371 TB, Requires<[HasSSE2]>;
3373 def : Pat<(X86SFence), (SFENCE)>;
3374 def : Pat<(X86LFence), (LFENCE)>;
3375 def : Pat<(X86MFence), (MFENCE)>;
3377 //===----------------------------------------------------------------------===//
3378 // SSE 1 & 2 - Load/Store XCSR register
3379 //===----------------------------------------------------------------------===//
3381 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3382 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3383 IIC_SSE_LDMXCSR>, VEX;
3384 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3385 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3386 IIC_SSE_STMXCSR>, VEX;
3388 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3389 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3391 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3392 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3395 //===---------------------------------------------------------------------===//
3396 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3397 //===---------------------------------------------------------------------===//
3399 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3401 let neverHasSideEffects = 1 in {
3402 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3403 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3405 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3406 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3409 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3410 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3412 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3413 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3417 let isCodeGenOnly = 1 in {
3418 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3419 "movdqa\t{$src, $dst|$dst, $src}", [],
3422 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3423 "movdqa\t{$src, $dst|$dst, $src}", [],
3426 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3427 "movdqu\t{$src, $dst|$dst, $src}", [],
3430 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3431 "movdqu\t{$src, $dst|$dst, $src}", [],
3436 let canFoldAsLoad = 1, mayLoad = 1 in {
3437 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3438 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3440 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3441 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3443 let Predicates = [HasAVX] in {
3444 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3445 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3447 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3448 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3453 let mayStore = 1 in {
3454 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3455 (ins i128mem:$dst, VR128:$src),
3456 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3458 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3459 (ins i256mem:$dst, VR256:$src),
3460 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3462 let Predicates = [HasAVX] in {
3463 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3464 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3466 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3467 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3472 let neverHasSideEffects = 1 in
3473 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3474 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3476 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3477 "movdqu\t{$src, $dst|$dst, $src}",
3478 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3481 let isCodeGenOnly = 1 in {
3482 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3483 "movdqa\t{$src, $dst|$dst, $src}", [],
3486 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3487 "movdqu\t{$src, $dst|$dst, $src}",
3488 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3491 let canFoldAsLoad = 1, mayLoad = 1 in {
3492 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3493 "movdqa\t{$src, $dst|$dst, $src}",
3494 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3496 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3497 "movdqu\t{$src, $dst|$dst, $src}",
3498 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3500 XS, Requires<[HasSSE2]>;
3503 let mayStore = 1 in {
3504 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3505 "movdqa\t{$src, $dst|$dst, $src}",
3506 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3508 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3509 "movdqu\t{$src, $dst|$dst, $src}",
3510 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3512 XS, Requires<[HasSSE2]>;
3515 // Intrinsic forms of MOVDQU load and store
3516 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3517 "vmovdqu\t{$src, $dst|$dst, $src}",
3518 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3520 XS, VEX, Requires<[HasAVX]>;
3522 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3523 "movdqu\t{$src, $dst|$dst, $src}",
3524 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3526 XS, Requires<[HasSSE2]>;
3528 } // ExeDomain = SSEPackedInt
3530 let Predicates = [HasAVX] in {
3531 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3532 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3535 //===---------------------------------------------------------------------===//
3536 // SSE2 - Packed Integer Arithmetic Instructions
3537 //===---------------------------------------------------------------------===//
3539 def SSE_PMADD : OpndItins<
3540 IIC_SSE_PMADD, IIC_SSE_PMADD
3543 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3545 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3546 RegisterClass RC, PatFrag memop_frag,
3547 X86MemOperand x86memop,
3549 bit IsCommutable = 0,
3551 let isCommutable = IsCommutable in
3552 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3553 (ins RC:$src1, RC:$src2),
3555 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3556 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3557 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3558 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3559 (ins RC:$src1, x86memop:$src2),
3561 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3562 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3563 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3567 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3568 string OpcodeStr, SDNode OpNode,
3569 SDNode OpNode2, RegisterClass RC,
3570 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3571 ShiftOpndItins itins,
3573 // src2 is always 128-bit
3574 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3575 (ins RC:$src1, VR128:$src2),
3577 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3578 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3579 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3581 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3582 (ins RC:$src1, i128mem:$src2),
3584 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3585 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3586 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3587 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3588 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3589 (ins RC:$src1, i32i8imm:$src2),
3591 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3592 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3593 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3596 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3597 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3598 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3599 PatFrag memop_frag, X86MemOperand x86memop,
3601 bit IsCommutable = 0, bit Is2Addr = 1> {
3602 let isCommutable = IsCommutable in
3603 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3604 (ins RC:$src1, RC:$src2),
3606 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3607 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3608 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3609 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3610 (ins RC:$src1, x86memop:$src2),
3612 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3613 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3614 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3615 (bitconvert (memop_frag addr:$src2)))))]>;
3617 } // ExeDomain = SSEPackedInt
3619 // 128-bit Integer Arithmetic
3621 let Predicates = [HasAVX] in {
3622 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3623 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3625 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3626 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3627 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3628 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3629 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3630 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3631 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3632 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3633 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3634 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3635 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3636 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3637 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3638 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3639 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3640 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3641 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3642 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3646 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3647 VR128, memopv2i64, i128mem,
3648 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3649 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3650 VR128, memopv2i64, i128mem,
3651 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3652 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3653 VR128, memopv2i64, i128mem,
3654 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3655 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3656 VR128, memopv2i64, i128mem,
3657 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3658 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3659 VR128, memopv2i64, i128mem,
3660 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3661 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3662 VR128, memopv2i64, i128mem,
3663 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3664 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3665 VR128, memopv2i64, i128mem,
3666 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3667 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3668 VR128, memopv2i64, i128mem,
3669 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3670 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3671 VR128, memopv2i64, i128mem,
3672 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3673 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3674 VR128, memopv2i64, i128mem,
3675 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3676 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3677 VR128, memopv2i64, i128mem,
3678 SSE_PMADD, 1, 0>, VEX_4V;
3679 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3680 VR128, memopv2i64, i128mem,
3681 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3682 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3683 VR128, memopv2i64, i128mem,
3684 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3685 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3686 VR128, memopv2i64, i128mem,
3687 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3688 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3689 VR128, memopv2i64, i128mem,
3690 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3691 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3692 VR128, memopv2i64, i128mem,
3693 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3694 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3695 VR128, memopv2i64, i128mem,
3696 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3697 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3698 VR128, memopv2i64, i128mem,
3699 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3702 let Predicates = [HasAVX2] in {
3703 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3704 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3705 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3706 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3707 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3708 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3709 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3710 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3711 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3712 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3713 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3714 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3715 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3716 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3717 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3718 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3719 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3720 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3721 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3722 VR256, memopv4i64, i256mem,
3723 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3726 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3727 VR256, memopv4i64, i256mem,
3728 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3729 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3730 VR256, memopv4i64, i256mem,
3731 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3732 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3733 VR256, memopv4i64, i256mem,
3734 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3735 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3736 VR256, memopv4i64, i256mem,
3737 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3738 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3739 VR256, memopv4i64, i256mem,
3740 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3741 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3742 VR256, memopv4i64, i256mem,
3743 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3744 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3745 VR256, memopv4i64, i256mem,
3746 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3747 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3748 VR256, memopv4i64, i256mem,
3749 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3750 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3751 VR256, memopv4i64, i256mem,
3752 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3753 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3754 VR256, memopv4i64, i256mem,
3755 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3756 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3757 VR256, memopv4i64, i256mem,
3758 SSE_PMADD, 1, 0>, VEX_4V;
3759 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3760 VR256, memopv4i64, i256mem,
3761 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3762 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3763 VR256, memopv4i64, i256mem,
3764 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3765 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3766 VR256, memopv4i64, i256mem,
3767 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3768 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3769 VR256, memopv4i64, i256mem,
3770 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3771 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3772 VR256, memopv4i64, i256mem,
3773 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3774 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3775 VR256, memopv4i64, i256mem,
3776 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3777 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3778 VR256, memopv4i64, i256mem,
3779 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3782 let Constraints = "$src1 = $dst" in {
3783 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3784 i128mem, SSE_INTALU_ITINS_P, 1>;
3785 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3786 i128mem, SSE_INTALU_ITINS_P, 1>;
3787 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3788 i128mem, SSE_INTALU_ITINS_P, 1>;
3789 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3790 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3791 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3792 i128mem, SSE_INTMUL_ITINS_P, 1>;
3793 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3794 i128mem, SSE_INTALU_ITINS_P>;
3795 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3796 i128mem, SSE_INTALU_ITINS_P>;
3797 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3798 i128mem, SSE_INTALU_ITINS_P>;
3799 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3800 i128mem, SSE_INTALUQ_ITINS_P>;
3801 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3802 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3805 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3806 VR128, memopv2i64, i128mem,
3807 SSE_INTALU_ITINS_P>;
3808 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3809 VR128, memopv2i64, i128mem,
3810 SSE_INTALU_ITINS_P>;
3811 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3812 VR128, memopv2i64, i128mem,
3813 SSE_INTALU_ITINS_P>;
3814 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3815 VR128, memopv2i64, i128mem,
3816 SSE_INTALU_ITINS_P>;
3817 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3818 VR128, memopv2i64, i128mem,
3819 SSE_INTALU_ITINS_P, 1>;
3820 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3821 VR128, memopv2i64, i128mem,
3822 SSE_INTALU_ITINS_P, 1>;
3823 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3824 VR128, memopv2i64, i128mem,
3825 SSE_INTALU_ITINS_P, 1>;
3826 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3827 VR128, memopv2i64, i128mem,
3828 SSE_INTALU_ITINS_P, 1>;
3829 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3830 VR128, memopv2i64, i128mem,
3831 SSE_INTMUL_ITINS_P, 1>;
3832 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3833 VR128, memopv2i64, i128mem,
3834 SSE_INTMUL_ITINS_P, 1>;
3835 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3836 VR128, memopv2i64, i128mem,
3838 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3839 VR128, memopv2i64, i128mem,
3840 SSE_INTALU_ITINS_P, 1>;
3841 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3842 VR128, memopv2i64, i128mem,
3843 SSE_INTALU_ITINS_P, 1>;
3844 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3845 VR128, memopv2i64, i128mem,
3846 SSE_INTALU_ITINS_P, 1>;
3847 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3848 VR128, memopv2i64, i128mem,
3849 SSE_INTALU_ITINS_P, 1>;
3850 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3851 VR128, memopv2i64, i128mem,
3852 SSE_INTALU_ITINS_P, 1>;
3853 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3854 VR128, memopv2i64, i128mem,
3855 SSE_INTALU_ITINS_P, 1>;
3856 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3857 VR128, memopv2i64, i128mem,
3858 SSE_INTALU_ITINS_P, 1>;
3860 } // Constraints = "$src1 = $dst"
3862 //===---------------------------------------------------------------------===//
3863 // SSE2 - Packed Integer Logical Instructions
3864 //===---------------------------------------------------------------------===//
3866 let Predicates = [HasAVX] in {
3867 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3868 VR128, v8i16, v8i16, bc_v8i16,
3869 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3870 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3871 VR128, v4i32, v4i32, bc_v4i32,
3872 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3873 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3874 VR128, v2i64, v2i64, bc_v2i64,
3875 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3877 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3878 VR128, v8i16, v8i16, bc_v8i16,
3879 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3880 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3881 VR128, v4i32, v4i32, bc_v4i32,
3882 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3883 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3884 VR128, v2i64, v2i64, bc_v2i64,
3885 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3887 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3888 VR128, v8i16, v8i16, bc_v8i16,
3889 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3890 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3891 VR128, v4i32, v4i32, bc_v4i32,
3892 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3894 let ExeDomain = SSEPackedInt in {
3895 // 128-bit logical shifts.
3896 def VPSLLDQri : PDIi8<0x73, MRM7r,
3897 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3898 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3900 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3902 def VPSRLDQri : PDIi8<0x73, MRM3r,
3903 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3904 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3906 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3908 // PSRADQri doesn't exist in SSE[1-3].
3910 } // Predicates = [HasAVX]
3912 let Predicates = [HasAVX2] in {
3913 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3914 VR256, v16i16, v8i16, bc_v8i16,
3915 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3916 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3917 VR256, v8i32, v4i32, bc_v4i32,
3918 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3919 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3920 VR256, v4i64, v2i64, bc_v2i64,
3921 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3923 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3924 VR256, v16i16, v8i16, bc_v8i16,
3925 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3926 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3927 VR256, v8i32, v4i32, bc_v4i32,
3928 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3929 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3930 VR256, v4i64, v2i64, bc_v2i64,
3931 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3933 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3934 VR256, v16i16, v8i16, bc_v8i16,
3935 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3936 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3937 VR256, v8i32, v4i32, bc_v4i32,
3938 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3940 let ExeDomain = SSEPackedInt in {
3941 // 256-bit logical shifts.
3942 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3943 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3944 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3946 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3948 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3949 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3950 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3952 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3954 // PSRADQYri doesn't exist in SSE[1-3].
3956 } // Predicates = [HasAVX2]
3958 let Constraints = "$src1 = $dst" in {
3959 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3960 VR128, v8i16, v8i16, bc_v8i16,
3961 SSE_INTSHIFT_ITINS_P>;
3962 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3963 VR128, v4i32, v4i32, bc_v4i32,
3964 SSE_INTSHIFT_ITINS_P>;
3965 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3966 VR128, v2i64, v2i64, bc_v2i64,
3967 SSE_INTSHIFT_ITINS_P>;
3969 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3970 VR128, v8i16, v8i16, bc_v8i16,
3971 SSE_INTSHIFT_ITINS_P>;
3972 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3973 VR128, v4i32, v4i32, bc_v4i32,
3974 SSE_INTSHIFT_ITINS_P>;
3975 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3976 VR128, v2i64, v2i64, bc_v2i64,
3977 SSE_INTSHIFT_ITINS_P>;
3979 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3980 VR128, v8i16, v8i16, bc_v8i16,
3981 SSE_INTSHIFT_ITINS_P>;
3982 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3983 VR128, v4i32, v4i32, bc_v4i32,
3984 SSE_INTSHIFT_ITINS_P>;
3986 let ExeDomain = SSEPackedInt in {
3987 // 128-bit logical shifts.
3988 def PSLLDQri : PDIi8<0x73, MRM7r,
3989 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3990 "pslldq\t{$src2, $dst|$dst, $src2}",
3992 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3993 def PSRLDQri : PDIi8<0x73, MRM3r,
3994 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3995 "psrldq\t{$src2, $dst|$dst, $src2}",
3997 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3998 // PSRADQri doesn't exist in SSE[1-3].
4000 } // Constraints = "$src1 = $dst"
4002 let Predicates = [HasAVX] in {
4003 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4004 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4005 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4006 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4007 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4008 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4010 // Shift up / down and insert zero's.
4011 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4012 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4013 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4014 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4017 let Predicates = [HasAVX2] in {
4018 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4019 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4020 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4021 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4024 let Predicates = [HasSSE2] in {
4025 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4026 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4027 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4028 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4029 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4030 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4032 // Shift up / down and insert zero's.
4033 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4034 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4035 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4036 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4039 //===---------------------------------------------------------------------===//
4040 // SSE2 - Packed Integer Comparison Instructions
4041 //===---------------------------------------------------------------------===//
4043 let Predicates = [HasAVX] in {
4044 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4045 VR128, memopv2i64, i128mem,
4046 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4047 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4048 VR128, memopv2i64, i128mem,
4049 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4050 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4051 VR128, memopv2i64, i128mem,
4052 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4053 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4054 VR128, memopv2i64, i128mem,
4055 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4056 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4057 VR128, memopv2i64, i128mem,
4058 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4059 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4060 VR128, memopv2i64, i128mem,
4061 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4064 let Predicates = [HasAVX2] in {
4065 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4066 VR256, memopv4i64, i256mem,
4067 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4068 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4069 VR256, memopv4i64, i256mem,
4070 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4071 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4072 VR256, memopv4i64, i256mem,
4073 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4074 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4075 VR256, memopv4i64, i256mem,
4076 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4077 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4078 VR256, memopv4i64, i256mem,
4079 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4080 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4081 VR256, memopv4i64, i256mem,
4082 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4085 let Constraints = "$src1 = $dst" in {
4086 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4087 VR128, memopv2i64, i128mem,
4088 SSE_INTALU_ITINS_P, 1>;
4089 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4090 VR128, memopv2i64, i128mem,
4091 SSE_INTALU_ITINS_P, 1>;
4092 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4093 VR128, memopv2i64, i128mem,
4094 SSE_INTALU_ITINS_P, 1>;
4095 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4096 VR128, memopv2i64, i128mem,
4097 SSE_INTALU_ITINS_P>;
4098 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4099 VR128, memopv2i64, i128mem,
4100 SSE_INTALU_ITINS_P>;
4101 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4102 VR128, memopv2i64, i128mem,
4103 SSE_INTALU_ITINS_P>;
4104 } // Constraints = "$src1 = $dst"
4106 //===---------------------------------------------------------------------===//
4107 // SSE2 - Packed Integer Pack Instructions
4108 //===---------------------------------------------------------------------===//
4110 let Predicates = [HasAVX] in {
4111 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4112 VR128, memopv2i64, i128mem,
4113 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4114 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4115 VR128, memopv2i64, i128mem,
4116 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4117 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4118 VR128, memopv2i64, i128mem,
4119 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4122 let Predicates = [HasAVX2] in {
4123 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4124 VR256, memopv4i64, i256mem,
4125 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4126 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4127 VR256, memopv4i64, i256mem,
4128 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4129 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4130 VR256, memopv4i64, i256mem,
4131 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4134 let Constraints = "$src1 = $dst" in {
4135 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4136 VR128, memopv2i64, i128mem,
4137 SSE_INTALU_ITINS_P>;
4138 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4139 VR128, memopv2i64, i128mem,
4140 SSE_INTALU_ITINS_P>;
4141 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4142 VR128, memopv2i64, i128mem,
4143 SSE_INTALU_ITINS_P>;
4144 } // Constraints = "$src1 = $dst"
4146 //===---------------------------------------------------------------------===//
4147 // SSE2 - Packed Integer Shuffle Instructions
4148 //===---------------------------------------------------------------------===//
4150 let ExeDomain = SSEPackedInt in {
4151 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4152 def ri : Ii8<0x70, MRMSrcReg,
4153 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4154 !strconcat(OpcodeStr,
4155 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4156 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4158 def mi : Ii8<0x70, MRMSrcMem,
4159 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4160 !strconcat(OpcodeStr,
4161 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4163 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4168 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4169 def Yri : Ii8<0x70, MRMSrcReg,
4170 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4171 !strconcat(OpcodeStr,
4172 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4173 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4174 def Ymi : Ii8<0x70, MRMSrcMem,
4175 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4176 !strconcat(OpcodeStr,
4177 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4179 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4180 (i8 imm:$src2))))]>;
4182 } // ExeDomain = SSEPackedInt
4184 let Predicates = [HasAVX] in {
4185 let AddedComplexity = 5 in
4186 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4188 // SSE2 with ImmT == Imm8 and XS prefix.
4189 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4191 // SSE2 with ImmT == Imm8 and XD prefix.
4192 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4194 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4195 (VPSHUFDmi addr:$src1, imm:$imm)>;
4196 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4197 (VPSHUFDri VR128:$src1, imm:$imm)>;
4200 let Predicates = [HasAVX2] in {
4201 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4202 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4203 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4206 let Predicates = [HasSSE2] in {
4207 let AddedComplexity = 5 in
4208 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4210 // SSE2 with ImmT == Imm8 and XS prefix.
4211 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4213 // SSE2 with ImmT == Imm8 and XD prefix.
4214 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4216 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4217 (PSHUFDmi addr:$src1, imm:$imm)>;
4218 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4219 (PSHUFDri VR128:$src1, imm:$imm)>;
4222 //===---------------------------------------------------------------------===//
4223 // SSE2 - Packed Integer Unpack Instructions
4224 //===---------------------------------------------------------------------===//
4226 let ExeDomain = SSEPackedInt in {
4227 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4228 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4229 def rr : PDI<opc, MRMSrcReg,
4230 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4232 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4233 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4234 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4236 def rm : PDI<opc, MRMSrcMem,
4237 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4239 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4240 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4241 [(set VR128:$dst, (OpNode VR128:$src1,
4242 (bc_frag (memopv2i64
4247 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4248 SDNode OpNode, PatFrag bc_frag> {
4249 def Yrr : PDI<opc, MRMSrcReg,
4250 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4251 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4252 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4253 def Yrm : PDI<opc, MRMSrcMem,
4254 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4255 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4256 [(set VR256:$dst, (OpNode VR256:$src1,
4257 (bc_frag (memopv4i64 addr:$src2))))]>;
4260 let Predicates = [HasAVX] in {
4261 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4262 bc_v16i8, 0>, VEX_4V;
4263 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4264 bc_v8i16, 0>, VEX_4V;
4265 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4266 bc_v4i32, 0>, VEX_4V;
4267 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4268 bc_v2i64, 0>, VEX_4V;
4270 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4271 bc_v16i8, 0>, VEX_4V;
4272 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4273 bc_v8i16, 0>, VEX_4V;
4274 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4275 bc_v4i32, 0>, VEX_4V;
4276 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4277 bc_v2i64, 0>, VEX_4V;
4280 let Predicates = [HasAVX2] in {
4281 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4283 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4285 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4287 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4290 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4292 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4294 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4296 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4300 let Constraints = "$src1 = $dst" in {
4301 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4303 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4305 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4307 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4310 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4312 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4314 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4316 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4319 } // ExeDomain = SSEPackedInt
4321 // Patterns for using AVX1 instructions with integer vectors
4322 // Here to give AVX2 priority
4323 let Predicates = [HasAVX] in {
4324 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4325 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4326 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4327 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4328 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4329 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4330 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4331 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4333 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4334 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4335 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4336 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4337 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4338 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4339 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4340 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4343 //===---------------------------------------------------------------------===//
4344 // SSE2 - Packed Integer Extract and Insert
4345 //===---------------------------------------------------------------------===//
4347 let ExeDomain = SSEPackedInt in {
4348 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4349 def rri : Ii8<0xC4, MRMSrcReg,
4350 (outs VR128:$dst), (ins VR128:$src1,
4351 GR32:$src2, i32i8imm:$src3),
4353 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4354 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4356 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4357 def rmi : Ii8<0xC4, MRMSrcMem,
4358 (outs VR128:$dst), (ins VR128:$src1,
4359 i16mem:$src2, i32i8imm:$src3),
4361 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4362 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4364 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4365 imm:$src3))], IIC_SSE_PINSRW>;
4369 let Predicates = [HasAVX] in
4370 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4371 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4372 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4373 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4374 imm:$src2))]>, TB, OpSize, VEX;
4375 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4376 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4377 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4378 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4379 imm:$src2))], IIC_SSE_PEXTRW>;
4382 let Predicates = [HasAVX] in {
4383 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4384 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4385 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4386 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4387 []>, TB, OpSize, VEX_4V;
4390 let Constraints = "$src1 = $dst" in
4391 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4393 } // ExeDomain = SSEPackedInt
4395 //===---------------------------------------------------------------------===//
4396 // SSE2 - Packed Mask Creation
4397 //===---------------------------------------------------------------------===//
4399 let ExeDomain = SSEPackedInt in {
4401 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4402 "pmovmskb\t{$src, $dst|$dst, $src}",
4403 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4404 IIC_SSE_MOVMSK>, VEX;
4405 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4406 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4408 let Predicates = [HasAVX2] in {
4409 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4410 "pmovmskb\t{$src, $dst|$dst, $src}",
4411 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4412 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4413 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4416 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4417 "pmovmskb\t{$src, $dst|$dst, $src}",
4418 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4421 } // ExeDomain = SSEPackedInt
4423 //===---------------------------------------------------------------------===//
4424 // SSE2 - Conditional Store
4425 //===---------------------------------------------------------------------===//
4427 let ExeDomain = SSEPackedInt in {
4430 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4431 (ins VR128:$src, VR128:$mask),
4432 "maskmovdqu\t{$mask, $src|$src, $mask}",
4433 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4434 IIC_SSE_MASKMOV>, VEX;
4436 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4437 (ins VR128:$src, VR128:$mask),
4438 "maskmovdqu\t{$mask, $src|$src, $mask}",
4439 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4440 IIC_SSE_MASKMOV>, VEX;
4443 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4444 "maskmovdqu\t{$mask, $src|$src, $mask}",
4445 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4448 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4449 "maskmovdqu\t{$mask, $src|$src, $mask}",
4450 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4453 } // ExeDomain = SSEPackedInt
4455 //===---------------------------------------------------------------------===//
4456 // SSE2 - Move Doubleword
4457 //===---------------------------------------------------------------------===//
4459 //===---------------------------------------------------------------------===//
4460 // Move Int Doubleword to Packed Double Int
4462 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4463 "movd\t{$src, $dst|$dst, $src}",
4465 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4467 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4468 "movd\t{$src, $dst|$dst, $src}",
4470 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4473 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4474 "mov{d|q}\t{$src, $dst|$dst, $src}",
4476 (v2i64 (scalar_to_vector GR64:$src)))],
4477 IIC_SSE_MOVDQ>, VEX;
4478 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4479 "mov{d|q}\t{$src, $dst|$dst, $src}",
4480 [(set FR64:$dst, (bitconvert GR64:$src))],
4481 IIC_SSE_MOVDQ>, VEX;
4483 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4484 "movd\t{$src, $dst|$dst, $src}",
4486 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4487 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4488 "movd\t{$src, $dst|$dst, $src}",
4490 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4492 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4493 "mov{d|q}\t{$src, $dst|$dst, $src}",
4495 (v2i64 (scalar_to_vector GR64:$src)))],
4497 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4498 "mov{d|q}\t{$src, $dst|$dst, $src}",
4499 [(set FR64:$dst, (bitconvert GR64:$src))],
4502 //===---------------------------------------------------------------------===//
4503 // Move Int Doubleword to Single Scalar
4505 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4506 "movd\t{$src, $dst|$dst, $src}",
4507 [(set FR32:$dst, (bitconvert GR32:$src))],
4508 IIC_SSE_MOVDQ>, VEX;
4510 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4511 "movd\t{$src, $dst|$dst, $src}",
4512 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4515 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4516 "movd\t{$src, $dst|$dst, $src}",
4517 [(set FR32:$dst, (bitconvert GR32:$src))],
4520 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4521 "movd\t{$src, $dst|$dst, $src}",
4522 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4525 //===---------------------------------------------------------------------===//
4526 // Move Packed Doubleword Int to Packed Double Int
4528 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4529 "movd\t{$src, $dst|$dst, $src}",
4530 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4531 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4532 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4533 (ins i32mem:$dst, VR128:$src),
4534 "movd\t{$src, $dst|$dst, $src}",
4535 [(store (i32 (vector_extract (v4i32 VR128:$src),
4536 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4538 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4539 "movd\t{$src, $dst|$dst, $src}",
4540 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4541 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4542 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4543 "movd\t{$src, $dst|$dst, $src}",
4544 [(store (i32 (vector_extract (v4i32 VR128:$src),
4545 (iPTR 0))), addr:$dst)],
4548 //===---------------------------------------------------------------------===//
4549 // Move Packed Doubleword Int first element to Doubleword Int
4551 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4552 "mov{d|q}\t{$src, $dst|$dst, $src}",
4553 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4556 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4558 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4559 "mov{d|q}\t{$src, $dst|$dst, $src}",
4560 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4564 //===---------------------------------------------------------------------===//
4565 // Bitcast FR64 <-> GR64
4567 let Predicates = [HasAVX] in
4568 def VMOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4569 "vmovq\t{$src, $dst|$dst, $src}",
4570 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4572 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4573 "mov{d|q}\t{$src, $dst|$dst, $src}",
4574 [(set GR64:$dst, (bitconvert FR64:$src))],
4575 IIC_SSE_MOVDQ>, VEX;
4576 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4577 "movq\t{$src, $dst|$dst, $src}",
4578 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4579 IIC_SSE_MOVDQ>, VEX;
4581 def MOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4582 "movq\t{$src, $dst|$dst, $src}",
4583 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4585 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4586 "mov{d|q}\t{$src, $dst|$dst, $src}",
4587 [(set GR64:$dst, (bitconvert FR64:$src))],
4589 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4590 "movq\t{$src, $dst|$dst, $src}",
4591 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4594 //===---------------------------------------------------------------------===//
4595 // Move Scalar Single to Double Int
4597 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4598 "movd\t{$src, $dst|$dst, $src}",
4599 [(set GR32:$dst, (bitconvert FR32:$src))],
4600 IIC_SSE_MOVD_ToGP>, VEX;
4601 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4602 "movd\t{$src, $dst|$dst, $src}",
4603 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4604 IIC_SSE_MOVDQ>, VEX;
4605 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4606 "movd\t{$src, $dst|$dst, $src}",
4607 [(set GR32:$dst, (bitconvert FR32:$src))],
4609 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4610 "movd\t{$src, $dst|$dst, $src}",
4611 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4614 //===---------------------------------------------------------------------===//
4615 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4617 let AddedComplexity = 15 in {
4618 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4619 "movd\t{$src, $dst|$dst, $src}",
4620 [(set VR128:$dst, (v4i32 (X86vzmovl
4621 (v4i32 (scalar_to_vector GR32:$src)))))],
4622 IIC_SSE_MOVDQ>, VEX;
4623 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4624 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4625 [(set VR128:$dst, (v2i64 (X86vzmovl
4626 (v2i64 (scalar_to_vector GR64:$src)))))],
4630 let AddedComplexity = 15 in {
4631 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4632 "movd\t{$src, $dst|$dst, $src}",
4633 [(set VR128:$dst, (v4i32 (X86vzmovl
4634 (v4i32 (scalar_to_vector GR32:$src)))))],
4636 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4637 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4638 [(set VR128:$dst, (v2i64 (X86vzmovl
4639 (v2i64 (scalar_to_vector GR64:$src)))))],
4643 let AddedComplexity = 20 in {
4644 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4645 "movd\t{$src, $dst|$dst, $src}",
4647 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4648 (loadi32 addr:$src))))))],
4649 IIC_SSE_MOVDQ>, VEX;
4650 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4651 "movd\t{$src, $dst|$dst, $src}",
4653 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4654 (loadi32 addr:$src))))))],
4658 let Predicates = [HasAVX] in {
4659 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4660 let AddedComplexity = 20 in {
4661 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4662 (VMOVZDI2PDIrm addr:$src)>;
4663 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4664 (VMOVZDI2PDIrm addr:$src)>;
4666 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4667 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4668 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4669 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4670 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4671 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4672 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4675 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4676 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4677 (MOVZDI2PDIrm addr:$src)>;
4678 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4679 (MOVZDI2PDIrm addr:$src)>;
4682 // These are the correct encodings of the instructions so that we know how to
4683 // read correct assembly, even though we continue to emit the wrong ones for
4684 // compatibility with Darwin's buggy assembler.
4685 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4686 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4687 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4688 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4689 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4690 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4691 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4692 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4693 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4694 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4695 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4696 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4698 //===---------------------------------------------------------------------===//
4699 // SSE2 - Move Quadword
4700 //===---------------------------------------------------------------------===//
4702 //===---------------------------------------------------------------------===//
4703 // Move Quadword Int to Packed Quadword Int
4705 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4706 "vmovq\t{$src, $dst|$dst, $src}",
4708 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4709 VEX, Requires<[HasAVX]>;
4710 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4711 "movq\t{$src, $dst|$dst, $src}",
4713 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4715 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4717 //===---------------------------------------------------------------------===//
4718 // Move Packed Quadword Int to Quadword Int
4720 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4721 "movq\t{$src, $dst|$dst, $src}",
4722 [(store (i64 (vector_extract (v2i64 VR128:$src),
4723 (iPTR 0))), addr:$dst)],
4724 IIC_SSE_MOVDQ>, VEX;
4725 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4726 "movq\t{$src, $dst|$dst, $src}",
4727 [(store (i64 (vector_extract (v2i64 VR128:$src),
4728 (iPTR 0))), addr:$dst)],
4731 //===---------------------------------------------------------------------===//
4732 // Store / copy lower 64-bits of a XMM register.
4734 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4735 "movq\t{$src, $dst|$dst, $src}",
4736 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4737 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4738 "movq\t{$src, $dst|$dst, $src}",
4739 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4742 let AddedComplexity = 20 in
4743 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4744 "vmovq\t{$src, $dst|$dst, $src}",
4746 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4747 (loadi64 addr:$src))))))],
4749 XS, VEX, Requires<[HasAVX]>;
4751 let AddedComplexity = 20 in
4752 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4753 "movq\t{$src, $dst|$dst, $src}",
4755 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4756 (loadi64 addr:$src))))))],
4758 XS, Requires<[HasSSE2]>;
4760 let Predicates = [HasAVX], AddedComplexity = 20 in {
4761 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4762 (VMOVZQI2PQIrm addr:$src)>;
4763 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4764 (VMOVZQI2PQIrm addr:$src)>;
4765 def : Pat<(v2i64 (X86vzload addr:$src)),
4766 (VMOVZQI2PQIrm addr:$src)>;
4769 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4770 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4771 (MOVZQI2PQIrm addr:$src)>;
4772 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4773 (MOVZQI2PQIrm addr:$src)>;
4774 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4777 let Predicates = [HasAVX] in {
4778 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4779 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4780 def : Pat<(v4i64 (X86vzload addr:$src)),
4781 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4784 //===---------------------------------------------------------------------===//
4785 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4786 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4788 let AddedComplexity = 15 in
4789 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4790 "vmovq\t{$src, $dst|$dst, $src}",
4791 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4793 XS, VEX, Requires<[HasAVX]>;
4794 let AddedComplexity = 15 in
4795 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4796 "movq\t{$src, $dst|$dst, $src}",
4797 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4799 XS, Requires<[HasSSE2]>;
4801 let AddedComplexity = 20 in
4802 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4803 "vmovq\t{$src, $dst|$dst, $src}",
4804 [(set VR128:$dst, (v2i64 (X86vzmovl
4805 (loadv2i64 addr:$src))))],
4807 XS, VEX, Requires<[HasAVX]>;
4808 let AddedComplexity = 20 in {
4809 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4810 "movq\t{$src, $dst|$dst, $src}",
4811 [(set VR128:$dst, (v2i64 (X86vzmovl
4812 (loadv2i64 addr:$src))))],
4814 XS, Requires<[HasSSE2]>;
4817 let AddedComplexity = 20 in {
4818 let Predicates = [HasAVX] in {
4819 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4820 (VMOVZPQILo2PQIrm addr:$src)>;
4821 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4822 (VMOVZPQILo2PQIrr VR128:$src)>;
4824 let Predicates = [HasSSE2] in {
4825 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4826 (MOVZPQILo2PQIrm addr:$src)>;
4827 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4828 (MOVZPQILo2PQIrr VR128:$src)>;
4832 // Instructions to match in the assembler
4833 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4834 "movq\t{$src, $dst|$dst, $src}", [],
4835 IIC_SSE_MOVDQ>, VEX, VEX_W;
4836 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4837 "movq\t{$src, $dst|$dst, $src}", [],
4838 IIC_SSE_MOVDQ>, VEX, VEX_W;
4839 // Recognize "movd" with GR64 destination, but encode as a "movq"
4840 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4841 "movd\t{$src, $dst|$dst, $src}", [],
4842 IIC_SSE_MOVDQ>, VEX, VEX_W;
4844 // Instructions for the disassembler
4845 // xr = XMM register
4848 let Predicates = [HasAVX] in
4849 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4850 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4851 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4852 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4854 //===---------------------------------------------------------------------===//
4855 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4856 //===---------------------------------------------------------------------===//
4857 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4858 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4859 X86MemOperand x86memop> {
4860 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4861 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4862 [(set RC:$dst, (vt (OpNode RC:$src)))],
4864 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4865 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4866 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4870 let Predicates = [HasAVX] in {
4871 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4872 v4f32, VR128, memopv4f32, f128mem>, VEX;
4873 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4874 v4f32, VR128, memopv4f32, f128mem>, VEX;
4875 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4876 v8f32, VR256, memopv8f32, f256mem>, VEX;
4877 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4878 v8f32, VR256, memopv8f32, f256mem>, VEX;
4880 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4881 memopv4f32, f128mem>;
4882 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4883 memopv4f32, f128mem>;
4885 let Predicates = [HasAVX] in {
4886 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4887 (VMOVSHDUPrr VR128:$src)>;
4888 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4889 (VMOVSHDUPrm addr:$src)>;
4890 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4891 (VMOVSLDUPrr VR128:$src)>;
4892 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4893 (VMOVSLDUPrm addr:$src)>;
4894 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4895 (VMOVSHDUPYrr VR256:$src)>;
4896 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4897 (VMOVSHDUPYrm addr:$src)>;
4898 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4899 (VMOVSLDUPYrr VR256:$src)>;
4900 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4901 (VMOVSLDUPYrm addr:$src)>;
4904 let Predicates = [HasSSE3] in {
4905 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4906 (MOVSHDUPrr VR128:$src)>;
4907 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4908 (MOVSHDUPrm addr:$src)>;
4909 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4910 (MOVSLDUPrr VR128:$src)>;
4911 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4912 (MOVSLDUPrm addr:$src)>;
4915 //===---------------------------------------------------------------------===//
4916 // SSE3 - Replicate Double FP - MOVDDUP
4917 //===---------------------------------------------------------------------===//
4919 multiclass sse3_replicate_dfp<string OpcodeStr> {
4920 let neverHasSideEffects = 1 in
4921 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4922 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4923 [], IIC_SSE_MOV_LH>;
4924 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4925 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4928 (scalar_to_vector (loadf64 addr:$src)))))],
4932 // FIXME: Merge with above classe when there're patterns for the ymm version
4933 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4934 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4935 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4936 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4937 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4938 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4941 (scalar_to_vector (loadf64 addr:$src)))))]>;
4944 let Predicates = [HasAVX] in {
4945 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4946 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4949 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4951 let Predicates = [HasAVX] in {
4952 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4953 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4954 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4955 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4956 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4957 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4958 def : Pat<(X86Movddup (bc_v2f64
4959 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4960 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4963 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4964 (VMOVDDUPYrm addr:$src)>;
4965 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4966 (VMOVDDUPYrm addr:$src)>;
4967 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4968 (VMOVDDUPYrm addr:$src)>;
4969 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4970 (VMOVDDUPYrr VR256:$src)>;
4973 let Predicates = [HasSSE3] in {
4974 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4975 (MOVDDUPrm addr:$src)>;
4976 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4977 (MOVDDUPrm addr:$src)>;
4978 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4979 (MOVDDUPrm addr:$src)>;
4980 def : Pat<(X86Movddup (bc_v2f64
4981 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4982 (MOVDDUPrm addr:$src)>;
4985 //===---------------------------------------------------------------------===//
4986 // SSE3 - Move Unaligned Integer
4987 //===---------------------------------------------------------------------===//
4989 let Predicates = [HasAVX] in {
4990 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4991 "vlddqu\t{$src, $dst|$dst, $src}",
4992 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4993 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4994 "vlddqu\t{$src, $dst|$dst, $src}",
4995 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4997 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4998 "lddqu\t{$src, $dst|$dst, $src}",
4999 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5002 //===---------------------------------------------------------------------===//
5003 // SSE3 - Arithmetic
5004 //===---------------------------------------------------------------------===//
5006 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5007 X86MemOperand x86memop, OpndItins itins,
5009 def rr : I<0xD0, MRMSrcReg,
5010 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5012 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5013 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5014 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
5015 def rm : I<0xD0, MRMSrcMem,
5016 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5018 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5019 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5020 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5023 let Predicates = [HasAVX] in {
5024 let ExeDomain = SSEPackedSingle in {
5025 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5026 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5027 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5028 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5030 let ExeDomain = SSEPackedDouble in {
5031 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5032 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5033 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5034 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5037 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5038 let ExeDomain = SSEPackedSingle in
5039 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5040 f128mem, SSE_ALU_F32P>, TB, XD;
5041 let ExeDomain = SSEPackedDouble in
5042 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5043 f128mem, SSE_ALU_F64P>, TB, OpSize;
5046 //===---------------------------------------------------------------------===//
5047 // SSE3 Instructions
5048 //===---------------------------------------------------------------------===//
5051 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5052 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5053 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5055 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5056 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5057 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5059 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5061 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5062 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5063 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5064 IIC_SSE_HADDSUB_RM>;
5066 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5067 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5068 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5070 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5071 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5072 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5074 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5076 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5077 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5078 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5079 IIC_SSE_HADDSUB_RM>;
5082 let Predicates = [HasAVX] in {
5083 let ExeDomain = SSEPackedSingle in {
5084 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5085 X86fhadd, 0>, VEX_4V;
5086 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5087 X86fhsub, 0>, VEX_4V;
5088 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5089 X86fhadd, 0>, VEX_4V;
5090 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5091 X86fhsub, 0>, VEX_4V;
5093 let ExeDomain = SSEPackedDouble in {
5094 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5095 X86fhadd, 0>, VEX_4V;
5096 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5097 X86fhsub, 0>, VEX_4V;
5098 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5099 X86fhadd, 0>, VEX_4V;
5100 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5101 X86fhsub, 0>, VEX_4V;
5105 let Constraints = "$src1 = $dst" in {
5106 let ExeDomain = SSEPackedSingle in {
5107 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5108 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5110 let ExeDomain = SSEPackedDouble in {
5111 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5112 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5116 //===---------------------------------------------------------------------===//
5117 // SSSE3 - Packed Absolute Instructions
5118 //===---------------------------------------------------------------------===//
5121 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5122 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5123 Intrinsic IntId128> {
5124 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5126 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5127 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5130 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5132 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5135 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5139 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5140 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5141 Intrinsic IntId256> {
5142 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5144 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5145 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5148 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5150 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5153 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5156 let Predicates = [HasAVX] in {
5157 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5158 int_x86_ssse3_pabs_b_128>, VEX;
5159 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5160 int_x86_ssse3_pabs_w_128>, VEX;
5161 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5162 int_x86_ssse3_pabs_d_128>, VEX;
5165 let Predicates = [HasAVX2] in {
5166 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5167 int_x86_avx2_pabs_b>, VEX;
5168 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5169 int_x86_avx2_pabs_w>, VEX;
5170 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5171 int_x86_avx2_pabs_d>, VEX;
5174 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5175 int_x86_ssse3_pabs_b_128>;
5176 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5177 int_x86_ssse3_pabs_w_128>;
5178 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5179 int_x86_ssse3_pabs_d_128>;
5181 //===---------------------------------------------------------------------===//
5182 // SSSE3 - Packed Binary Operator Instructions
5183 //===---------------------------------------------------------------------===//
5185 def SSE_PHADDSUBD : OpndItins<
5186 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5188 def SSE_PHADDSUBSW : OpndItins<
5189 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5191 def SSE_PHADDSUBW : OpndItins<
5192 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5194 def SSE_PSHUFB : OpndItins<
5195 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5197 def SSE_PSIGN : OpndItins<
5198 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5200 def SSE_PMULHRSW : OpndItins<
5201 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5204 /// SS3I_binop_rm - Simple SSSE3 bin op
5205 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5206 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5207 X86MemOperand x86memop, OpndItins itins,
5209 let isCommutable = 1 in
5210 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5211 (ins RC:$src1, RC:$src2),
5213 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5214 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5215 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5217 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5218 (ins RC:$src1, x86memop:$src2),
5220 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5221 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5223 (OpVT (OpNode RC:$src1,
5224 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5227 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5228 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5229 Intrinsic IntId128, OpndItins itins,
5231 let isCommutable = 1 in
5232 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5233 (ins VR128:$src1, VR128:$src2),
5235 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5236 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5237 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5239 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5240 (ins VR128:$src1, i128mem:$src2),
5242 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5243 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5245 (IntId128 VR128:$src1,
5246 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5249 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5250 Intrinsic IntId256> {
5251 let isCommutable = 1 in
5252 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5253 (ins VR256:$src1, VR256:$src2),
5254 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5255 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5257 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5258 (ins VR256:$src1, i256mem:$src2),
5259 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5261 (IntId256 VR256:$src1,
5262 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5265 let ImmT = NoImm, Predicates = [HasAVX] in {
5266 let isCommutable = 0 in {
5267 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5268 memopv2i64, i128mem,
5269 SSE_PHADDSUBW, 0>, VEX_4V;
5270 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5271 memopv2i64, i128mem,
5272 SSE_PHADDSUBD, 0>, VEX_4V;
5273 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5274 memopv2i64, i128mem,
5275 SSE_PHADDSUBW, 0>, VEX_4V;
5276 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5277 memopv2i64, i128mem,
5278 SSE_PHADDSUBD, 0>, VEX_4V;
5279 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5280 memopv2i64, i128mem,
5281 SSE_PSIGN, 0>, VEX_4V;
5282 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5283 memopv2i64, i128mem,
5284 SSE_PSIGN, 0>, VEX_4V;
5285 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5286 memopv2i64, i128mem,
5287 SSE_PSIGN, 0>, VEX_4V;
5288 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5289 memopv2i64, i128mem,
5290 SSE_PSHUFB, 0>, VEX_4V;
5291 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5292 int_x86_ssse3_phadd_sw_128,
5293 SSE_PHADDSUBSW, 0>, VEX_4V;
5294 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5295 int_x86_ssse3_phsub_sw_128,
5296 SSE_PHADDSUBSW, 0>, VEX_4V;
5297 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5298 int_x86_ssse3_pmadd_ub_sw_128,
5299 SSE_PMADD, 0>, VEX_4V;
5301 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5302 int_x86_ssse3_pmul_hr_sw_128,
5303 SSE_PMULHRSW, 0>, VEX_4V;
5306 let ImmT = NoImm, Predicates = [HasAVX2] in {
5307 let isCommutable = 0 in {
5308 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5309 memopv4i64, i256mem,
5310 SSE_PHADDSUBW, 0>, VEX_4V;
5311 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5312 memopv4i64, i256mem,
5313 SSE_PHADDSUBW, 0>, VEX_4V;
5314 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5315 memopv4i64, i256mem,
5316 SSE_PHADDSUBW, 0>, VEX_4V;
5317 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5318 memopv4i64, i256mem,
5319 SSE_PHADDSUBW, 0>, VEX_4V;
5320 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5321 memopv4i64, i256mem,
5322 SSE_PHADDSUBW, 0>, VEX_4V;
5323 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5324 memopv4i64, i256mem,
5325 SSE_PHADDSUBW, 0>, VEX_4V;
5326 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5327 memopv4i64, i256mem,
5328 SSE_PHADDSUBW, 0>, VEX_4V;
5329 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5330 memopv4i64, i256mem,
5331 SSE_PHADDSUBW, 0>, VEX_4V;
5332 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5333 int_x86_avx2_phadd_sw>, VEX_4V;
5334 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5335 int_x86_avx2_phsub_sw>, VEX_4V;
5336 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5337 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5339 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5340 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5343 // None of these have i8 immediate fields.
5344 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5345 let isCommutable = 0 in {
5346 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5347 memopv2i64, i128mem, SSE_PHADDSUBW>;
5348 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5349 memopv2i64, i128mem, SSE_PHADDSUBD>;
5350 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5351 memopv2i64, i128mem, SSE_PHADDSUBW>;
5352 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5353 memopv2i64, i128mem, SSE_PHADDSUBD>;
5354 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5355 memopv2i64, i128mem, SSE_PSIGN>;
5356 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5357 memopv2i64, i128mem, SSE_PSIGN>;
5358 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5359 memopv2i64, i128mem, SSE_PSIGN>;
5360 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5361 memopv2i64, i128mem, SSE_PSHUFB>;
5362 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5363 int_x86_ssse3_phadd_sw_128,
5365 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5366 int_x86_ssse3_phsub_sw_128,
5368 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5369 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5371 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5372 int_x86_ssse3_pmul_hr_sw_128,
5376 //===---------------------------------------------------------------------===//
5377 // SSSE3 - Packed Align Instruction Patterns
5378 //===---------------------------------------------------------------------===//
5380 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5381 let neverHasSideEffects = 1 in {
5382 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5383 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5385 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5387 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5388 [], IIC_SSE_PALIGNR>, OpSize;
5390 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5391 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5393 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5395 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5396 [], IIC_SSE_PALIGNR>, OpSize;
5400 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5401 let neverHasSideEffects = 1 in {
5402 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5403 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5405 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5408 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5409 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5411 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5416 let Predicates = [HasAVX] in
5417 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5418 let Predicates = [HasAVX2] in
5419 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5420 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5421 defm PALIGN : ssse3_palign<"palignr">;
5423 let Predicates = [HasAVX2] in {
5424 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5425 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5426 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5427 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5428 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5429 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5430 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5431 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5434 let Predicates = [HasAVX] in {
5435 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5436 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5437 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5438 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5439 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5440 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5441 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5442 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5445 let Predicates = [HasSSSE3] in {
5446 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5447 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5448 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5449 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5450 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5451 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5452 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5453 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5456 //===---------------------------------------------------------------------===//
5457 // SSSE3 - Thread synchronization
5458 //===---------------------------------------------------------------------===//
5460 let usesCustomInserter = 1 in {
5461 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5462 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5463 Requires<[HasSSE3]>;
5464 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5465 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5466 Requires<[HasSSE3]>;
5469 let Uses = [EAX, ECX, EDX] in
5470 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5471 TB, Requires<[HasSSE3]>;
5472 let Uses = [ECX, EAX] in
5473 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5474 TB, Requires<[HasSSE3]>;
5476 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5477 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5479 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5480 Requires<[In32BitMode]>;
5481 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5482 Requires<[In64BitMode]>;
5484 //===----------------------------------------------------------------------===//
5485 // SSE4.1 - Packed Move with Sign/Zero Extend
5486 //===----------------------------------------------------------------------===//
5488 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5489 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5491 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5493 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5496 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5500 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5502 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5504 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5506 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5507 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5508 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5511 let Predicates = [HasAVX] in {
5512 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5514 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5516 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5518 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5520 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5522 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5526 let Predicates = [HasAVX2] in {
5527 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5528 int_x86_avx2_pmovsxbw>, VEX;
5529 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5530 int_x86_avx2_pmovsxwd>, VEX;
5531 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5532 int_x86_avx2_pmovsxdq>, VEX;
5533 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5534 int_x86_avx2_pmovzxbw>, VEX;
5535 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5536 int_x86_avx2_pmovzxwd>, VEX;
5537 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5538 int_x86_avx2_pmovzxdq>, VEX;
5541 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5542 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5543 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5544 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5545 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5546 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5548 let Predicates = [HasAVX] in {
5549 // Common patterns involving scalar load.
5550 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5551 (VPMOVSXBWrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5553 (VPMOVSXBWrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5556 (VPMOVSXWDrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5558 (VPMOVSXWDrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5561 (VPMOVSXDQrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5563 (VPMOVSXDQrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5566 (VPMOVZXBWrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5568 (VPMOVZXBWrm addr:$src)>;
5570 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5571 (VPMOVZXWDrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5573 (VPMOVZXWDrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5576 (VPMOVZXDQrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5578 (VPMOVZXDQrm addr:$src)>;
5581 let Predicates = [HasSSE41] in {
5582 // Common patterns involving scalar load.
5583 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5584 (PMOVSXBWrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5586 (PMOVSXBWrm addr:$src)>;
5588 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5589 (PMOVSXWDrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5591 (PMOVSXWDrm addr:$src)>;
5593 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5594 (PMOVSXDQrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5596 (PMOVSXDQrm addr:$src)>;
5598 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5599 (PMOVZXBWrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5601 (PMOVZXBWrm addr:$src)>;
5603 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5604 (PMOVZXWDrm addr:$src)>;
5605 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5606 (PMOVZXWDrm addr:$src)>;
5608 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5609 (PMOVZXDQrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5611 (PMOVZXDQrm addr:$src)>;
5614 let Predicates = [HasAVX2] in {
5615 let AddedComplexity = 15 in {
5616 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5617 (VPMOVZXDQYrr VR128:$src)>;
5618 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5619 (VPMOVZXWDYrr VR128:$src)>;
5622 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5623 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5626 let Predicates = [HasAVX] in {
5627 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5628 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5631 let Predicates = [HasSSE41] in {
5632 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5633 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5637 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5638 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5639 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5640 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5642 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5643 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5645 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5649 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5651 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5652 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5653 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5655 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5656 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5658 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5662 let Predicates = [HasAVX] in {
5663 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5665 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5667 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5669 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5673 let Predicates = [HasAVX2] in {
5674 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5675 int_x86_avx2_pmovsxbd>, VEX;
5676 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5677 int_x86_avx2_pmovsxwq>, VEX;
5678 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5679 int_x86_avx2_pmovzxbd>, VEX;
5680 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5681 int_x86_avx2_pmovzxwq>, VEX;
5684 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5685 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5686 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5687 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5689 let Predicates = [HasAVX] in {
5690 // Common patterns involving scalar load
5691 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5692 (VPMOVSXBDrm addr:$src)>;
5693 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5694 (VPMOVSXWQrm addr:$src)>;
5696 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5697 (VPMOVZXBDrm addr:$src)>;
5698 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5699 (VPMOVZXWQrm addr:$src)>;
5702 let Predicates = [HasSSE41] in {
5703 // Common patterns involving scalar load
5704 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5705 (PMOVSXBDrm addr:$src)>;
5706 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5707 (PMOVSXWQrm addr:$src)>;
5709 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5710 (PMOVZXBDrm addr:$src)>;
5711 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5712 (PMOVZXWQrm addr:$src)>;
5715 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5716 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5717 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5718 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5720 // Expecting a i16 load any extended to i32 value.
5721 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5722 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5723 [(set VR128:$dst, (IntId (bitconvert
5724 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5728 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5730 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5731 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5732 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5734 // Expecting a i16 load any extended to i32 value.
5735 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5736 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5737 [(set VR256:$dst, (IntId (bitconvert
5738 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5742 let Predicates = [HasAVX] in {
5743 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5745 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5748 let Predicates = [HasAVX2] in {
5749 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5750 int_x86_avx2_pmovsxbq>, VEX;
5751 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5752 int_x86_avx2_pmovzxbq>, VEX;
5754 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5755 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5757 let Predicates = [HasAVX] in {
5758 // Common patterns involving scalar load
5759 def : Pat<(int_x86_sse41_pmovsxbq
5760 (bitconvert (v4i32 (X86vzmovl
5761 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5762 (VPMOVSXBQrm addr:$src)>;
5764 def : Pat<(int_x86_sse41_pmovzxbq
5765 (bitconvert (v4i32 (X86vzmovl
5766 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5767 (VPMOVZXBQrm addr:$src)>;
5770 let Predicates = [HasSSE41] in {
5771 // Common patterns involving scalar load
5772 def : Pat<(int_x86_sse41_pmovsxbq
5773 (bitconvert (v4i32 (X86vzmovl
5774 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5775 (PMOVSXBQrm addr:$src)>;
5777 def : Pat<(int_x86_sse41_pmovzxbq
5778 (bitconvert (v4i32 (X86vzmovl
5779 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5780 (PMOVZXBQrm addr:$src)>;
5783 //===----------------------------------------------------------------------===//
5784 // SSE4.1 - Extract Instructions
5785 //===----------------------------------------------------------------------===//
5787 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5788 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5789 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5790 (ins VR128:$src1, i32i8imm:$src2),
5791 !strconcat(OpcodeStr,
5792 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5793 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5795 let neverHasSideEffects = 1, mayStore = 1 in
5796 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5797 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5798 !strconcat(OpcodeStr,
5799 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5802 // There's an AssertZext in the way of writing the store pattern
5803 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5806 let Predicates = [HasAVX] in {
5807 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5808 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5809 (ins VR128:$src1, i32i8imm:$src2),
5810 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5813 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5816 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5817 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5818 let neverHasSideEffects = 1, mayStore = 1 in
5819 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5820 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5821 !strconcat(OpcodeStr,
5822 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5825 // There's an AssertZext in the way of writing the store pattern
5826 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5829 let Predicates = [HasAVX] in
5830 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5832 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5835 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5836 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5837 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5838 (ins VR128:$src1, i32i8imm:$src2),
5839 !strconcat(OpcodeStr,
5840 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5842 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5843 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5844 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5845 !strconcat(OpcodeStr,
5846 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5847 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5848 addr:$dst)]>, OpSize;
5851 let Predicates = [HasAVX] in
5852 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5854 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5856 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5857 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5858 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5859 (ins VR128:$src1, i32i8imm:$src2),
5860 !strconcat(OpcodeStr,
5861 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5863 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5864 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5865 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5866 !strconcat(OpcodeStr,
5867 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5868 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5869 addr:$dst)]>, OpSize, REX_W;
5872 let Predicates = [HasAVX] in
5873 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5875 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5877 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5879 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5880 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5881 (ins VR128:$src1, i32i8imm:$src2),
5882 !strconcat(OpcodeStr,
5883 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5885 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5887 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5888 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5889 !strconcat(OpcodeStr,
5890 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5891 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5892 addr:$dst)]>, OpSize;
5895 let ExeDomain = SSEPackedSingle in {
5896 let Predicates = [HasAVX] in {
5897 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5898 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5899 (ins VR128:$src1, i32i8imm:$src2),
5900 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5903 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5906 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5907 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5910 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5912 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5915 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5916 Requires<[HasSSE41]>;
5918 //===----------------------------------------------------------------------===//
5919 // SSE4.1 - Insert Instructions
5920 //===----------------------------------------------------------------------===//
5922 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5923 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5924 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5926 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5928 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5930 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5931 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5932 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5934 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5936 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5938 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5939 imm:$src3))]>, OpSize;
5942 let Predicates = [HasAVX] in
5943 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5944 let Constraints = "$src1 = $dst" in
5945 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5947 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5948 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5949 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5951 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5953 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5955 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5957 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5958 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5960 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5962 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5964 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5965 imm:$src3)))]>, OpSize;
5968 let Predicates = [HasAVX] in
5969 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5970 let Constraints = "$src1 = $dst" in
5971 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5973 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5974 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5975 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5977 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5979 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5981 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5983 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5984 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5986 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5988 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5990 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5991 imm:$src3)))]>, OpSize;
5994 let Predicates = [HasAVX] in
5995 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5996 let Constraints = "$src1 = $dst" in
5997 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5999 // insertps has a few different modes, there's the first two here below which
6000 // are optimized inserts that won't zero arbitrary elements in the destination
6001 // vector. The next one matches the intrinsic and could zero arbitrary elements
6002 // in the target vector.
6003 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6004 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6005 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6007 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6009 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6011 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6013 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6014 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6016 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6018 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6020 (X86insrtps VR128:$src1,
6021 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6022 imm:$src3))]>, OpSize;
6025 let ExeDomain = SSEPackedSingle in {
6026 let Predicates = [HasAVX] in
6027 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6028 let Constraints = "$src1 = $dst" in
6029 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6032 //===----------------------------------------------------------------------===//
6033 // SSE4.1 - Round Instructions
6034 //===----------------------------------------------------------------------===//
6036 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6037 X86MemOperand x86memop, RegisterClass RC,
6038 PatFrag mem_frag32, PatFrag mem_frag64,
6039 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6040 let ExeDomain = SSEPackedSingle in {
6041 // Intrinsic operation, reg.
6042 // Vector intrinsic operation, reg
6043 def PSr : SS4AIi8<opcps, MRMSrcReg,
6044 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6045 !strconcat(OpcodeStr,
6046 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6047 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6050 // Vector intrinsic operation, mem
6051 def PSm : SS4AIi8<opcps, MRMSrcMem,
6052 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6053 !strconcat(OpcodeStr,
6054 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6056 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6058 } // ExeDomain = SSEPackedSingle
6060 let ExeDomain = SSEPackedDouble in {
6061 // Vector intrinsic operation, reg
6062 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6063 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6064 !strconcat(OpcodeStr,
6065 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6066 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6069 // Vector intrinsic operation, mem
6070 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6071 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6072 !strconcat(OpcodeStr,
6073 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6075 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6077 } // ExeDomain = SSEPackedDouble
6080 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6083 Intrinsic F64Int, bit Is2Addr = 1> {
6084 let ExeDomain = GenericDomain in {
6086 def SSr : SS4AIi8<opcss, MRMSrcReg,
6087 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6089 !strconcat(OpcodeStr,
6090 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6091 !strconcat(OpcodeStr,
6092 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6095 // Intrinsic operation, reg.
6096 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6097 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6099 !strconcat(OpcodeStr,
6100 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6101 !strconcat(OpcodeStr,
6102 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6103 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6106 // Intrinsic operation, mem.
6107 def SSm : SS4AIi8<opcss, MRMSrcMem,
6108 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6110 !strconcat(OpcodeStr,
6111 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6112 !strconcat(OpcodeStr,
6113 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6115 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6119 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6120 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6122 !strconcat(OpcodeStr,
6123 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6124 !strconcat(OpcodeStr,
6125 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6128 // Intrinsic operation, reg.
6129 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6130 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6132 !strconcat(OpcodeStr,
6133 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6134 !strconcat(OpcodeStr,
6135 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6136 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6139 // Intrinsic operation, mem.
6140 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6141 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6143 !strconcat(OpcodeStr,
6144 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6145 !strconcat(OpcodeStr,
6146 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6148 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6150 } // ExeDomain = GenericDomain
6153 // FP round - roundss, roundps, roundsd, roundpd
6154 let Predicates = [HasAVX] in {
6156 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6157 memopv4f32, memopv2f64,
6158 int_x86_sse41_round_ps,
6159 int_x86_sse41_round_pd>, VEX;
6160 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6161 memopv8f32, memopv4f64,
6162 int_x86_avx_round_ps_256,
6163 int_x86_avx_round_pd_256>, VEX;
6164 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6165 int_x86_sse41_round_ss,
6166 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6168 def : Pat<(ffloor FR32:$src),
6169 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6170 def : Pat<(f64 (ffloor FR64:$src)),
6171 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6172 def : Pat<(f32 (fnearbyint FR32:$src)),
6173 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6174 def : Pat<(f64 (fnearbyint FR64:$src)),
6175 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6176 def : Pat<(f32 (fceil FR32:$src)),
6177 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6178 def : Pat<(f64 (fceil FR64:$src)),
6179 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6180 def : Pat<(f32 (frint FR32:$src)),
6181 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6182 def : Pat<(f64 (frint FR64:$src)),
6183 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6184 def : Pat<(f32 (ftrunc FR32:$src)),
6185 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6186 def : Pat<(f64 (ftrunc FR64:$src)),
6187 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6190 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6191 memopv4f32, memopv2f64,
6192 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6193 let Constraints = "$src1 = $dst" in
6194 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6195 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6197 def : Pat<(ffloor FR32:$src),
6198 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6199 def : Pat<(f64 (ffloor FR64:$src)),
6200 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6201 def : Pat<(f32 (fnearbyint FR32:$src)),
6202 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6203 def : Pat<(f64 (fnearbyint FR64:$src)),
6204 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6205 def : Pat<(f32 (fceil FR32:$src)),
6206 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6207 def : Pat<(f64 (fceil FR64:$src)),
6208 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6209 def : Pat<(f32 (frint FR32:$src)),
6210 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6211 def : Pat<(f64 (frint FR64:$src)),
6212 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6213 def : Pat<(f32 (ftrunc FR32:$src)),
6214 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6215 def : Pat<(f64 (ftrunc FR64:$src)),
6216 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6218 //===----------------------------------------------------------------------===//
6219 // SSE4.1 - Packed Bit Test
6220 //===----------------------------------------------------------------------===//
6222 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6223 // the intel intrinsic that corresponds to this.
6224 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6225 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6226 "vptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6229 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6230 "vptest\t{$src2, $src1|$src1, $src2}",
6231 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6234 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6235 "vptest\t{$src2, $src1|$src1, $src2}",
6236 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6238 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6239 "vptest\t{$src2, $src1|$src1, $src2}",
6240 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6244 let Defs = [EFLAGS] in {
6245 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6246 "ptest\t{$src2, $src1|$src1, $src2}",
6247 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6249 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6250 "ptest\t{$src2, $src1|$src1, $src2}",
6251 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6255 // The bit test instructions below are AVX only
6256 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6257 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6258 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6259 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6260 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6261 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6262 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6263 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6267 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6268 let ExeDomain = SSEPackedSingle in {
6269 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6270 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6272 let ExeDomain = SSEPackedDouble in {
6273 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6274 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6278 //===----------------------------------------------------------------------===//
6279 // SSE4.1 - Misc Instructions
6280 //===----------------------------------------------------------------------===//
6282 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6283 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6284 "popcnt{w}\t{$src, $dst|$dst, $src}",
6285 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6287 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6288 "popcnt{w}\t{$src, $dst|$dst, $src}",
6289 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6290 (implicit EFLAGS)]>, OpSize, XS;
6292 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6293 "popcnt{l}\t{$src, $dst|$dst, $src}",
6294 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6296 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6297 "popcnt{l}\t{$src, $dst|$dst, $src}",
6298 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6299 (implicit EFLAGS)]>, XS;
6301 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6302 "popcnt{q}\t{$src, $dst|$dst, $src}",
6303 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6305 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6306 "popcnt{q}\t{$src, $dst|$dst, $src}",
6307 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6308 (implicit EFLAGS)]>, XS;
6313 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6314 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6315 Intrinsic IntId128> {
6316 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6318 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6319 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6320 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6322 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6325 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6328 let Predicates = [HasAVX] in
6329 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6330 int_x86_sse41_phminposuw>, VEX;
6331 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6332 int_x86_sse41_phminposuw>;
6334 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6335 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6336 Intrinsic IntId128, bit Is2Addr = 1> {
6337 let isCommutable = 1 in
6338 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6339 (ins VR128:$src1, VR128:$src2),
6341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6343 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6344 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6345 (ins VR128:$src1, i128mem:$src2),
6347 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6350 (IntId128 VR128:$src1,
6351 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6354 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6355 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6356 Intrinsic IntId256> {
6357 let isCommutable = 1 in
6358 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6359 (ins VR256:$src1, VR256:$src2),
6360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6361 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6362 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6363 (ins VR256:$src1, i256mem:$src2),
6364 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6366 (IntId256 VR256:$src1,
6367 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6370 let Predicates = [HasAVX] in {
6371 let isCommutable = 0 in
6372 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6374 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6376 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6378 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6380 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6382 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6384 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6386 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6388 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6390 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6394 let Predicates = [HasAVX2] in {
6395 let isCommutable = 0 in
6396 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6397 int_x86_avx2_packusdw>, VEX_4V;
6398 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6399 int_x86_avx2_pmins_b>, VEX_4V;
6400 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6401 int_x86_avx2_pmins_d>, VEX_4V;
6402 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6403 int_x86_avx2_pminu_d>, VEX_4V;
6404 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6405 int_x86_avx2_pminu_w>, VEX_4V;
6406 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6407 int_x86_avx2_pmaxs_b>, VEX_4V;
6408 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6409 int_x86_avx2_pmaxs_d>, VEX_4V;
6410 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6411 int_x86_avx2_pmaxu_d>, VEX_4V;
6412 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6413 int_x86_avx2_pmaxu_w>, VEX_4V;
6414 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6415 int_x86_avx2_pmul_dq>, VEX_4V;
6418 let Constraints = "$src1 = $dst" in {
6419 let isCommutable = 0 in
6420 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6421 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6422 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6423 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6424 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6425 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6426 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6427 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6428 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6429 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6432 /// SS48I_binop_rm - Simple SSE41 binary operator.
6433 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6434 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6435 X86MemOperand x86memop, bit Is2Addr = 1> {
6436 let isCommutable = 1 in
6437 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6438 (ins RC:$src1, RC:$src2),
6440 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6441 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6442 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6443 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6444 (ins RC:$src1, x86memop:$src2),
6446 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6447 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6449 (OpVT (OpNode RC:$src1,
6450 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6453 let Predicates = [HasAVX] in {
6454 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6455 memopv2i64, i128mem, 0>, VEX_4V;
6456 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6457 memopv2i64, i128mem, 0>, VEX_4V;
6459 let Predicates = [HasAVX2] in {
6460 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6461 memopv4i64, i256mem, 0>, VEX_4V;
6462 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6463 memopv4i64, i256mem, 0>, VEX_4V;
6466 let Constraints = "$src1 = $dst" in {
6467 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6468 memopv2i64, i128mem>;
6469 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6470 memopv2i64, i128mem>;
6473 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6474 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6475 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6476 X86MemOperand x86memop, bit Is2Addr = 1> {
6477 let isCommutable = 1 in
6478 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6479 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6481 !strconcat(OpcodeStr,
6482 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6483 !strconcat(OpcodeStr,
6484 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6485 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6487 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6488 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6490 !strconcat(OpcodeStr,
6491 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6492 !strconcat(OpcodeStr,
6493 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6496 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6500 let Predicates = [HasAVX] in {
6501 let isCommutable = 0 in {
6502 let ExeDomain = SSEPackedSingle in {
6503 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6504 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6505 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6506 int_x86_avx_blend_ps_256, VR256, memopv8f32, f256mem, 0>, VEX_4V;
6508 let ExeDomain = SSEPackedDouble in {
6509 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6510 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6511 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6512 int_x86_avx_blend_pd_256, VR256, memopv4f64, f256mem, 0>, VEX_4V;
6514 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6515 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6516 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6517 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6519 let ExeDomain = SSEPackedSingle in
6520 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6521 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6522 let ExeDomain = SSEPackedDouble in
6523 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6524 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6525 let ExeDomain = SSEPackedSingle in
6526 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6527 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6530 let Predicates = [HasAVX2] in {
6531 let isCommutable = 0 in {
6532 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6533 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6534 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6535 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6539 let Constraints = "$src1 = $dst" in {
6540 let isCommutable = 0 in {
6541 let ExeDomain = SSEPackedSingle in
6542 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6543 VR128, memopv4f32, f128mem>;
6544 let ExeDomain = SSEPackedDouble in
6545 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6546 VR128, memopv2f64, f128mem>;
6547 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6548 VR128, memopv2i64, i128mem>;
6549 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6550 VR128, memopv2i64, i128mem>;
6552 let ExeDomain = SSEPackedSingle in
6553 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6554 VR128, memopv4f32, f128mem>;
6555 let ExeDomain = SSEPackedDouble in
6556 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6557 VR128, memopv2f64, f128mem>;
6560 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6561 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6562 RegisterClass RC, X86MemOperand x86memop,
6563 PatFrag mem_frag, Intrinsic IntId> {
6564 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6565 (ins RC:$src1, RC:$src2, RC:$src3),
6566 !strconcat(OpcodeStr,
6567 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6568 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6569 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6571 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6572 (ins RC:$src1, x86memop:$src2, RC:$src3),
6573 !strconcat(OpcodeStr,
6574 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6576 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6578 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6581 let Predicates = [HasAVX] in {
6582 let ExeDomain = SSEPackedDouble in {
6583 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6584 memopv2f64, int_x86_sse41_blendvpd>;
6585 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6586 memopv4f64, int_x86_avx_blendv_pd_256>;
6587 } // ExeDomain = SSEPackedDouble
6588 let ExeDomain = SSEPackedSingle in {
6589 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6590 memopv4f32, int_x86_sse41_blendvps>;
6591 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6592 memopv8f32, int_x86_avx_blendv_ps_256>;
6593 } // ExeDomain = SSEPackedSingle
6594 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6595 memopv2i64, int_x86_sse41_pblendvb>;
6598 let Predicates = [HasAVX2] in {
6599 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6600 memopv4i64, int_x86_avx2_pblendvb>;
6603 let Predicates = [HasAVX] in {
6604 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6605 (v16i8 VR128:$src2))),
6606 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6607 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6608 (v4i32 VR128:$src2))),
6609 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6610 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6611 (v4f32 VR128:$src2))),
6612 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6613 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6614 (v2i64 VR128:$src2))),
6615 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6616 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6617 (v2f64 VR128:$src2))),
6618 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6619 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6620 (v8i32 VR256:$src2))),
6621 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6622 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6623 (v8f32 VR256:$src2))),
6624 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6625 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6626 (v4i64 VR256:$src2))),
6627 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6628 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6629 (v4f64 VR256:$src2))),
6630 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6632 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6634 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6635 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6637 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6639 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6641 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6642 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6644 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6645 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6647 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6650 let Predicates = [HasAVX2] in {
6651 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6652 (v32i8 VR256:$src2))),
6653 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6654 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6656 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6659 /// SS41I_ternary_int - SSE 4.1 ternary operator
6660 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6661 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6662 X86MemOperand x86memop, Intrinsic IntId> {
6663 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6664 (ins VR128:$src1, VR128:$src2),
6665 !strconcat(OpcodeStr,
6666 "\t{$src2, $dst|$dst, $src2}"),
6667 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6670 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6671 (ins VR128:$src1, x86memop:$src2),
6672 !strconcat(OpcodeStr,
6673 "\t{$src2, $dst|$dst, $src2}"),
6676 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6680 let ExeDomain = SSEPackedDouble in
6681 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6682 int_x86_sse41_blendvpd>;
6683 let ExeDomain = SSEPackedSingle in
6684 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6685 int_x86_sse41_blendvps>;
6686 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6687 int_x86_sse41_pblendvb>;
6689 // Aliases with the implicit xmm0 argument
6690 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6691 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6692 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6693 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6694 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6695 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6696 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6697 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6698 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6699 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6700 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6701 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6703 let Predicates = [HasSSE41] in {
6704 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6705 (v16i8 VR128:$src2))),
6706 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6707 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6708 (v4i32 VR128:$src2))),
6709 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6710 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6711 (v4f32 VR128:$src2))),
6712 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6713 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6714 (v2i64 VR128:$src2))),
6715 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6716 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6717 (v2f64 VR128:$src2))),
6718 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6720 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6722 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6723 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6725 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6726 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6728 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6732 let Predicates = [HasAVX] in
6733 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6734 "vmovntdqa\t{$src, $dst|$dst, $src}",
6735 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6737 let Predicates = [HasAVX2] in
6738 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6739 "vmovntdqa\t{$src, $dst|$dst, $src}",
6740 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6742 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6743 "movntdqa\t{$src, $dst|$dst, $src}",
6744 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6747 //===----------------------------------------------------------------------===//
6748 // SSE4.2 - Compare Instructions
6749 //===----------------------------------------------------------------------===//
6751 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6752 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6753 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6754 X86MemOperand x86memop, bit Is2Addr = 1> {
6755 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6756 (ins RC:$src1, RC:$src2),
6758 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6759 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6760 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6762 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6763 (ins RC:$src1, x86memop:$src2),
6765 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6766 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6768 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6771 let Predicates = [HasAVX] in
6772 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6773 memopv2i64, i128mem, 0>, VEX_4V;
6775 let Predicates = [HasAVX2] in
6776 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6777 memopv4i64, i256mem, 0>, VEX_4V;
6779 let Constraints = "$src1 = $dst" in
6780 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6781 memopv2i64, i128mem>;
6783 //===----------------------------------------------------------------------===//
6784 // SSE4.2 - String/text Processing Instructions
6785 //===----------------------------------------------------------------------===//
6787 // Packed Compare Implicit Length Strings, Return Mask
6788 multiclass pseudo_pcmpistrm<string asm> {
6789 def REG : PseudoI<(outs VR128:$dst),
6790 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6791 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6793 def MEM : PseudoI<(outs VR128:$dst),
6794 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6795 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6796 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6799 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6800 let AddedComplexity = 1 in
6801 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6802 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6805 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6806 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6807 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6808 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6810 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6811 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6812 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6815 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6816 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6817 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6818 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6820 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6821 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6822 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6825 // Packed Compare Explicit Length Strings, Return Mask
6826 multiclass pseudo_pcmpestrm<string asm> {
6827 def REG : PseudoI<(outs VR128:$dst),
6828 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6829 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6830 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6831 def MEM : PseudoI<(outs VR128:$dst),
6832 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6833 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6834 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6837 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6838 let AddedComplexity = 1 in
6839 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6840 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6843 let Predicates = [HasAVX],
6844 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6845 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6846 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6847 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6849 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6850 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6851 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6854 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6855 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6856 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6857 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6859 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6860 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6861 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6864 // Packed Compare Implicit Length Strings, Return Index
6865 let Defs = [ECX, EFLAGS] in {
6866 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6867 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6868 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6869 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6870 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6871 (implicit EFLAGS)]>, OpSize;
6872 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6873 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6874 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6875 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6876 (implicit EFLAGS)]>, OpSize;
6880 let Predicates = [HasAVX] in {
6881 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6883 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6885 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6887 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6889 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6891 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6895 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6896 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6897 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6898 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6899 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6900 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6902 // Packed Compare Explicit Length Strings, Return Index
6903 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6904 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6905 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6906 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6907 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6908 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6909 (implicit EFLAGS)]>, OpSize;
6910 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6911 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6912 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6914 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6915 (implicit EFLAGS)]>, OpSize;
6919 let Predicates = [HasAVX] in {
6920 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6922 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6924 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6926 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6928 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6930 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6934 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6935 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6936 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6937 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6938 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6939 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6941 //===----------------------------------------------------------------------===//
6942 // SSE4.2 - CRC Instructions
6943 //===----------------------------------------------------------------------===//
6945 // No CRC instructions have AVX equivalents
6947 // crc intrinsic instruction
6948 // This set of instructions are only rm, the only difference is the size
6950 let Constraints = "$src1 = $dst" in {
6951 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6952 (ins GR32:$src1, i8mem:$src2),
6953 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6955 (int_x86_sse42_crc32_32_8 GR32:$src1,
6956 (load addr:$src2)))]>;
6957 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6958 (ins GR32:$src1, GR8:$src2),
6959 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6961 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6962 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6963 (ins GR32:$src1, i16mem:$src2),
6964 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6966 (int_x86_sse42_crc32_32_16 GR32:$src1,
6967 (load addr:$src2)))]>,
6969 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6970 (ins GR32:$src1, GR16:$src2),
6971 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6973 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6975 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6976 (ins GR32:$src1, i32mem:$src2),
6977 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6979 (int_x86_sse42_crc32_32_32 GR32:$src1,
6980 (load addr:$src2)))]>;
6981 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6982 (ins GR32:$src1, GR32:$src2),
6983 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6985 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6986 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6987 (ins GR64:$src1, i8mem:$src2),
6988 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_64_8 GR64:$src1,
6991 (load addr:$src2)))]>,
6993 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6994 (ins GR64:$src1, GR8:$src2),
6995 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6997 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6999 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7000 (ins GR64:$src1, i64mem:$src2),
7001 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7003 (int_x86_sse42_crc32_64_64 GR64:$src1,
7004 (load addr:$src2)))]>,
7006 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7007 (ins GR64:$src1, GR64:$src2),
7008 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7010 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7014 //===----------------------------------------------------------------------===//
7015 // AES-NI Instructions
7016 //===----------------------------------------------------------------------===//
7018 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7019 Intrinsic IntId128, bit Is2Addr = 1> {
7020 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7021 (ins VR128:$src1, VR128:$src2),
7023 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7024 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7025 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7027 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7028 (ins VR128:$src1, i128mem:$src2),
7030 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7031 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7033 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7036 // Perform One Round of an AES Encryption/Decryption Flow
7037 let Predicates = [HasAVX, HasAES] in {
7038 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7039 int_x86_aesni_aesenc, 0>, VEX_4V;
7040 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7041 int_x86_aesni_aesenclast, 0>, VEX_4V;
7042 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7043 int_x86_aesni_aesdec, 0>, VEX_4V;
7044 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7045 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7048 let Constraints = "$src1 = $dst" in {
7049 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7050 int_x86_aesni_aesenc>;
7051 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7052 int_x86_aesni_aesenclast>;
7053 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7054 int_x86_aesni_aesdec>;
7055 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7056 int_x86_aesni_aesdeclast>;
7059 // Perform the AES InvMixColumn Transformation
7060 let Predicates = [HasAVX, HasAES] in {
7061 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7063 "vaesimc\t{$src1, $dst|$dst, $src1}",
7065 (int_x86_aesni_aesimc VR128:$src1))]>,
7067 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7068 (ins i128mem:$src1),
7069 "vaesimc\t{$src1, $dst|$dst, $src1}",
7070 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7073 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7075 "aesimc\t{$src1, $dst|$dst, $src1}",
7077 (int_x86_aesni_aesimc VR128:$src1))]>,
7079 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7080 (ins i128mem:$src1),
7081 "aesimc\t{$src1, $dst|$dst, $src1}",
7082 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7085 // AES Round Key Generation Assist
7086 let Predicates = [HasAVX, HasAES] in {
7087 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7088 (ins VR128:$src1, i8imm:$src2),
7089 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7091 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7093 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7094 (ins i128mem:$src1, i8imm:$src2),
7095 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7097 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7100 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7101 (ins VR128:$src1, i8imm:$src2),
7102 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7104 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7106 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7107 (ins i128mem:$src1, i8imm:$src2),
7108 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7110 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7113 //===----------------------------------------------------------------------===//
7114 // PCLMUL Instructions
7115 //===----------------------------------------------------------------------===//
7117 // AVX carry-less Multiplication instructions
7118 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7119 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7120 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7122 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7124 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7125 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7126 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7127 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7128 (memopv2i64 addr:$src2), imm:$src3))]>;
7130 // Carry-less Multiplication instructions
7131 let Constraints = "$src1 = $dst" in {
7132 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7133 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7134 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7136 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7138 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7139 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7140 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7141 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7142 (memopv2i64 addr:$src2), imm:$src3))]>;
7143 } // Constraints = "$src1 = $dst"
7146 multiclass pclmul_alias<string asm, int immop> {
7147 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7148 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7150 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7151 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7153 def : InstAlias<!strconcat("vpclmul", asm,
7154 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7155 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7157 def : InstAlias<!strconcat("vpclmul", asm,
7158 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7159 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7161 defm : pclmul_alias<"hqhq", 0x11>;
7162 defm : pclmul_alias<"hqlq", 0x01>;
7163 defm : pclmul_alias<"lqhq", 0x10>;
7164 defm : pclmul_alias<"lqlq", 0x00>;
7166 //===----------------------------------------------------------------------===//
7167 // SSE4A Instructions
7168 //===----------------------------------------------------------------------===//
7170 let Predicates = [HasSSE4A] in {
7172 let Constraints = "$src = $dst" in {
7173 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7174 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7175 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7176 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7177 imm:$idx))]>, TB, OpSize;
7178 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7179 (ins VR128:$src, VR128:$mask),
7180 "extrq\t{$mask, $src|$src, $mask}",
7181 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7182 VR128:$mask))]>, TB, OpSize;
7184 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7185 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7186 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7187 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7188 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7189 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7190 (ins VR128:$src, VR128:$mask),
7191 "insertq\t{$mask, $src|$src, $mask}",
7192 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7193 VR128:$mask))]>, XD;
7196 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7197 "movntss\t{$src, $dst|$dst, $src}",
7198 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7200 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7201 "movntsd\t{$src, $dst|$dst, $src}",
7202 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7205 //===----------------------------------------------------------------------===//
7207 //===----------------------------------------------------------------------===//
7209 //===----------------------------------------------------------------------===//
7210 // VBROADCAST - Load from memory and broadcast to all elements of the
7211 // destination operand
7213 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7214 X86MemOperand x86memop, Intrinsic Int> :
7215 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7216 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7217 [(set RC:$dst, (Int addr:$src))]>, VEX;
7219 // AVX2 adds register forms
7220 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7222 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7223 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7224 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7226 let ExeDomain = SSEPackedSingle in {
7227 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7228 int_x86_avx_vbroadcast_ss>;
7229 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7230 int_x86_avx_vbroadcast_ss_256>;
7232 let ExeDomain = SSEPackedDouble in
7233 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7234 int_x86_avx_vbroadcast_sd_256>;
7235 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7236 int_x86_avx_vbroadcastf128_pd_256>;
7238 let ExeDomain = SSEPackedSingle in {
7239 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7240 int_x86_avx2_vbroadcast_ss_ps>;
7241 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7242 int_x86_avx2_vbroadcast_ss_ps_256>;
7244 let ExeDomain = SSEPackedDouble in
7245 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7246 int_x86_avx2_vbroadcast_sd_pd_256>;
7248 let Predicates = [HasAVX2] in
7249 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7250 int_x86_avx2_vbroadcasti128>;
7252 let Predicates = [HasAVX] in
7253 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7254 (VBROADCASTF128 addr:$src)>;
7257 //===----------------------------------------------------------------------===//
7258 // VINSERTF128 - Insert packed floating-point values
7260 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7261 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7262 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7263 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7266 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7267 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7268 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7272 let Predicates = [HasAVX] in {
7273 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7275 (VINSERTF128rr VR256:$src1, VR128:$src2,
7276 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7277 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7279 (VINSERTF128rr VR256:$src1, VR128:$src2,
7280 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7281 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7283 (VINSERTF128rr VR256:$src1, VR128:$src2,
7284 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7285 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7287 (VINSERTF128rr VR256:$src1, VR128:$src2,
7288 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7289 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7291 (VINSERTF128rr VR256:$src1, VR128:$src2,
7292 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7293 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7295 (VINSERTF128rr VR256:$src1, VR128:$src2,
7296 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7298 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7300 (VINSERTF128rm VR256:$src1, addr:$src2,
7301 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7302 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7304 (VINSERTF128rm VR256:$src1, addr:$src2,
7305 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7306 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7308 (VINSERTF128rm VR256:$src1, addr:$src2,
7309 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7312 //===----------------------------------------------------------------------===//
7313 // VEXTRACTF128 - Extract packed floating-point values
7315 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7316 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7317 (ins VR256:$src1, i8imm:$src2),
7318 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7321 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7322 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7323 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7327 // Extract and store.
7328 let Predicates = [HasAVX] in {
7329 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7330 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7331 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7332 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7333 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7334 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7336 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7337 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7338 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7339 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7340 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7341 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7345 let Predicates = [HasAVX] in {
7346 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7347 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7348 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7349 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7350 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7351 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7353 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7354 (v4f32 (VEXTRACTF128rr
7355 (v8f32 VR256:$src1),
7356 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7357 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7358 (v2f64 (VEXTRACTF128rr
7359 (v4f64 VR256:$src1),
7360 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7361 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7362 (v2i64 (VEXTRACTF128rr
7363 (v4i64 VR256:$src1),
7364 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7365 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7366 (v4i32 (VEXTRACTF128rr
7367 (v8i32 VR256:$src1),
7368 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7369 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7370 (v8i16 (VEXTRACTF128rr
7371 (v16i16 VR256:$src1),
7372 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7373 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7374 (v16i8 (VEXTRACTF128rr
7375 (v32i8 VR256:$src1),
7376 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7379 //===----------------------------------------------------------------------===//
7380 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7382 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7383 Intrinsic IntLd, Intrinsic IntLd256,
7384 Intrinsic IntSt, Intrinsic IntSt256> {
7385 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7386 (ins VR128:$src1, f128mem:$src2),
7387 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7388 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7390 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7391 (ins VR256:$src1, f256mem:$src2),
7392 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7393 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7395 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7396 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7398 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7399 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7400 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7401 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7402 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7405 let ExeDomain = SSEPackedSingle in
7406 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7407 int_x86_avx_maskload_ps,
7408 int_x86_avx_maskload_ps_256,
7409 int_x86_avx_maskstore_ps,
7410 int_x86_avx_maskstore_ps_256>;
7411 let ExeDomain = SSEPackedDouble in
7412 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7413 int_x86_avx_maskload_pd,
7414 int_x86_avx_maskload_pd_256,
7415 int_x86_avx_maskstore_pd,
7416 int_x86_avx_maskstore_pd_256>;
7418 //===----------------------------------------------------------------------===//
7419 // VPERMIL - Permute Single and Double Floating-Point Values
7421 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7422 RegisterClass RC, X86MemOperand x86memop_f,
7423 X86MemOperand x86memop_i, PatFrag i_frag,
7424 Intrinsic IntVar, ValueType vt> {
7425 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7426 (ins RC:$src1, RC:$src2),
7427 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7428 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7429 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7430 (ins RC:$src1, x86memop_i:$src2),
7431 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7432 [(set RC:$dst, (IntVar RC:$src1,
7433 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7435 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7436 (ins RC:$src1, i8imm:$src2),
7437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7438 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7439 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7440 (ins x86memop_f:$src1, i8imm:$src2),
7441 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7443 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7446 let ExeDomain = SSEPackedSingle in {
7447 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7448 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7449 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7450 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7452 let ExeDomain = SSEPackedDouble in {
7453 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7454 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7455 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7456 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7459 let Predicates = [HasAVX] in {
7460 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7461 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7462 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7463 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7464 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7466 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7467 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7468 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7470 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7471 (VPERMILPDri VR128:$src1, imm:$imm)>;
7472 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7473 (VPERMILPDmi addr:$src1, imm:$imm)>;
7476 //===----------------------------------------------------------------------===//
7477 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7479 let ExeDomain = SSEPackedSingle in {
7480 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7481 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7482 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7483 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7484 (i8 imm:$src3))))]>, VEX_4V;
7485 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7486 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7487 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7488 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7489 (i8 imm:$src3)))]>, VEX_4V;
7492 let Predicates = [HasAVX] in {
7493 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7494 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7495 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7496 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7497 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7498 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7499 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7500 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7501 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7502 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7504 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7505 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7506 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7507 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7508 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7509 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7510 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7511 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7512 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7513 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7514 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7515 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7516 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7517 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7518 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7519 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7520 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7521 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7524 //===----------------------------------------------------------------------===//
7525 // VZERO - Zero YMM registers
7527 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7528 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7529 // Zero All YMM registers
7530 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7531 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7533 // Zero Upper bits of YMM registers
7534 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7535 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7538 //===----------------------------------------------------------------------===//
7539 // Half precision conversion instructions
7540 //===----------------------------------------------------------------------===//
7541 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7542 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7543 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7544 [(set RC:$dst, (Int VR128:$src))]>,
7546 let neverHasSideEffects = 1, mayLoad = 1 in
7547 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7548 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7551 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7552 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7553 (ins RC:$src1, i32i8imm:$src2),
7554 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7555 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7557 let neverHasSideEffects = 1, mayStore = 1 in
7558 def mr : Ii8<0x1D, MRMDestMem, (outs),
7559 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7560 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7564 let Predicates = [HasAVX, HasF16C] in {
7565 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7566 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7567 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7568 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7571 //===----------------------------------------------------------------------===//
7572 // AVX2 Instructions
7573 //===----------------------------------------------------------------------===//
7575 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7576 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7577 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7578 X86MemOperand x86memop> {
7579 let isCommutable = 1 in
7580 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7581 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7582 !strconcat(OpcodeStr,
7583 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7584 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7586 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7587 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7588 !strconcat(OpcodeStr,
7589 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7592 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7596 let isCommutable = 0 in {
7597 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7598 VR128, memopv2i64, i128mem>;
7599 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7600 VR256, memopv4i64, i256mem>;
7603 //===----------------------------------------------------------------------===//
7604 // VPBROADCAST - Load from memory and broadcast to all elements of the
7605 // destination operand
7607 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7608 X86MemOperand x86memop, PatFrag ld_frag,
7609 Intrinsic Int128, Intrinsic Int256> {
7610 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7611 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7612 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7613 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7614 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7616 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7617 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7618 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7619 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7620 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7621 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7623 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7626 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7627 int_x86_avx2_pbroadcastb_128,
7628 int_x86_avx2_pbroadcastb_256>;
7629 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7630 int_x86_avx2_pbroadcastw_128,
7631 int_x86_avx2_pbroadcastw_256>;
7632 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7633 int_x86_avx2_pbroadcastd_128,
7634 int_x86_avx2_pbroadcastd_256>;
7635 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7636 int_x86_avx2_pbroadcastq_128,
7637 int_x86_avx2_pbroadcastq_256>;
7639 let Predicates = [HasAVX2] in {
7640 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7641 (VPBROADCASTBrm addr:$src)>;
7642 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7643 (VPBROADCASTBYrm addr:$src)>;
7644 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7645 (VPBROADCASTWrm addr:$src)>;
7646 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7647 (VPBROADCASTWYrm addr:$src)>;
7648 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7649 (VPBROADCASTDrm addr:$src)>;
7650 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7651 (VPBROADCASTDYrm addr:$src)>;
7652 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7653 (VPBROADCASTQrm addr:$src)>;
7654 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7655 (VPBROADCASTQYrm addr:$src)>;
7657 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7658 (VPBROADCASTBrr VR128:$src)>;
7659 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7660 (VPBROADCASTBYrr VR128:$src)>;
7661 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7662 (VPBROADCASTWrr VR128:$src)>;
7663 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7664 (VPBROADCASTWYrr VR128:$src)>;
7665 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7666 (VPBROADCASTDrr VR128:$src)>;
7667 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7668 (VPBROADCASTDYrr VR128:$src)>;
7669 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7670 (VPBROADCASTQrr VR128:$src)>;
7671 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7672 (VPBROADCASTQYrr VR128:$src)>;
7673 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7674 (VBROADCASTSSrr VR128:$src)>;
7675 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7676 (VBROADCASTSSYrr VR128:$src)>;
7677 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7678 (VPBROADCASTQrr VR128:$src)>;
7679 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7680 (VBROADCASTSDYrr VR128:$src)>;
7682 // Provide fallback in case the load node that is used in the patterns above
7683 // is used by additional users, which prevents the pattern selection.
7684 let AddedComplexity = 20 in {
7685 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7686 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7687 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7688 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7689 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7690 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7692 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7693 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7694 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7695 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7696 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7697 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7701 // AVX1 broadcast patterns
7702 let Predicates = [HasAVX] in {
7703 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7704 (VBROADCASTSSYrm addr:$src)>;
7705 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7706 (VBROADCASTSDYrm addr:$src)>;
7707 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7708 (VBROADCASTSSYrm addr:$src)>;
7709 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7710 (VBROADCASTSDYrm addr:$src)>;
7711 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7712 (VBROADCASTSSrm addr:$src)>;
7713 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7714 (VBROADCASTSSrm addr:$src)>;
7716 // Provide fallback in case the load node that is used in the patterns above
7717 // is used by additional users, which prevents the pattern selection.
7718 let AddedComplexity = 20 in {
7719 // 128bit broadcasts:
7720 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7721 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7722 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7723 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7724 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7725 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7726 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7727 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7728 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7729 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7731 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7732 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7733 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7734 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7735 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7736 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7737 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7738 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7739 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7740 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7744 //===----------------------------------------------------------------------===//
7745 // VPERM - Permute instructions
7748 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7750 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7751 (ins VR256:$src1, VR256:$src2),
7752 !strconcat(OpcodeStr,
7753 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7755 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7756 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7757 (ins VR256:$src1, i256mem:$src2),
7758 !strconcat(OpcodeStr,
7759 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7761 (OpVT (X86VPermv VR256:$src1,
7762 (bitconvert (mem_frag addr:$src2)))))]>,
7766 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7767 let ExeDomain = SSEPackedSingle in
7768 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7770 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7772 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7773 (ins VR256:$src1, i8imm:$src2),
7774 !strconcat(OpcodeStr,
7775 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7777 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7778 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7779 (ins i256mem:$src1, i8imm:$src2),
7780 !strconcat(OpcodeStr,
7781 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7783 (OpVT (X86VPermi (mem_frag addr:$src1),
7784 (i8 imm:$src2))))]>, VEX;
7787 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7788 let ExeDomain = SSEPackedDouble in
7789 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7791 //===----------------------------------------------------------------------===//
7792 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7794 let AddedComplexity = 1 in {
7795 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7796 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7797 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7798 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7799 (i8 imm:$src3))))]>, VEX_4V;
7800 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7801 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7802 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7803 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7804 (i8 imm:$src3)))]>, VEX_4V;
7807 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7808 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7809 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7810 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7811 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7812 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7813 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7815 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7817 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7818 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7819 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7820 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7821 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7823 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7827 //===----------------------------------------------------------------------===//
7828 // VINSERTI128 - Insert packed integer values
7830 let neverHasSideEffects = 1 in {
7831 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7832 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7833 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7836 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7837 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7838 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7842 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7843 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7845 (VINSERTI128rr VR256:$src1, VR128:$src2,
7846 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7847 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7849 (VINSERTI128rr VR256:$src1, VR128:$src2,
7850 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7851 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7853 (VINSERTI128rr VR256:$src1, VR128:$src2,
7854 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7855 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7857 (VINSERTI128rr VR256:$src1, VR128:$src2,
7858 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7861 //===----------------------------------------------------------------------===//
7862 // VEXTRACTI128 - Extract packed integer values
7864 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7865 (ins VR256:$src1, i8imm:$src2),
7866 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7868 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7870 let neverHasSideEffects = 1, mayStore = 1 in
7871 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7872 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7873 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7875 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7876 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7877 (v2i64 (VEXTRACTI128rr
7878 (v4i64 VR256:$src1),
7879 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7880 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7881 (v4i32 (VEXTRACTI128rr
7882 (v8i32 VR256:$src1),
7883 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7884 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7885 (v8i16 (VEXTRACTI128rr
7886 (v16i16 VR256:$src1),
7887 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7888 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7889 (v16i8 (VEXTRACTI128rr
7890 (v32i8 VR256:$src1),
7891 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7894 //===----------------------------------------------------------------------===//
7895 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7897 multiclass avx2_pmovmask<string OpcodeStr,
7898 Intrinsic IntLd128, Intrinsic IntLd256,
7899 Intrinsic IntSt128, Intrinsic IntSt256> {
7900 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7901 (ins VR128:$src1, i128mem:$src2),
7902 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7903 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7904 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7905 (ins VR256:$src1, i256mem:$src2),
7906 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7907 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7908 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7909 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7910 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7911 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7912 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7913 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7914 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7915 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7918 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7919 int_x86_avx2_maskload_d,
7920 int_x86_avx2_maskload_d_256,
7921 int_x86_avx2_maskstore_d,
7922 int_x86_avx2_maskstore_d_256>;
7923 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7924 int_x86_avx2_maskload_q,
7925 int_x86_avx2_maskload_q_256,
7926 int_x86_avx2_maskstore_q,
7927 int_x86_avx2_maskstore_q_256>, VEX_W;
7930 //===----------------------------------------------------------------------===//
7931 // Variable Bit Shifts
7933 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7934 ValueType vt128, ValueType vt256> {
7935 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7936 (ins VR128:$src1, VR128:$src2),
7937 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7939 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7941 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7942 (ins VR128:$src1, i128mem:$src2),
7943 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7945 (vt128 (OpNode VR128:$src1,
7946 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7948 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7949 (ins VR256:$src1, VR256:$src2),
7950 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7952 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7954 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7955 (ins VR256:$src1, i256mem:$src2),
7956 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7958 (vt256 (OpNode VR256:$src1,
7959 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7963 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7964 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7965 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7966 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7967 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
7969 //===----------------------------------------------------------------------===//
7970 // VGATHER - GATHER Operations
7971 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
7972 X86MemOperand memop128, X86MemOperand memop256> {
7973 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
7974 (ins VR128:$src1, memop128:$src2, VR128:$mask),
7975 !strconcat(OpcodeStr,
7976 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7978 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
7979 (ins RC256:$src1, memop256:$src2, RC256:$mask),
7980 !strconcat(OpcodeStr,
7981 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
7982 []>, VEX_4VOp3, VEX_L;
7985 let Constraints = "$src1 = $dst, $mask = $mask_wb" in {
7986 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
7987 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
7988 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
7989 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
7990 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
7991 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
7992 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
7993 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;