1 //===-- X86InstrSSE.td - SSE Instruction Set ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
21 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
27 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
28 InstrItinClass arg_ri> {
29 InstrItinClass rr = arg_rr;
30 InstrItinClass rm = arg_rm;
31 InstrItinClass ri = arg_ri;
36 def SSE_ALU_F32S : OpndItins<
37 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
40 def SSE_ALU_F64S : OpndItins<
41 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
44 def SSE_ALU_ITINS_S : SizeItins<
45 SSE_ALU_F32S, SSE_ALU_F64S
48 def SSE_MUL_F32S : OpndItins<
49 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
52 def SSE_MUL_F64S : OpndItins<
53 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
56 def SSE_MUL_ITINS_S : SizeItins<
57 SSE_MUL_F32S, SSE_MUL_F64S
60 def SSE_DIV_F32S : OpndItins<
61 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
64 def SSE_DIV_F64S : OpndItins<
65 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
68 def SSE_DIV_ITINS_S : SizeItins<
69 SSE_DIV_F32S, SSE_DIV_F64S
73 def SSE_ALU_F32P : OpndItins<
74 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
77 def SSE_ALU_F64P : OpndItins<
78 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
81 def SSE_ALU_ITINS_P : SizeItins<
82 SSE_ALU_F32P, SSE_ALU_F64P
85 def SSE_MUL_F32P : OpndItins<
86 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
89 def SSE_MUL_F64P : OpndItins<
90 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
93 def SSE_MUL_ITINS_P : SizeItins<
94 SSE_MUL_F32P, SSE_MUL_F64P
97 def SSE_DIV_F32P : OpndItins<
98 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
101 def SSE_DIV_F64P : OpndItins<
102 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
105 def SSE_DIV_ITINS_P : SizeItins<
106 SSE_DIV_F32P, SSE_DIV_F64P
109 def SSE_BIT_ITINS_P : OpndItins<
110 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
113 def SSE_INTALU_ITINS_P : OpndItins<
114 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
117 def SSE_INTALUQ_ITINS_P : OpndItins<
118 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
121 def SSE_INTMUL_ITINS_P : OpndItins<
122 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
125 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
126 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
129 def SSE_MOVA_ITINS : OpndItins<
130 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
133 def SSE_MOVU_ITINS : OpndItins<
134 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
137 //===----------------------------------------------------------------------===//
138 // SSE 1 & 2 Instructions Classes
139 //===----------------------------------------------------------------------===//
141 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
142 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
143 RegisterClass RC, X86MemOperand x86memop,
146 let isCommutable = 1 in {
147 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
149 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
150 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
151 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
153 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
157 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
160 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
161 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
162 string asm, string SSEVer, string FPSizeStr,
163 Operand memopr, ComplexPattern mem_cpat,
166 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
168 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
169 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
170 [(set RC:$dst, (!cast<Intrinsic>(
171 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
172 RC:$src1, RC:$src2))], itins.rr>;
173 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
175 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
176 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
177 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
178 SSEVer, "_", OpcodeStr, FPSizeStr))
179 RC:$src1, mem_cpat:$src2))], itins.rm>;
182 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
183 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
184 RegisterClass RC, ValueType vt,
185 X86MemOperand x86memop, PatFrag mem_frag,
186 Domain d, OpndItins itins, bit Is2Addr = 1> {
187 let isCommutable = 1 in
188 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
190 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
191 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
192 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
194 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
196 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
197 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
198 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
202 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
203 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
204 string OpcodeStr, X86MemOperand x86memop,
205 list<dag> pat_rr, list<dag> pat_rm,
207 bit rr_hasSideEffects = 0> {
208 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
209 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
213 pat_rr, IIC_DEFAULT, d>;
214 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
218 pat_rm, IIC_DEFAULT, d>;
221 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
222 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
223 string asm, string SSEVer, string FPSizeStr,
224 X86MemOperand x86memop, PatFrag mem_frag,
225 Domain d, OpndItins itins, bit Is2Addr = 1> {
226 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
228 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
229 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
230 [(set RC:$dst, (!cast<Intrinsic>(
231 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
232 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
233 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
235 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
236 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
237 [(set RC:$dst, (!cast<Intrinsic>(
238 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
239 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
242 //===----------------------------------------------------------------------===//
243 // Non-instruction patterns
244 //===----------------------------------------------------------------------===//
246 // A vector extract of the first f32/f64 position is a subregister copy
247 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
248 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
249 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
250 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
252 // A 128-bit subvector extract from the first 256-bit vector position
253 // is a subregister copy that needs no instruction.
254 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
255 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
256 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
257 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
259 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
260 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
261 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
262 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
264 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
265 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
266 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
267 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
269 // A 128-bit subvector insert to the first 256-bit vector position
270 // is a subregister copy that needs no instruction.
271 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
272 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
273 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
274 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
275 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
276 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
277 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
278 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
279 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
280 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
281 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
282 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
284 // Implicitly promote a 32-bit scalar to a vector.
285 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
286 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
287 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
288 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
289 // Implicitly promote a 64-bit scalar to a vector.
290 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
291 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
292 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
293 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
295 // Bitcasts between 128-bit vector types. Return the original type since
296 // no instruction is needed for the conversion
297 let Predicates = [HasSSE2] in {
298 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
299 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
300 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
301 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
302 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
303 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
304 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
305 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
306 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
307 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
308 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
309 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
310 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
311 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
312 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
313 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
314 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
315 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
316 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
317 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
318 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
319 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
320 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
321 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
322 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
323 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
324 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
325 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
326 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
327 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
330 // Bitcasts between 256-bit vector types. Return the original type since
331 // no instruction is needed for the conversion
332 let Predicates = [HasAVX] in {
333 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
334 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
335 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
336 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
337 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
338 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
339 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
340 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
341 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
342 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
343 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
344 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
345 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
346 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
347 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
348 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
349 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
350 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
351 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
352 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
353 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
354 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
355 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
356 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
357 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
358 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
359 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
360 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
361 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
362 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
365 // Alias instructions that map fld0 to pxor for sse.
366 // This is expanded by ExpandPostRAPseudos.
367 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
369 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
370 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
371 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
372 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
375 //===----------------------------------------------------------------------===//
376 // AVX & SSE - Zero/One Vectors
377 //===----------------------------------------------------------------------===//
379 // Alias instruction that maps zero vector to pxor / xorp* for sse.
380 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
381 // swizzled by ExecutionDepsFix to pxor.
382 // We set canFoldAsLoad because this can be converted to a constant-pool
383 // load of an all-zeros value if folding it would be beneficial.
384 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
385 isPseudo = 1, neverHasSideEffects = 1 in {
386 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
389 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
390 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
391 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
392 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
393 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
394 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
397 // The same as done above but for AVX. The 256-bit ISA does not support PI,
398 // and doesn't need it because on sandy bridge the register is set to zero
399 // at the rename stage without using any execution unit, so SET0PSY
400 // and SET0PDY can be used for vector int instructions without penalty
401 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
402 // JIT implementatioan, it does not expand the instructions below like
403 // X86MCInstLower does.
404 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
405 isCodeGenOnly = 1 in {
406 let Predicates = [HasAVX] in {
407 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
408 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
409 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
410 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
412 let Predicates = [HasAVX2], neverHasSideEffects = 1 in
413 def AVX2_SET0 : PDI<0xef, MRMInitReg, (outs VR256:$dst), (ins), "",
417 let Predicates = [HasAVX2], AddedComplexity = 5 in {
418 def : Pat<(v4i64 immAllZerosV), (AVX2_SET0)>;
419 def : Pat<(v8i32 immAllZerosV), (AVX2_SET0)>;
420 def : Pat<(v16i16 immAllZerosV), (AVX2_SET0)>;
421 def : Pat<(v32i8 immAllZerosV), (AVX2_SET0)>;
424 // AVX has no support for 256-bit integer instructions, but since the 128-bit
425 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
426 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
427 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
428 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
430 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
431 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
432 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
434 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
435 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
436 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
438 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
439 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
440 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
442 // We set canFoldAsLoad because this can be converted to a constant-pool
443 // load of an all-ones value if folding it would be beneficial.
444 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
445 // JIT implementation, it does not expand the instructions below like
446 // X86MCInstLower does.
447 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
448 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
449 let Predicates = [HasAVX] in
450 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
451 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
452 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
453 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
454 let Predicates = [HasAVX2] in
455 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
456 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
460 //===----------------------------------------------------------------------===//
461 // SSE 1 & 2 - Move FP Scalar Instructions
463 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
464 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
465 // is used instead. Register-to-register movss/movsd is not modeled as an
466 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
467 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
468 //===----------------------------------------------------------------------===//
470 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
471 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
472 [(set VR128:$dst, (vt (OpNode VR128:$src1,
473 (scalar_to_vector RC:$src2))))],
476 // Loading from memory automatically zeroing upper bits.
477 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
478 PatFrag mem_pat, string OpcodeStr> :
479 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
481 [(set RC:$dst, (mem_pat addr:$src))],
485 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
486 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
488 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
489 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
492 // For the disassembler
493 let isCodeGenOnly = 1 in {
494 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
495 (ins VR128:$src1, FR32:$src2),
496 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
499 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
500 (ins VR128:$src1, FR64:$src2),
501 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
506 let canFoldAsLoad = 1, isReMaterializable = 1 in {
507 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
509 let AddedComplexity = 20 in
510 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
514 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
515 "movss\t{$src, $dst|$dst, $src}",
516 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
518 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
519 "movsd\t{$src, $dst|$dst, $src}",
520 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
524 let Constraints = "$src1 = $dst" in {
525 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
526 "movss\t{$src2, $dst|$dst, $src2}">, XS;
527 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
528 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
530 // For the disassembler
531 let isCodeGenOnly = 1 in {
532 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
533 (ins VR128:$src1, FR32:$src2),
534 "movss\t{$src2, $dst|$dst, $src2}", [],
535 IIC_SSE_MOV_S_RR>, XS;
536 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
537 (ins VR128:$src1, FR64:$src2),
538 "movsd\t{$src2, $dst|$dst, $src2}", [],
539 IIC_SSE_MOV_S_RR>, XD;
543 let canFoldAsLoad = 1, isReMaterializable = 1 in {
544 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
546 let AddedComplexity = 20 in
547 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
550 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
551 "movss\t{$src, $dst|$dst, $src}",
552 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
553 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
554 "movsd\t{$src, $dst|$dst, $src}",
555 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
558 let Predicates = [HasAVX] in {
559 let AddedComplexity = 15 in {
560 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
561 // MOVS{S,D} to the lower bits.
562 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
563 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
564 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
565 (VMOVSSrr (v4f32 (V_SET0)),
566 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
567 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
568 (VMOVSSrr (v4i32 (V_SET0)),
569 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
570 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
571 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
573 // Move low f32 and clear high bits.
574 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
575 (SUBREG_TO_REG (i32 0),
576 (VMOVSSrr (v4f32 (V_SET0)),
577 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
578 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
579 (SUBREG_TO_REG (i32 0),
580 (VMOVSSrr (v4i32 (V_SET0)),
581 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
584 let AddedComplexity = 20 in {
585 // MOVSSrm zeros the high parts of the register; represent this
586 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
587 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
588 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
589 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
590 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
591 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
592 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
594 // MOVSDrm zeros the high parts of the register; represent this
595 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
596 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
597 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
598 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
599 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
600 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
601 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
602 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
603 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
604 def : Pat<(v2f64 (X86vzload addr:$src)),
605 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
607 // Represent the same patterns above but in the form they appear for
609 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
610 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
611 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
612 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
613 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
614 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
615 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
616 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
617 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
619 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
620 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
621 (SUBREG_TO_REG (i32 0),
622 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
624 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
625 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
626 (SUBREG_TO_REG (i64 0),
627 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
629 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
630 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
631 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
633 // Move low f64 and clear high bits.
634 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
635 (SUBREG_TO_REG (i32 0),
636 (VMOVSDrr (v2f64 (V_SET0)),
637 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
639 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
640 (SUBREG_TO_REG (i32 0),
641 (VMOVSDrr (v2i64 (V_SET0)),
642 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
644 // Extract and store.
645 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
648 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
649 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
652 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
654 // Shuffle with VMOVSS
655 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
656 (VMOVSSrr (v4i32 VR128:$src1),
657 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
658 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
659 (VMOVSSrr (v4f32 VR128:$src1),
660 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
663 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
664 (SUBREG_TO_REG (i32 0),
665 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
666 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
667 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
668 (SUBREG_TO_REG (i32 0),
669 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
670 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
672 // Shuffle with VMOVSD
673 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
674 (VMOVSDrr (v2i64 VR128:$src1),
675 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
676 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
677 (VMOVSDrr (v2f64 VR128:$src1),
678 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
679 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
680 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
682 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
683 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
687 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
688 (SUBREG_TO_REG (i32 0),
689 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
690 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
691 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
692 (SUBREG_TO_REG (i32 0),
693 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
694 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
697 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
698 // is during lowering, where it's not possible to recognize the fold cause
699 // it has two uses through a bitcast. One use disappears at isel time and the
700 // fold opportunity reappears.
701 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
702 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
704 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
705 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
707 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
708 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
710 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
711 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
715 let Predicates = [HasSSE1] in {
716 let AddedComplexity = 15 in {
717 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
718 // MOVSS to the lower bits.
719 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
720 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
721 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
722 (MOVSSrr (v4f32 (V_SET0)),
723 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
724 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
725 (MOVSSrr (v4i32 (V_SET0)),
726 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
729 let AddedComplexity = 20 in {
730 // MOVSSrm zeros the high parts of the register; represent this
731 // with SUBREG_TO_REG.
732 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
733 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
734 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
735 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
736 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
737 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
740 // Extract and store.
741 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
744 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
746 // Shuffle with MOVSS
747 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
748 (MOVSSrr (v4i32 VR128:$src1),
749 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
750 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
751 (MOVSSrr (v4f32 VR128:$src1),
752 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
755 let Predicates = [HasSSE2] in {
756 let AddedComplexity = 15 in {
757 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
758 // MOVSD to the lower bits.
759 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
760 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
763 let AddedComplexity = 20 in {
764 // MOVSDrm zeros the high parts of the register; represent this
765 // with SUBREG_TO_REG.
766 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
767 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
768 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
769 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
770 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
771 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
772 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
773 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
774 def : Pat<(v2f64 (X86vzload addr:$src)),
775 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
778 // Extract and store.
779 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
782 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
784 // Shuffle with MOVSD
785 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
786 (MOVSDrr (v2i64 VR128:$src1),
787 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
788 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
789 (MOVSDrr (v2f64 VR128:$src1),
790 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
791 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
792 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
793 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
794 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
796 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
797 // is during lowering, where it's not possible to recognize the fold cause
798 // it has two uses through a bitcast. One use disappears at isel time and the
799 // fold opportunity reappears.
800 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
801 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
802 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
803 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
804 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
805 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
806 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
807 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
810 //===----------------------------------------------------------------------===//
811 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
812 //===----------------------------------------------------------------------===//
814 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
815 X86MemOperand x86memop, PatFrag ld_frag,
816 string asm, Domain d,
818 bit IsReMaterializable = 1> {
819 let neverHasSideEffects = 1 in
820 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
821 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
822 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
823 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
824 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
825 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
828 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
829 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
831 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
832 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
834 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
835 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
837 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
838 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
841 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
842 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
844 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
845 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
847 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
848 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
850 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
851 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
853 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
854 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
856 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
857 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
859 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
860 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
862 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
863 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
866 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
867 "movaps\t{$src, $dst|$dst, $src}",
868 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
869 IIC_SSE_MOVA_P_MR>, VEX;
870 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
871 "movapd\t{$src, $dst|$dst, $src}",
872 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
873 IIC_SSE_MOVA_P_MR>, VEX;
874 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
875 "movups\t{$src, $dst|$dst, $src}",
876 [(store (v4f32 VR128:$src), addr:$dst)],
877 IIC_SSE_MOVU_P_MR>, VEX;
878 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
879 "movupd\t{$src, $dst|$dst, $src}",
880 [(store (v2f64 VR128:$src), addr:$dst)],
881 IIC_SSE_MOVU_P_MR>, VEX;
882 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
883 "movaps\t{$src, $dst|$dst, $src}",
884 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
885 IIC_SSE_MOVA_P_MR>, VEX;
886 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
887 "movapd\t{$src, $dst|$dst, $src}",
888 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
889 IIC_SSE_MOVA_P_MR>, VEX;
890 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
891 "movups\t{$src, $dst|$dst, $src}",
892 [(store (v8f32 VR256:$src), addr:$dst)],
893 IIC_SSE_MOVU_P_MR>, VEX;
894 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
895 "movupd\t{$src, $dst|$dst, $src}",
896 [(store (v4f64 VR256:$src), addr:$dst)],
897 IIC_SSE_MOVU_P_MR>, VEX;
900 let isCodeGenOnly = 1 in {
901 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
903 "movaps\t{$src, $dst|$dst, $src}", [],
904 IIC_SSE_MOVA_P_RR>, VEX;
905 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
907 "movapd\t{$src, $dst|$dst, $src}", [],
908 IIC_SSE_MOVA_P_RR>, VEX;
909 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
911 "movups\t{$src, $dst|$dst, $src}", [],
912 IIC_SSE_MOVU_P_RR>, VEX;
913 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
915 "movupd\t{$src, $dst|$dst, $src}", [],
916 IIC_SSE_MOVU_P_RR>, VEX;
917 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
919 "movaps\t{$src, $dst|$dst, $src}", [],
920 IIC_SSE_MOVA_P_RR>, VEX;
921 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
923 "movapd\t{$src, $dst|$dst, $src}", [],
924 IIC_SSE_MOVA_P_RR>, VEX;
925 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
927 "movups\t{$src, $dst|$dst, $src}", [],
928 IIC_SSE_MOVU_P_RR>, VEX;
929 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
931 "movupd\t{$src, $dst|$dst, $src}", [],
932 IIC_SSE_MOVU_P_RR>, VEX;
935 let Predicates = [HasAVX] in {
936 def : Pat<(v8i32 (X86vzmovl
937 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
938 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
939 def : Pat<(v4i64 (X86vzmovl
940 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
941 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
942 def : Pat<(v8f32 (X86vzmovl
943 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
944 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
945 def : Pat<(v4f64 (X86vzmovl
946 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
947 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
951 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
952 (VMOVUPSYmr addr:$dst, VR256:$src)>;
953 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
954 (VMOVUPDYmr addr:$dst, VR256:$src)>;
956 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
957 "movaps\t{$src, $dst|$dst, $src}",
958 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
960 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
961 "movapd\t{$src, $dst|$dst, $src}",
962 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
964 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
965 "movups\t{$src, $dst|$dst, $src}",
966 [(store (v4f32 VR128:$src), addr:$dst)],
968 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
969 "movupd\t{$src, $dst|$dst, $src}",
970 [(store (v2f64 VR128:$src), addr:$dst)],
974 let isCodeGenOnly = 1 in {
975 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
976 "movaps\t{$src, $dst|$dst, $src}", [],
978 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
979 "movapd\t{$src, $dst|$dst, $src}", [],
981 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
982 "movups\t{$src, $dst|$dst, $src}", [],
984 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
985 "movupd\t{$src, $dst|$dst, $src}", [],
989 let Predicates = [HasAVX] in {
990 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
991 (VMOVUPSmr addr:$dst, VR128:$src)>;
992 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
993 (VMOVUPDmr addr:$dst, VR128:$src)>;
996 let Predicates = [HasSSE1] in
997 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
998 (MOVUPSmr addr:$dst, VR128:$src)>;
999 let Predicates = [HasSSE2] in
1000 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
1001 (MOVUPDmr addr:$dst, VR128:$src)>;
1003 // Use vmovaps/vmovups for AVX integer load/store.
1004 let Predicates = [HasAVX] in {
1005 // 128-bit load/store
1006 def : Pat<(alignedloadv2i64 addr:$src),
1007 (VMOVAPSrm addr:$src)>;
1008 def : Pat<(loadv2i64 addr:$src),
1009 (VMOVUPSrm addr:$src)>;
1011 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1012 (VMOVAPSmr addr:$dst, VR128:$src)>;
1013 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1014 (VMOVAPSmr addr:$dst, VR128:$src)>;
1015 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1016 (VMOVAPSmr addr:$dst, VR128:$src)>;
1017 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1018 (VMOVAPSmr addr:$dst, VR128:$src)>;
1019 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1020 (VMOVUPSmr addr:$dst, VR128:$src)>;
1021 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1022 (VMOVUPSmr addr:$dst, VR128:$src)>;
1023 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1024 (VMOVUPSmr addr:$dst, VR128:$src)>;
1025 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1026 (VMOVUPSmr addr:$dst, VR128:$src)>;
1028 // 256-bit load/store
1029 def : Pat<(alignedloadv4i64 addr:$src),
1030 (VMOVAPSYrm addr:$src)>;
1031 def : Pat<(loadv4i64 addr:$src),
1032 (VMOVUPSYrm addr:$src)>;
1033 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1034 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1035 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1036 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1037 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1038 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1039 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1040 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1041 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1042 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1043 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1044 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1045 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1046 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1047 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1048 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1051 // Use movaps / movups for SSE integer load / store (one byte shorter).
1052 // The instructions selected below are then converted to MOVDQA/MOVDQU
1053 // during the SSE domain pass.
1054 let Predicates = [HasSSE1] in {
1055 def : Pat<(alignedloadv2i64 addr:$src),
1056 (MOVAPSrm addr:$src)>;
1057 def : Pat<(loadv2i64 addr:$src),
1058 (MOVUPSrm addr:$src)>;
1060 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1061 (MOVAPSmr addr:$dst, VR128:$src)>;
1062 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1063 (MOVAPSmr addr:$dst, VR128:$src)>;
1064 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1065 (MOVAPSmr addr:$dst, VR128:$src)>;
1066 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1067 (MOVAPSmr addr:$dst, VR128:$src)>;
1068 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1069 (MOVUPSmr addr:$dst, VR128:$src)>;
1070 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1071 (MOVUPSmr addr:$dst, VR128:$src)>;
1072 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1073 (MOVUPSmr addr:$dst, VR128:$src)>;
1074 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1075 (MOVUPSmr addr:$dst, VR128:$src)>;
1078 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1079 // bits are disregarded. FIXME: Set encoding to pseudo!
1080 let neverHasSideEffects = 1 in {
1081 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1082 "movaps\t{$src, $dst|$dst, $src}", [],
1083 IIC_SSE_MOVA_P_RR>, VEX;
1084 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1085 "movapd\t{$src, $dst|$dst, $src}", [],
1086 IIC_SSE_MOVA_P_RR>, VEX;
1087 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1088 "movaps\t{$src, $dst|$dst, $src}", [],
1090 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1091 "movapd\t{$src, $dst|$dst, $src}", [],
1095 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1096 // bits are disregarded. FIXME: Set encoding to pseudo!
1097 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1098 let isCodeGenOnly = 1 in {
1099 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1100 "movaps\t{$src, $dst|$dst, $src}",
1101 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1102 IIC_SSE_MOVA_P_RM>, VEX;
1103 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1104 "movapd\t{$src, $dst|$dst, $src}",
1105 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1106 IIC_SSE_MOVA_P_RM>, VEX;
1108 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1109 "movaps\t{$src, $dst|$dst, $src}",
1110 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1112 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1113 "movapd\t{$src, $dst|$dst, $src}",
1114 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1118 //===----------------------------------------------------------------------===//
1119 // SSE 1 & 2 - Move Low packed FP Instructions
1120 //===----------------------------------------------------------------------===//
1122 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1123 SDNode psnode, SDNode pdnode, string base_opc,
1124 string asm_opr, InstrItinClass itin> {
1125 def PSrm : PI<opc, MRMSrcMem,
1126 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1127 !strconcat(base_opc, "s", asm_opr),
1130 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1131 itin, SSEPackedSingle>, TB;
1133 def PDrm : PI<opc, MRMSrcMem,
1134 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1135 !strconcat(base_opc, "d", asm_opr),
1136 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1137 (scalar_to_vector (loadf64 addr:$src2)))))],
1138 itin, SSEPackedDouble>, TB, OpSize;
1141 let AddedComplexity = 20 in {
1142 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1143 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1144 IIC_SSE_MOV_LH>, VEX_4V;
1146 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1147 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1148 "\t{$src2, $dst|$dst, $src2}",
1152 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1153 "movlps\t{$src, $dst|$dst, $src}",
1154 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1155 (iPTR 0))), addr:$dst)],
1156 IIC_SSE_MOV_LH>, VEX;
1157 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1158 "movlpd\t{$src, $dst|$dst, $src}",
1159 [(store (f64 (vector_extract (v2f64 VR128:$src),
1160 (iPTR 0))), addr:$dst)],
1161 IIC_SSE_MOV_LH>, VEX;
1162 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1163 "movlps\t{$src, $dst|$dst, $src}",
1164 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1165 (iPTR 0))), addr:$dst)],
1167 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1168 "movlpd\t{$src, $dst|$dst, $src}",
1169 [(store (f64 (vector_extract (v2f64 VR128:$src),
1170 (iPTR 0))), addr:$dst)],
1173 let Predicates = [HasAVX] in {
1174 // Shuffle with VMOVLPS
1175 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1176 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1177 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1178 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1180 // Shuffle with VMOVLPD
1181 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1182 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1183 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1184 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1187 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1189 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1190 def : Pat<(store (v4i32 (X86Movlps
1191 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1192 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1193 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1195 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1196 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1198 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1201 let Predicates = [HasSSE1] in {
1202 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1203 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1204 (iPTR 0))), addr:$src1),
1205 (MOVLPSmr addr:$src1, VR128:$src2)>;
1207 // Shuffle with MOVLPS
1208 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1209 (MOVLPSrm VR128:$src1, addr:$src2)>;
1210 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1211 (MOVLPSrm VR128:$src1, addr:$src2)>;
1212 def : Pat<(X86Movlps VR128:$src1,
1213 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1214 (MOVLPSrm VR128:$src1, addr:$src2)>;
1217 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1219 (MOVLPSmr addr:$src1, VR128:$src2)>;
1220 def : Pat<(store (v4i32 (X86Movlps
1221 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1223 (MOVLPSmr addr:$src1, VR128:$src2)>;
1226 let Predicates = [HasSSE2] in {
1227 // Shuffle with MOVLPD
1228 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1229 (MOVLPDrm VR128:$src1, addr:$src2)>;
1230 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1231 (MOVLPDrm VR128:$src1, addr:$src2)>;
1234 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1236 (MOVLPDmr addr:$src1, VR128:$src2)>;
1237 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1239 (MOVLPDmr addr:$src1, VR128:$src2)>;
1242 //===----------------------------------------------------------------------===//
1243 // SSE 1 & 2 - Move Hi packed FP Instructions
1244 //===----------------------------------------------------------------------===//
1246 let AddedComplexity = 20 in {
1247 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1248 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1249 IIC_SSE_MOV_LH>, VEX_4V;
1251 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1252 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1253 "\t{$src2, $dst|$dst, $src2}",
1257 // v2f64 extract element 1 is always custom lowered to unpack high to low
1258 // and extract element 0 so the non-store version isn't too horrible.
1259 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1260 "movhps\t{$src, $dst|$dst, $src}",
1261 [(store (f64 (vector_extract
1262 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1263 (bc_v2f64 (v4f32 VR128:$src))),
1264 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1265 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1266 "movhpd\t{$src, $dst|$dst, $src}",
1267 [(store (f64 (vector_extract
1268 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1269 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1270 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1271 "movhps\t{$src, $dst|$dst, $src}",
1272 [(store (f64 (vector_extract
1273 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1274 (bc_v2f64 (v4f32 VR128:$src))),
1275 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1276 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1277 "movhpd\t{$src, $dst|$dst, $src}",
1278 [(store (f64 (vector_extract
1279 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1280 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1282 let Predicates = [HasAVX] in {
1284 def : Pat<(X86Movlhps VR128:$src1,
1285 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1286 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1287 def : Pat<(X86Movlhps VR128:$src1,
1288 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1289 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1291 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1292 // is during lowering, where it's not possible to recognize the load fold
1293 // cause it has two uses through a bitcast. One use disappears at isel time
1294 // and the fold opportunity reappears.
1295 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1296 (scalar_to_vector (loadf64 addr:$src2)))),
1297 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1300 let Predicates = [HasSSE1] in {
1302 def : Pat<(X86Movlhps VR128:$src1,
1303 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1304 (MOVHPSrm VR128:$src1, addr:$src2)>;
1305 def : Pat<(X86Movlhps VR128:$src1,
1306 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1307 (MOVHPSrm VR128:$src1, addr:$src2)>;
1310 let Predicates = [HasSSE2] in {
1311 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1312 // is during lowering, where it's not possible to recognize the load fold
1313 // cause it has two uses through a bitcast. One use disappears at isel time
1314 // and the fold opportunity reappears.
1315 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1316 (scalar_to_vector (loadf64 addr:$src2)))),
1317 (MOVHPDrm VR128:$src1, addr:$src2)>;
1320 //===----------------------------------------------------------------------===//
1321 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1322 //===----------------------------------------------------------------------===//
1324 let AddedComplexity = 20 in {
1325 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1326 (ins VR128:$src1, VR128:$src2),
1327 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1329 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1332 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1333 (ins VR128:$src1, VR128:$src2),
1334 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1336 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1340 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1341 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1342 (ins VR128:$src1, VR128:$src2),
1343 "movlhps\t{$src2, $dst|$dst, $src2}",
1345 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1347 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1348 (ins VR128:$src1, VR128:$src2),
1349 "movhlps\t{$src2, $dst|$dst, $src2}",
1351 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1355 let Predicates = [HasAVX] in {
1357 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1358 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1359 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1360 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1363 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1364 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1367 let Predicates = [HasSSE1] in {
1369 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1370 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1371 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1372 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1375 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1376 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1379 //===----------------------------------------------------------------------===//
1380 // SSE 1 & 2 - Conversion Instructions
1381 //===----------------------------------------------------------------------===//
1383 def SSE_CVT_PD : OpndItins<
1384 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1387 def SSE_CVT_PS : OpndItins<
1388 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1391 def SSE_CVT_Scalar : OpndItins<
1392 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1395 def SSE_CVT_SS2SI_32 : OpndItins<
1396 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1399 def SSE_CVT_SS2SI_64 : OpndItins<
1400 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1403 def SSE_CVT_SD2SI : OpndItins<
1404 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1407 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1408 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1409 string asm, OpndItins itins> {
1410 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1411 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1413 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1414 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1418 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1419 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1420 string asm, Domain d, OpndItins itins> {
1421 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1422 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1424 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1425 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1429 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1430 X86MemOperand x86memop, string asm> {
1431 let neverHasSideEffects = 1 in {
1432 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1433 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1435 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1436 (ins DstRC:$src1, x86memop:$src),
1437 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1438 } // neverHasSideEffects = 1
1441 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1442 "cvttss2si\t{$src, $dst|$dst, $src}",
1445 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1446 "cvttss2si\t{$src, $dst|$dst, $src}",
1448 XS, VEX, VEX_W, VEX_LIG;
1449 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1450 "cvttsd2si\t{$src, $dst|$dst, $src}",
1453 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1454 "cvttsd2si\t{$src, $dst|$dst, $src}",
1456 XD, VEX, VEX_W, VEX_LIG;
1458 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1459 // register, but the same isn't true when only using memory operands,
1460 // provide other assembly "l" and "q" forms to address this explicitly
1461 // where appropriate to do so.
1462 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1463 XS, VEX_4V, VEX_LIG;
1464 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1465 XS, VEX_4V, VEX_W, VEX_LIG;
1466 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1467 XD, VEX_4V, VEX_LIG;
1468 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1469 XD, VEX_4V, VEX_LIG;
1470 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1471 XD, VEX_4V, VEX_W, VEX_LIG;
1473 let Predicates = [HasAVX], AddedComplexity = 1 in {
1474 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1475 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1476 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1477 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1478 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1479 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1480 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1481 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1483 def : Pat<(f32 (sint_to_fp GR32:$src)),
1484 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1485 def : Pat<(f32 (sint_to_fp GR64:$src)),
1486 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1487 def : Pat<(f64 (sint_to_fp GR32:$src)),
1488 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1489 def : Pat<(f64 (sint_to_fp GR64:$src)),
1490 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1493 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1494 "cvttss2si\t{$src, $dst|$dst, $src}",
1495 SSE_CVT_SS2SI_32>, XS;
1496 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1497 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1498 SSE_CVT_SS2SI_64>, XS, REX_W;
1499 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1500 "cvttsd2si\t{$src, $dst|$dst, $src}",
1502 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1503 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1504 SSE_CVT_SD2SI>, XD, REX_W;
1505 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1506 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1507 SSE_CVT_Scalar>, XS;
1508 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1509 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1510 SSE_CVT_Scalar>, XS, REX_W;
1511 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1512 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1513 SSE_CVT_Scalar>, XD;
1514 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1515 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1516 SSE_CVT_Scalar>, XD, REX_W;
1518 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1519 // and/or XMM operand(s).
1521 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1522 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1523 string asm, OpndItins itins> {
1524 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1525 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1526 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1527 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1528 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1529 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], itins.rm>;
1532 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1533 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1534 PatFrag ld_frag, string asm, OpndItins itins,
1536 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1538 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1539 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1540 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1542 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1543 (ins DstRC:$src1, x86memop:$src2),
1545 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1546 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1547 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1551 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1552 f128mem, load, "cvtsd2si", SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1553 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1554 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si",
1555 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1557 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1558 f128mem, load, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1559 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1560 f128mem, load, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1563 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1564 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1565 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1566 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1567 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss",
1568 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1570 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1571 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1572 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1573 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1574 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd",
1575 SSE_CVT_Scalar, 0>, XD,
1578 let Constraints = "$src1 = $dst" in {
1579 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1580 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1581 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1582 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1583 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1584 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1585 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1586 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1587 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1588 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1589 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1590 "cvtsi2sd", SSE_CVT_Scalar>, XD, REX_W;
1595 // Aliases for intrinsics
1596 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1597 f32mem, load, "cvttss2si",
1598 SSE_CVT_SS2SI_32>, XS, VEX;
1599 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1600 int_x86_sse_cvttss2si64, f32mem, load,
1601 "cvttss2si", SSE_CVT_SS2SI_64>,
1603 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1604 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1606 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1607 int_x86_sse2_cvttsd2si64, f128mem, load,
1608 "cvttsd2si", SSE_CVT_SD2SI>,
1610 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1611 f32mem, load, "cvttss2si",
1612 SSE_CVT_SS2SI_32>, XS;
1613 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1614 int_x86_sse_cvttss2si64, f32mem, load,
1615 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1617 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1618 f128mem, load, "cvttsd2si", SSE_CVT_SD2SI>,
1620 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1621 int_x86_sse2_cvttsd2si64, f128mem, load,
1622 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1625 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1626 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1627 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1628 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1629 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1630 "cvtss2si\t{$src, $dst|$dst, $src}",
1631 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1632 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1633 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1634 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1636 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1637 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1638 SSEPackedSingle, SSE_CVT_PS>, TB, VEX,
1642 let Pattern = []<dag>, neverHasSideEffects = 1 in {
1643 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1644 "cvtss2si{l}\t{$src, $dst|$dst, $src}",
1645 SSE_CVT_SS2SI_32>, XS;
1646 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1647 "cvtss2si{q}\t{$src, $dst|$dst, $src}",
1648 SSE_CVT_SS2SI_64>, XS, REX_W;
1649 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1650 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1651 SSEPackedSingle, SSE_CVT_PS>, TB,
1652 Requires<[HasSSE2]>;
1655 let Predicates = [HasAVX] in {
1656 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1657 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1658 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1659 (VCVTSS2SIrm addr:$src)>;
1660 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1661 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1662 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1663 (VCVTSS2SI64rm addr:$src)>;
1666 let Predicates = [HasSSE1] in {
1667 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1668 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1669 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1670 (CVTSS2SIrm addr:$src)>;
1671 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1672 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1673 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1674 (CVTSS2SI64rm addr:$src)>;
1679 // Convert scalar double to scalar single
1680 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1681 (ins FR64:$src1, FR64:$src2),
1682 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1683 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1685 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1686 (ins FR64:$src1, f64mem:$src2),
1687 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1688 [], IIC_SSE_CVT_Scalar_RM>,
1689 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1691 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1694 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1695 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1696 [(set FR32:$dst, (fround FR64:$src))],
1697 IIC_SSE_CVT_Scalar_RR>;
1698 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1699 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1700 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1701 IIC_SSE_CVT_Scalar_RM>,
1703 Requires<[HasSSE2, OptForSize]>;
1705 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1706 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1709 let Constraints = "$src1 = $dst" in
1710 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1711 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss",
1712 SSE_CVT_Scalar>, XS;
1714 // Convert scalar single to scalar double
1715 // SSE2 instructions with XS prefix
1716 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1717 (ins FR32:$src1, FR32:$src2),
1718 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1719 [], IIC_SSE_CVT_Scalar_RR>,
1720 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1722 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1723 (ins FR32:$src1, f32mem:$src2),
1724 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1725 [], IIC_SSE_CVT_Scalar_RM>,
1726 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1728 let Predicates = [HasAVX] in {
1729 def : Pat<(f64 (fextend FR32:$src)),
1730 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1731 def : Pat<(fextend (loadf32 addr:$src)),
1732 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1733 def : Pat<(extloadf32 addr:$src),
1734 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1737 def : Pat<(extloadf32 addr:$src),
1738 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1739 Requires<[HasAVX, OptForSpeed]>;
1741 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1742 "cvtss2sd\t{$src, $dst|$dst, $src}",
1743 [(set FR64:$dst, (fextend FR32:$src))],
1744 IIC_SSE_CVT_Scalar_RR>, XS,
1745 Requires<[HasSSE2]>;
1746 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1747 "cvtss2sd\t{$src, $dst|$dst, $src}",
1748 [(set FR64:$dst, (extloadf32 addr:$src))],
1749 IIC_SSE_CVT_Scalar_RM>, XS,
1750 Requires<[HasSSE2, OptForSize]>;
1752 // extload f32 -> f64. This matches load+fextend because we have a hack in
1753 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1755 // Since these loads aren't folded into the fextend, we have to match it
1757 def : Pat<(fextend (loadf32 addr:$src)),
1758 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1759 def : Pat<(extloadf32 addr:$src),
1760 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1762 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1763 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1764 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1767 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V,
1769 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1770 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1771 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1772 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1773 (load addr:$src2)))],
1774 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V,
1776 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1777 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1778 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1779 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1780 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1782 IIC_SSE_CVT_Scalar_RR>, XS,
1783 Requires<[HasSSE2]>;
1784 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1785 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1786 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1787 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1788 (load addr:$src2)))],
1789 IIC_SSE_CVT_Scalar_RM>, XS,
1790 Requires<[HasSSE2]>;
1793 // Convert packed single/double fp to doubleword
1794 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1796 IIC_SSE_CVT_PS_RR>, VEX;
1797 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1798 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1799 IIC_SSE_CVT_PS_RM>, VEX;
1800 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1801 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1802 IIC_SSE_CVT_PS_RR>, VEX;
1803 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1804 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1805 IIC_SSE_CVT_PS_RM>, VEX;
1806 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1807 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1809 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1810 "cvtps2dq\t{$src, $dst|$dst, $src}", [],
1813 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1814 "cvtps2dq\t{$src, $dst|$dst, $src}",
1815 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1818 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1820 "cvtps2dq\t{$src, $dst|$dst, $src}",
1821 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1822 (memop addr:$src)))],
1823 IIC_SSE_CVT_PS_RM>, VEX;
1824 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1825 "cvtps2dq\t{$src, $dst|$dst, $src}",
1826 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1828 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1829 "cvtps2dq\t{$src, $dst|$dst, $src}",
1830 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1831 (memop addr:$src)))],
1834 // Convert Packed Double FP to Packed DW Integers
1835 let Predicates = [HasAVX] in {
1836 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1837 // register, but the same isn't true when using memory operands instead.
1838 // Provide other assembly rr and rm forms to address this explicitly.
1839 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1840 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1841 def VCVTPD2DQXrYr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1842 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1845 def VCVTPD2DQXrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1847 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1848 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1851 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1852 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1854 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1857 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1858 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1860 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1861 "cvtpd2dq\t{$src, $dst|$dst, $src}", [],
1864 // SSE2 packed instructions with XD prefix
1865 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1866 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1867 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1869 XD, VEX, Requires<[HasAVX]>;
1870 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1871 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1872 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1873 (memop addr:$src)))],
1875 XD, VEX, Requires<[HasAVX]>;
1876 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1877 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1878 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1880 XD, Requires<[HasSSE2]>;
1881 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1882 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1883 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1884 (memop addr:$src)))],
1886 XD, Requires<[HasSSE2]>;
1889 // Convert with truncation packed single/double fp to doubleword
1890 // SSE2 packed instructions with XS prefix
1891 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1892 "cvttps2dq\t{$src, $dst|$dst, $src}",
1894 (int_x86_sse2_cvttps2dq VR128:$src))],
1895 IIC_SSE_CVT_PS_RR>, VEX;
1896 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1897 "cvttps2dq\t{$src, $dst|$dst, $src}",
1898 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1899 (memop addr:$src)))],
1900 IIC_SSE_CVT_PS_RM>, VEX;
1901 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1902 "cvttps2dq\t{$src, $dst|$dst, $src}",
1904 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1905 IIC_SSE_CVT_PS_RR>, VEX;
1906 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1907 "cvttps2dq\t{$src, $dst|$dst, $src}",
1908 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1909 (memopv8f32 addr:$src)))],
1910 IIC_SSE_CVT_PS_RM>, VEX;
1912 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1913 "cvttps2dq\t{$src, $dst|$dst, $src}",
1915 (int_x86_sse2_cvttps2dq VR128:$src))],
1917 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1918 "cvttps2dq\t{$src, $dst|$dst, $src}",
1920 (int_x86_sse2_cvttps2dq (memop addr:$src)))],
1923 let Predicates = [HasAVX] in {
1924 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1925 (VCVTDQ2PSrr VR128:$src)>;
1926 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1927 (VCVTDQ2PSrm addr:$src)>;
1929 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1930 (VCVTDQ2PSrr VR128:$src)>;
1931 def : Pat<(int_x86_sse2_cvtdq2ps (bitconvert (memopv2i64 addr:$src))),
1932 (VCVTDQ2PSrm addr:$src)>;
1934 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1935 (VCVTTPS2DQrr VR128:$src)>;
1936 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1937 (VCVTTPS2DQrm addr:$src)>;
1939 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1940 (VCVTDQ2PSYrr VR256:$src)>;
1941 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1942 (VCVTDQ2PSYrm addr:$src)>;
1944 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1945 (VCVTTPS2DQYrr VR256:$src)>;
1946 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1947 (VCVTTPS2DQYrm addr:$src)>;
1950 let Predicates = [HasSSE2] in {
1951 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1952 (CVTDQ2PSrr VR128:$src)>;
1953 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1954 (CVTDQ2PSrm addr:$src)>;
1956 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1957 (CVTDQ2PSrr VR128:$src)>;
1958 def : Pat<(int_x86_sse2_cvtdq2ps (bitconvert (memopv2i64 addr:$src))),
1959 (CVTDQ2PSrm addr:$src)>;
1961 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1962 (CVTTPS2DQrr VR128:$src)>;
1963 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1964 (CVTTPS2DQrm addr:$src)>;
1967 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1968 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1970 (int_x86_sse2_cvttpd2dq VR128:$src))],
1971 IIC_SSE_CVT_PD_RR>, VEX;
1972 let isCodeGenOnly = 1 in
1973 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1974 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1975 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1976 (memop addr:$src)))],
1977 IIC_SSE_CVT_PD_RM>, VEX;
1978 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1979 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1980 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1982 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1983 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1984 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1985 (memop addr:$src)))],
1988 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1989 // register, but the same isn't true when using memory operands instead.
1990 // Provide other assembly rr and rm forms to address this explicitly.
1991 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1992 "cvttpd2dq\t{$src, $dst|$dst, $src}", [],
1993 IIC_SSE_CVT_PD_RR>, VEX;
1996 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1997 "cvttpd2dqx\t{$src, $dst|$dst, $src}", [],
1998 IIC_SSE_CVT_PD_RR>, VEX;
1999 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2000 "cvttpd2dqx\t{$src, $dst|$dst, $src}", [],
2001 IIC_SSE_CVT_PD_RM>, VEX;
2004 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2005 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
2006 IIC_SSE_CVT_PD_RR>, VEX;
2007 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2008 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}", [],
2009 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2011 let Predicates = [HasAVX] in {
2012 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2013 (VCVTTPD2DQYrr VR256:$src)>;
2014 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2015 (VCVTTPD2DQYrm addr:$src)>;
2016 } // Predicates = [HasAVX]
2018 // Convert packed single to packed double
2019 let Predicates = [HasAVX] in {
2020 // SSE2 instructions without OpSize prefix
2021 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2022 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2023 IIC_SSE_CVT_PD_RR>, TB, VEX;
2024 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2025 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2026 IIC_SSE_CVT_PD_RM>, TB, VEX;
2027 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2028 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2029 IIC_SSE_CVT_PD_RR>, TB, VEX;
2030 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2031 "vcvtps2pd\t{$src, $dst|$dst, $src}", [],
2032 IIC_SSE_CVT_PD_RM>, TB, VEX;
2034 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2035 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2036 IIC_SSE_CVT_PD_RR>, TB;
2037 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2038 "cvtps2pd\t{$src, $dst|$dst, $src}", [],
2039 IIC_SSE_CVT_PD_RM>, TB;
2041 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2042 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2043 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2045 TB, VEX, Requires<[HasAVX]>;
2046 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2047 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2048 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
2049 (load addr:$src)))],
2051 TB, VEX, Requires<[HasAVX]>;
2052 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2053 "cvtps2pd\t{$src, $dst|$dst, $src}",
2054 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2056 TB, Requires<[HasSSE2]>;
2057 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2058 "cvtps2pd\t{$src, $dst|$dst, $src}",
2059 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
2060 (load addr:$src)))],
2062 TB, Requires<[HasSSE2]>;
2064 // Convert Packed DW Integers to Packed Double FP
2065 let Predicates = [HasAVX] in {
2066 def VCVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2067 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2068 def VCVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2069 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2070 def VCVTDQ2PDYrm : SSDI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2071 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2072 def VCVTDQ2PDYrr : SSDI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2073 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2076 def CVTDQ2PDrm : SSDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2077 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2079 def CVTDQ2PDrr : SSDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2080 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2083 // 128 bit register conversion intrinsics
2084 let Predicates = [HasAVX] in
2085 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2086 (VCVTDQ2PDrr VR128:$src)>;
2088 let Predicates = [HasSSE2] in
2089 def : Pat<(int_x86_sse2_cvtdq2pd VR128:$src),
2090 (CVTDQ2PDrr VR128:$src)>;
2092 // AVX 256-bit register conversion intrinsics
2093 let Predicates = [HasAVX] in {
2094 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
2095 (VCVTDQ2PDYrr VR128:$src)>;
2096 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
2097 (VCVTDQ2PDYrm addr:$src)>;
2099 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
2100 (VCVTPD2DQYrr VR256:$src)>;
2101 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
2102 (VCVTPD2DQYrm addr:$src)>;
2104 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2105 (VCVTDQ2PDYrr VR128:$src)>;
2106 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2107 (VCVTDQ2PDYrm addr:$src)>;
2108 } // Predicates = [HasAVX]
2110 // Convert packed double to packed single
2111 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2112 // register, but the same isn't true when using memory operands instead.
2113 // Provide other assembly rr and rm forms to address this explicitly.
2114 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2115 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2116 IIC_SSE_CVT_PD_RR>, VEX;
2117 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2118 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2119 IIC_SSE_CVT_PD_RR>, VEX;
2122 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2123 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2124 IIC_SSE_CVT_PD_RR>, VEX;
2125 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2126 "cvtpd2psx\t{$src, $dst|$dst, $src}", [],
2127 IIC_SSE_CVT_PD_RM>, VEX;
2130 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2131 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2132 IIC_SSE_CVT_PD_RR>, VEX;
2133 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2134 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}", [],
2135 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2136 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2137 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2139 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2140 "cvtpd2ps\t{$src, $dst|$dst, $src}", [],
2144 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2145 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2146 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2148 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
2150 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2151 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2152 (memop addr:$src)))],
2154 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2155 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2156 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2158 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2159 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2160 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2161 (memop addr:$src)))],
2164 // AVX 256-bit register conversion intrinsics
2165 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2166 // whenever possible to avoid declaring two versions of each one.
2167 let Predicates = [HasAVX] in {
2168 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2169 (VCVTDQ2PSYrr VR256:$src)>;
2170 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2171 (VCVTDQ2PSYrm addr:$src)>;
2173 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2174 (VCVTPD2PSYrr VR256:$src)>;
2175 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2176 (VCVTPD2PSYrm addr:$src)>;
2178 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2179 (VCVTPS2DQYrr VR256:$src)>;
2180 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2181 (VCVTPS2DQYrm addr:$src)>;
2183 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2184 (VCVTPS2PDYrr VR128:$src)>;
2185 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2186 (VCVTPS2PDYrm addr:$src)>;
2188 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2189 (VCVTTPD2DQYrr VR256:$src)>;
2190 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2191 (VCVTTPD2DQYrm addr:$src)>;
2193 // Match fround and fextend for 128/256-bit conversions
2194 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2195 (VCVTPD2PSYrr VR256:$src)>;
2196 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2197 (VCVTPD2PSYrm addr:$src)>;
2199 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2200 (VCVTPS2PDYrr VR128:$src)>;
2201 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2202 (VCVTPS2PDYrm addr:$src)>;
2205 //===----------------------------------------------------------------------===//
2206 // SSE 1 & 2 - Compare Instructions
2207 //===----------------------------------------------------------------------===//
2209 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2210 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2211 Operand CC, SDNode OpNode, ValueType VT,
2212 PatFrag ld_frag, string asm, string asm_alt,
2214 def rr : SIi8<0xC2, MRMSrcReg,
2215 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2216 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2218 def rm : SIi8<0xC2, MRMSrcMem,
2219 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2220 [(set RC:$dst, (OpNode (VT RC:$src1),
2221 (ld_frag addr:$src2), imm:$cc))],
2224 // Accept explicit immediate argument form instead of comparison code.
2225 let neverHasSideEffects = 1 in {
2226 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2227 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2228 IIC_SSE_ALU_F32S_RR>;
2230 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2231 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2232 IIC_SSE_ALU_F32S_RM>;
2236 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2237 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2238 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2240 XS, VEX_4V, VEX_LIG;
2241 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2242 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2243 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2244 SSE_ALU_F32S>, // same latency as 32 bit compare
2245 XD, VEX_4V, VEX_LIG;
2247 let Constraints = "$src1 = $dst" in {
2248 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2249 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2250 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2252 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2253 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2254 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2255 SSE_ALU_F32S>, // same latency as 32 bit compare
2259 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2260 Intrinsic Int, string asm, OpndItins itins> {
2261 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2262 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2263 [(set VR128:$dst, (Int VR128:$src1,
2264 VR128:$src, imm:$cc))],
2266 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2267 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2268 [(set VR128:$dst, (Int VR128:$src1,
2269 (load addr:$src), imm:$cc))],
2273 // Aliases to match intrinsics which expect XMM operand(s).
2274 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2275 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2278 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2279 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2280 SSE_ALU_F32S>, // same latency as f32
2282 let Constraints = "$src1 = $dst" in {
2283 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2284 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2286 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2287 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2288 SSE_ALU_F32S>, // same latency as f32
2293 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2294 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2295 ValueType vt, X86MemOperand x86memop,
2296 PatFrag ld_frag, string OpcodeStr, Domain d> {
2297 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2298 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2299 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2300 IIC_SSE_COMIS_RR, d>;
2301 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2302 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2303 [(set EFLAGS, (OpNode (vt RC:$src1),
2304 (ld_frag addr:$src2)))],
2305 IIC_SSE_COMIS_RM, d>;
2308 let Defs = [EFLAGS] in {
2309 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2310 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2311 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2312 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2314 let Pattern = []<dag> in {
2315 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2316 "comiss", SSEPackedSingle>, TB, VEX,
2318 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2319 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2323 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2324 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2325 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2326 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2328 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2329 load, "comiss", SSEPackedSingle>, TB, VEX;
2330 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2331 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2332 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2333 "ucomiss", SSEPackedSingle>, TB;
2334 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2335 "ucomisd", SSEPackedDouble>, TB, OpSize;
2337 let Pattern = []<dag> in {
2338 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2339 "comiss", SSEPackedSingle>, TB;
2340 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2341 "comisd", SSEPackedDouble>, TB, OpSize;
2344 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2345 load, "ucomiss", SSEPackedSingle>, TB;
2346 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2347 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2349 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2350 "comiss", SSEPackedSingle>, TB;
2351 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2352 "comisd", SSEPackedDouble>, TB, OpSize;
2353 } // Defs = [EFLAGS]
2355 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2356 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2357 Operand CC, Intrinsic Int, string asm,
2358 string asm_alt, Domain d> {
2359 def rri : PIi8<0xC2, MRMSrcReg,
2360 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2361 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2362 IIC_SSE_CMPP_RR, d>;
2363 def rmi : PIi8<0xC2, MRMSrcMem,
2364 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2365 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2366 IIC_SSE_CMPP_RM, d>;
2368 // Accept explicit immediate argument form instead of comparison code.
2369 let neverHasSideEffects = 1 in {
2370 def rri_alt : PIi8<0xC2, MRMSrcReg,
2371 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2372 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2373 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2374 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2375 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2379 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2380 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2381 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2382 SSEPackedSingle>, TB, VEX_4V;
2383 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2384 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2385 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2386 SSEPackedDouble>, TB, OpSize, VEX_4V;
2387 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2388 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2389 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2390 SSEPackedSingle>, TB, VEX_4V;
2391 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2392 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2393 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2394 SSEPackedDouble>, TB, OpSize, VEX_4V;
2395 let Constraints = "$src1 = $dst" in {
2396 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2397 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2398 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2399 SSEPackedSingle>, TB;
2400 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2401 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2402 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2403 SSEPackedDouble>, TB, OpSize;
2406 let Predicates = [HasAVX] in {
2407 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2408 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2409 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2410 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2411 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2412 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2413 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2414 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2416 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2417 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2418 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2419 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2420 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2421 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2422 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2423 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2426 let Predicates = [HasSSE1] in {
2427 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2428 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2429 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2430 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2433 let Predicates = [HasSSE2] in {
2434 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2435 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2436 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2437 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2440 //===----------------------------------------------------------------------===//
2441 // SSE 1 & 2 - Shuffle Instructions
2442 //===----------------------------------------------------------------------===//
2444 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2445 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2446 ValueType vt, string asm, PatFrag mem_frag,
2447 Domain d, bit IsConvertibleToThreeAddress = 0> {
2448 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2449 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2450 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2451 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2452 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2453 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2454 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2455 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2456 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2459 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2460 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2461 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2462 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2463 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2464 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2465 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2466 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2467 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2468 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2469 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2470 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2472 let Constraints = "$src1 = $dst" in {
2473 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2474 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2475 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2477 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2478 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2479 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2483 let Predicates = [HasAVX] in {
2484 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2485 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2486 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2487 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2488 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2490 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2491 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2492 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2493 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2494 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2497 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2498 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2499 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2500 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2501 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2503 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2504 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2505 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2506 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2507 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2510 let Predicates = [HasSSE1] in {
2511 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2512 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2513 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2514 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2515 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2518 let Predicates = [HasSSE2] in {
2519 // Generic SHUFPD patterns
2520 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2521 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2522 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2523 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2524 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2527 //===----------------------------------------------------------------------===//
2528 // SSE 1 & 2 - Unpack Instructions
2529 //===----------------------------------------------------------------------===//
2531 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2532 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2533 PatFrag mem_frag, RegisterClass RC,
2534 X86MemOperand x86memop, string asm,
2536 def rr : PI<opc, MRMSrcReg,
2537 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2539 (vt (OpNode RC:$src1, RC:$src2)))],
2541 def rm : PI<opc, MRMSrcMem,
2542 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2544 (vt (OpNode RC:$src1,
2545 (mem_frag addr:$src2))))],
2549 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2550 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2551 SSEPackedSingle>, TB, VEX_4V;
2552 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2553 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2554 SSEPackedDouble>, TB, OpSize, VEX_4V;
2555 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2556 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2557 SSEPackedSingle>, TB, VEX_4V;
2558 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2559 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2560 SSEPackedDouble>, TB, OpSize, VEX_4V;
2562 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2563 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2564 SSEPackedSingle>, TB, VEX_4V;
2565 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2566 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2567 SSEPackedDouble>, TB, OpSize, VEX_4V;
2568 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2569 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2570 SSEPackedSingle>, TB, VEX_4V;
2571 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2572 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2573 SSEPackedDouble>, TB, OpSize, VEX_4V;
2575 let Constraints = "$src1 = $dst" in {
2576 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2577 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2578 SSEPackedSingle>, TB;
2579 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2580 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2581 SSEPackedDouble>, TB, OpSize;
2582 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2583 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2584 SSEPackedSingle>, TB;
2585 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2586 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2587 SSEPackedDouble>, TB, OpSize;
2588 } // Constraints = "$src1 = $dst"
2590 let Predicates = [HasAVX], AddedComplexity = 1 in {
2591 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2592 // problem is during lowering, where it's not possible to recognize the load
2593 // fold cause it has two uses through a bitcast. One use disappears at isel
2594 // time and the fold opportunity reappears.
2595 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2596 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2599 let Predicates = [HasSSE2] in {
2600 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2601 // problem is during lowering, where it's not possible to recognize the load
2602 // fold cause it has two uses through a bitcast. One use disappears at isel
2603 // time and the fold opportunity reappears.
2604 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2605 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2608 //===----------------------------------------------------------------------===//
2609 // SSE 1 & 2 - Extract Floating-Point Sign mask
2610 //===----------------------------------------------------------------------===//
2612 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2613 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2615 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2616 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2617 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2618 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2619 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2620 IIC_SSE_MOVMSK, d>, REX_W;
2623 let Predicates = [HasAVX] in {
2624 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2625 "movmskps", SSEPackedSingle>, TB, VEX;
2626 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2627 "movmskpd", SSEPackedDouble>, TB,
2629 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2630 "movmskps", SSEPackedSingle>, TB, VEX;
2631 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2632 "movmskpd", SSEPackedDouble>, TB,
2635 def : Pat<(i32 (X86fgetsign FR32:$src)),
2636 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2638 def : Pat<(i64 (X86fgetsign FR32:$src)),
2639 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2641 def : Pat<(i32 (X86fgetsign FR64:$src)),
2642 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2644 def : Pat<(i64 (X86fgetsign FR64:$src)),
2645 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2649 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2650 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2651 SSEPackedSingle>, TB, VEX;
2652 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2653 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2654 SSEPackedDouble>, TB,
2656 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2657 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2658 SSEPackedSingle>, TB, VEX;
2659 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2660 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2661 SSEPackedDouble>, TB,
2665 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2666 SSEPackedSingle>, TB;
2667 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2668 SSEPackedDouble>, TB, OpSize;
2670 def : Pat<(i32 (X86fgetsign FR32:$src)),
2671 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2672 sub_ss))>, Requires<[HasSSE1]>;
2673 def : Pat<(i64 (X86fgetsign FR32:$src)),
2674 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2675 sub_ss))>, Requires<[HasSSE1]>;
2676 def : Pat<(i32 (X86fgetsign FR64:$src)),
2677 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2678 sub_sd))>, Requires<[HasSSE2]>;
2679 def : Pat<(i64 (X86fgetsign FR64:$src)),
2680 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2681 sub_sd))>, Requires<[HasSSE2]>;
2683 //===---------------------------------------------------------------------===//
2684 // SSE2 - Packed Integer Logical Instructions
2685 //===---------------------------------------------------------------------===//
2687 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2689 /// PDI_binop_rm - Simple SSE2 binary operator.
2690 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2691 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2692 X86MemOperand x86memop,
2694 bit IsCommutable = 0,
2696 let isCommutable = IsCommutable in
2697 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2698 (ins RC:$src1, RC:$src2),
2700 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2701 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2702 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2703 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2704 (ins RC:$src1, x86memop:$src2),
2706 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2707 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2708 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2709 (bitconvert (memop_frag addr:$src2)))))],
2712 } // ExeDomain = SSEPackedInt
2714 // These are ordered here for pattern ordering requirements with the fp versions
2716 let Predicates = [HasAVX] in {
2717 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2718 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2719 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2720 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2721 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2722 i128mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2723 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2724 i128mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2727 let Constraints = "$src1 = $dst" in {
2728 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2729 i128mem, SSE_BIT_ITINS_P, 1>;
2730 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2731 i128mem, SSE_BIT_ITINS_P, 1>;
2732 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2733 i128mem, SSE_BIT_ITINS_P, 1>;
2734 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2735 i128mem, SSE_BIT_ITINS_P, 0>;
2736 } // Constraints = "$src1 = $dst"
2738 let Predicates = [HasAVX2] in {
2739 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2740 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2741 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2742 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2743 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2744 i256mem, SSE_BIT_ITINS_P, 1, 0>, VEX_4V;
2745 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2746 i256mem, SSE_BIT_ITINS_P, 0, 0>, VEX_4V;
2749 //===----------------------------------------------------------------------===//
2750 // SSE 1 & 2 - Logical Instructions
2751 //===----------------------------------------------------------------------===//
2753 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2755 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2756 SDNode OpNode, OpndItins itins> {
2757 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2758 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2761 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2762 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2765 let Constraints = "$src1 = $dst" in {
2766 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2767 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2770 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2771 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2776 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2777 let mayLoad = 0 in {
2778 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2780 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2782 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2786 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2787 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2790 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2792 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2794 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2795 // are all promoted to v2i64, and the patterns are covered by the int
2796 // version. This is needed in SSE only, because v2i64 isn't supported on
2797 // SSE1, but only on SSE2.
2798 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2799 !strconcat(OpcodeStr, "ps"), f128mem, [],
2800 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2801 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2803 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2804 !strconcat(OpcodeStr, "pd"), f128mem,
2805 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2806 (bc_v2i64 (v2f64 VR128:$src2))))],
2807 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2808 (memopv2i64 addr:$src2)))], 0>,
2810 let Constraints = "$src1 = $dst" in {
2811 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2812 !strconcat(OpcodeStr, "ps"), f128mem,
2813 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2814 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2815 (memopv2i64 addr:$src2)))]>, TB;
2817 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2818 !strconcat(OpcodeStr, "pd"), f128mem,
2819 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2820 (bc_v2i64 (v2f64 VR128:$src2))))],
2821 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2822 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2826 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2828 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2830 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2831 !strconcat(OpcodeStr, "ps"), f256mem,
2832 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2833 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2834 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2836 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2837 !strconcat(OpcodeStr, "pd"), f256mem,
2838 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2839 (bc_v4i64 (v4f64 VR256:$src2))))],
2840 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2841 (memopv4i64 addr:$src2)))], 0>,
2845 // AVX 256-bit packed logical ops forms
2846 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2847 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2848 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2849 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2851 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2852 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2853 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2854 let isCommutable = 0 in
2855 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2857 //===----------------------------------------------------------------------===//
2858 // SSE 1 & 2 - Arithmetic Instructions
2859 //===----------------------------------------------------------------------===//
2861 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2864 /// In addition, we also have a special variant of the scalar form here to
2865 /// represent the associated intrinsic operation. This form is unlike the
2866 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2867 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2869 /// These three forms can each be reg+reg or reg+mem.
2872 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2874 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2877 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2878 OpNode, FR32, f32mem,
2879 itins.s, Is2Addr>, XS;
2880 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2881 OpNode, FR64, f64mem,
2882 itins.d, Is2Addr>, XD;
2885 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2888 let mayLoad = 0 in {
2889 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2890 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2892 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2893 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2898 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2901 let mayLoad = 0 in {
2902 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2903 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2905 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2906 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2911 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2914 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2915 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2916 itins.s, Is2Addr>, XS;
2917 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2918 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2919 itins.d, Is2Addr>, XD;
2922 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2925 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2926 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2927 SSEPackedSingle, itins.s, Is2Addr>,
2930 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2931 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2932 SSEPackedDouble, itins.d, Is2Addr>,
2936 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2938 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2939 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2940 SSEPackedSingle, itins.s, 0>, TB;
2942 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2943 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2944 SSEPackedDouble, itins.d, 0>, TB, OpSize;
2947 // Binary Arithmetic instructions
2948 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2949 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2951 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2952 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2954 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2955 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2957 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2958 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2961 let isCommutable = 0 in {
2962 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2963 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2965 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2966 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>, VEX_4V;
2967 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2968 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2970 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2971 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2973 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2974 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2976 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2977 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2978 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2979 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2981 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2982 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2984 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2985 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2986 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2987 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2991 let Constraints = "$src1 = $dst" in {
2992 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2993 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2994 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2995 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2996 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2997 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2999 let isCommutable = 0 in {
3000 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
3001 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
3002 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
3003 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
3004 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
3005 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
3006 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
3007 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
3008 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
3009 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
3010 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
3011 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
3012 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
3013 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
3018 /// In addition, we also have a special variant of the scalar form here to
3019 /// represent the associated intrinsic operation. This form is unlike the
3020 /// plain scalar form, in that it takes an entire vector (instead of a
3021 /// scalar) and leaves the top elements undefined.
3023 /// And, we have a special variant form for a full-vector intrinsic form.
3025 def SSE_SQRTP : OpndItins<
3026 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
3029 def SSE_SQRTS : OpndItins<
3030 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
3033 def SSE_RCPP : OpndItins<
3034 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3037 def SSE_RCPS : OpndItins<
3038 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3041 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3042 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3043 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3044 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3045 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3046 [(set FR32:$dst, (OpNode FR32:$src))]>;
3047 // For scalar unary operations, fold a load into the operation
3048 // only in OptForSize mode. It eliminates an instruction, but it also
3049 // eliminates a whole-register clobber (the load), so it introduces a
3050 // partial register update condition.
3051 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3052 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3053 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3054 Requires<[HasSSE1, OptForSize]>;
3055 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3056 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3057 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
3058 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3059 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3060 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
3063 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3064 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3065 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3066 !strconcat(OpcodeStr,
3067 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3069 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3070 !strconcat(OpcodeStr,
3071 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3072 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3073 (ins VR128:$src1, ssmem:$src2),
3074 !strconcat(OpcodeStr,
3075 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3078 /// sse1_fp_unop_p - SSE1 unops in packed form.
3079 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3081 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3082 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3083 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3084 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3085 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3086 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3089 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3090 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3092 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3093 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3094 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3096 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3097 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3098 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3102 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3103 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3104 Intrinsic V4F32Int, OpndItins itins> {
3105 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3106 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3107 [(set VR128:$dst, (V4F32Int VR128:$src))],
3109 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3110 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3111 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3115 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3116 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3117 Intrinsic V4F32Int, OpndItins itins> {
3118 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3119 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3120 [(set VR256:$dst, (V4F32Int VR256:$src))],
3122 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3123 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3124 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3128 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3129 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3130 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3131 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3132 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3133 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3134 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3135 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3136 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3137 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3138 Requires<[HasSSE2, OptForSize]>;
3139 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3140 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3141 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3142 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3143 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3144 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3147 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3148 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3149 let neverHasSideEffects = 1 in {
3150 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3151 !strconcat(OpcodeStr,
3152 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3154 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3155 !strconcat(OpcodeStr,
3156 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3158 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3159 (ins VR128:$src1, sdmem:$src2),
3160 !strconcat(OpcodeStr,
3161 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3164 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3165 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3166 SDNode OpNode, OpndItins itins> {
3167 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3168 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3169 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3170 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3171 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3172 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3175 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3176 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3178 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3179 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3180 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3182 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3183 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3184 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3188 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3189 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3190 Intrinsic V2F64Int, OpndItins itins> {
3191 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3192 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3193 [(set VR128:$dst, (V2F64Int VR128:$src))],
3195 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3196 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3197 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3201 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3202 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3203 Intrinsic V2F64Int, OpndItins itins> {
3204 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3205 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3206 [(set VR256:$dst, (V2F64Int VR256:$src))],
3208 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3209 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3210 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3214 let Predicates = [HasAVX] in {
3216 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3217 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3219 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3220 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3221 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3222 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3223 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3225 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3227 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3229 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3233 // Reciprocal approximations. Note that these typically require refinement
3234 // in order to obtain suitable precision.
3235 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3236 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3237 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3238 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3240 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3243 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3244 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3245 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3246 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3248 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3252 let AddedComplexity = 1 in {
3253 def : Pat<(f32 (fsqrt FR32:$src)),
3254 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3255 def : Pat<(f32 (fsqrt (load addr:$src))),
3256 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3257 Requires<[HasAVX, OptForSize]>;
3258 def : Pat<(f64 (fsqrt FR64:$src)),
3259 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3260 def : Pat<(f64 (fsqrt (load addr:$src))),
3261 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3262 Requires<[HasAVX, OptForSize]>;
3264 def : Pat<(f32 (X86frsqrt FR32:$src)),
3265 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3266 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3267 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3268 Requires<[HasAVX, OptForSize]>;
3270 def : Pat<(f32 (X86frcp FR32:$src)),
3271 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3272 def : Pat<(f32 (X86frcp (load addr:$src))),
3273 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3274 Requires<[HasAVX, OptForSize]>;
3277 let Predicates = [HasAVX], AddedComplexity = 1 in {
3278 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3279 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3280 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3281 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3283 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3284 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3286 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3287 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3288 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3289 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3291 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3292 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3294 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3295 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3296 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3297 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3299 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3300 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3302 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3303 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3304 (VRCPSSr (f32 (IMPLICIT_DEF)),
3305 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3307 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3308 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3312 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3314 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3315 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3316 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3318 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3319 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3321 // Reciprocal approximations. Note that these typically require refinement
3322 // in order to obtain suitable precision.
3323 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3325 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3326 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3328 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3330 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3331 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3333 // There is no f64 version of the reciprocal approximation instructions.
3335 //===----------------------------------------------------------------------===//
3336 // SSE 1 & 2 - Non-temporal stores
3337 //===----------------------------------------------------------------------===//
3339 let AddedComplexity = 400 in { // Prefer non-temporal versions
3340 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3341 (ins f128mem:$dst, VR128:$src),
3342 "movntps\t{$src, $dst|$dst, $src}",
3343 [(alignednontemporalstore (v4f32 VR128:$src),
3345 IIC_SSE_MOVNT>, VEX;
3346 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3347 (ins f128mem:$dst, VR128:$src),
3348 "movntpd\t{$src, $dst|$dst, $src}",
3349 [(alignednontemporalstore (v2f64 VR128:$src),
3351 IIC_SSE_MOVNT>, VEX;
3353 let ExeDomain = SSEPackedInt in
3354 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3355 (ins f128mem:$dst, VR128:$src),
3356 "movntdq\t{$src, $dst|$dst, $src}",
3357 [(alignednontemporalstore (v2i64 VR128:$src),
3359 IIC_SSE_MOVNT>, VEX;
3361 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3362 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3364 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3365 (ins f256mem:$dst, VR256:$src),
3366 "movntps\t{$src, $dst|$dst, $src}",
3367 [(alignednontemporalstore (v8f32 VR256:$src),
3369 IIC_SSE_MOVNT>, VEX;
3370 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3371 (ins f256mem:$dst, VR256:$src),
3372 "movntpd\t{$src, $dst|$dst, $src}",
3373 [(alignednontemporalstore (v4f64 VR256:$src),
3375 IIC_SSE_MOVNT>, VEX;
3376 let ExeDomain = SSEPackedInt in
3377 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3378 (ins f256mem:$dst, VR256:$src),
3379 "movntdq\t{$src, $dst|$dst, $src}",
3380 [(alignednontemporalstore (v4i64 VR256:$src),
3382 IIC_SSE_MOVNT>, VEX;
3385 let AddedComplexity = 400 in { // Prefer non-temporal versions
3386 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3387 "movntps\t{$src, $dst|$dst, $src}",
3388 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3390 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3391 "movntpd\t{$src, $dst|$dst, $src}",
3392 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3395 let ExeDomain = SSEPackedInt in
3396 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3397 "movntdq\t{$src, $dst|$dst, $src}",
3398 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3401 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3402 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3404 // There is no AVX form for instructions below this point
3405 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3406 "movnti{l}\t{$src, $dst|$dst, $src}",
3407 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3409 TB, Requires<[HasSSE2]>;
3410 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3411 "movnti{q}\t{$src, $dst|$dst, $src}",
3412 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3414 TB, Requires<[HasSSE2]>;
3417 //===----------------------------------------------------------------------===//
3418 // SSE 1 & 2 - Prefetch and memory fence
3419 //===----------------------------------------------------------------------===//
3421 // Prefetch intrinsic.
3422 let Predicates = [HasSSE1] in {
3423 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3424 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3425 IIC_SSE_PREFETCH>, TB;
3426 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3427 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3428 IIC_SSE_PREFETCH>, TB;
3429 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3430 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3431 IIC_SSE_PREFETCH>, TB;
3432 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3433 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3434 IIC_SSE_PREFETCH>, TB;
3438 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3439 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3440 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3442 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3443 // was introduced with SSE2, it's backward compatible.
3444 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3446 // Load, store, and memory fence
3447 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3448 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3449 TB, Requires<[HasSSE1]>;
3450 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3451 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3452 TB, Requires<[HasSSE2]>;
3453 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3454 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3455 TB, Requires<[HasSSE2]>;
3457 def : Pat<(X86SFence), (SFENCE)>;
3458 def : Pat<(X86LFence), (LFENCE)>;
3459 def : Pat<(X86MFence), (MFENCE)>;
3461 //===----------------------------------------------------------------------===//
3462 // SSE 1 & 2 - Load/Store XCSR register
3463 //===----------------------------------------------------------------------===//
3465 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3466 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3467 IIC_SSE_LDMXCSR>, VEX;
3468 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3469 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3470 IIC_SSE_STMXCSR>, VEX;
3472 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3473 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3475 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3476 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3479 //===---------------------------------------------------------------------===//
3480 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3481 //===---------------------------------------------------------------------===//
3483 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3485 let neverHasSideEffects = 1 in {
3486 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3487 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3489 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3490 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3493 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3494 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3496 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3497 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3501 let isCodeGenOnly = 1 in {
3502 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3503 "movdqa\t{$src, $dst|$dst, $src}", [],
3506 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3507 "movdqa\t{$src, $dst|$dst, $src}", [],
3510 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3511 "movdqu\t{$src, $dst|$dst, $src}", [],
3514 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3515 "movdqu\t{$src, $dst|$dst, $src}", [],
3520 let canFoldAsLoad = 1, mayLoad = 1 in {
3521 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3522 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3524 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3525 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3527 let Predicates = [HasAVX] in {
3528 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3529 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3531 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3532 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3537 let mayStore = 1 in {
3538 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3539 (ins i128mem:$dst, VR128:$src),
3540 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3542 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3543 (ins i256mem:$dst, VR256:$src),
3544 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3546 let Predicates = [HasAVX] in {
3547 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3548 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3550 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3551 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3556 let neverHasSideEffects = 1 in
3557 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3558 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3560 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3561 "movdqu\t{$src, $dst|$dst, $src}",
3562 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3565 let isCodeGenOnly = 1 in {
3566 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3567 "movdqa\t{$src, $dst|$dst, $src}", [],
3570 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3571 "movdqu\t{$src, $dst|$dst, $src}",
3572 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[HasSSE2]>;
3575 let canFoldAsLoad = 1, mayLoad = 1 in {
3576 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3577 "movdqa\t{$src, $dst|$dst, $src}",
3578 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3580 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3581 "movdqu\t{$src, $dst|$dst, $src}",
3582 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3584 XS, Requires<[HasSSE2]>;
3587 let mayStore = 1 in {
3588 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3589 "movdqa\t{$src, $dst|$dst, $src}",
3590 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3592 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3593 "movdqu\t{$src, $dst|$dst, $src}",
3594 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3596 XS, Requires<[HasSSE2]>;
3599 // Intrinsic forms of MOVDQU load and store
3600 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3601 "vmovdqu\t{$src, $dst|$dst, $src}",
3602 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3604 XS, VEX, Requires<[HasAVX]>;
3606 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3607 "movdqu\t{$src, $dst|$dst, $src}",
3608 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)],
3610 XS, Requires<[HasSSE2]>;
3612 } // ExeDomain = SSEPackedInt
3614 let Predicates = [HasAVX] in {
3615 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3616 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3619 //===---------------------------------------------------------------------===//
3620 // SSE2 - Packed Integer Arithmetic Instructions
3621 //===---------------------------------------------------------------------===//
3623 def SSE_PMADD : OpndItins<
3624 IIC_SSE_PMADD, IIC_SSE_PMADD
3627 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3629 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3630 RegisterClass RC, PatFrag memop_frag,
3631 X86MemOperand x86memop,
3633 bit IsCommutable = 0,
3635 let isCommutable = IsCommutable in
3636 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3637 (ins RC:$src1, RC:$src2),
3639 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3640 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3641 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3642 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3643 (ins RC:$src1, x86memop:$src2),
3645 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3646 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3647 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3651 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3652 string OpcodeStr, SDNode OpNode,
3653 SDNode OpNode2, RegisterClass RC,
3654 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3655 ShiftOpndItins itins,
3657 // src2 is always 128-bit
3658 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3659 (ins RC:$src1, VR128:$src2),
3661 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3662 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3663 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3665 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3666 (ins RC:$src1, i128mem:$src2),
3668 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3669 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3670 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3671 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3672 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3673 (ins RC:$src1, i32i8imm:$src2),
3675 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3676 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3677 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3680 /// PDI_binop_rm - Simple SSE2 binary operator with different src and dst types
3681 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3682 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3683 PatFrag memop_frag, X86MemOperand x86memop,
3685 bit IsCommutable = 0, bit Is2Addr = 1> {
3686 let isCommutable = IsCommutable in
3687 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3688 (ins RC:$src1, RC:$src2),
3690 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3692 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3693 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3694 (ins RC:$src1, x86memop:$src2),
3696 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3697 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3698 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3699 (bitconvert (memop_frag addr:$src2)))))]>;
3701 } // ExeDomain = SSEPackedInt
3703 // 128-bit Integer Arithmetic
3705 let Predicates = [HasAVX] in {
3706 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3707 i128mem, SSE_INTALU_ITINS_P, 1, 0 /*3addr*/>,
3709 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3710 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3711 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3712 i128mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3713 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3714 i128mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3715 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3716 i128mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3717 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3718 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3719 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3720 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3721 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3722 i128mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3723 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3724 i128mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3725 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3726 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3730 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3731 VR128, memopv2i64, i128mem,
3732 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3733 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3734 VR128, memopv2i64, i128mem,
3735 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3736 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3737 VR128, memopv2i64, i128mem,
3738 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3739 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3740 VR128, memopv2i64, i128mem,
3741 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3742 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3743 VR128, memopv2i64, i128mem,
3744 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3745 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3746 VR128, memopv2i64, i128mem,
3747 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3748 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3749 VR128, memopv2i64, i128mem,
3750 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3751 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3752 VR128, memopv2i64, i128mem,
3753 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3754 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3755 VR128, memopv2i64, i128mem,
3756 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3757 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3758 VR128, memopv2i64, i128mem,
3759 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3760 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3761 VR128, memopv2i64, i128mem,
3762 SSE_PMADD, 1, 0>, VEX_4V;
3763 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3764 VR128, memopv2i64, i128mem,
3765 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3766 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3767 VR128, memopv2i64, i128mem,
3768 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3769 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3770 VR128, memopv2i64, i128mem,
3771 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3772 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3773 VR128, memopv2i64, i128mem,
3774 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3775 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3776 VR128, memopv2i64, i128mem,
3777 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3778 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3779 VR128, memopv2i64, i128mem,
3780 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3781 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3782 VR128, memopv2i64, i128mem,
3783 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3786 let Predicates = [HasAVX2] in {
3787 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3788 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3789 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3790 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3791 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3792 i256mem, SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3793 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3794 i256mem, SSE_INTALUQ_ITINS_P, 1, 0>, VEX_4V;
3795 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3796 i256mem, SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3797 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3798 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3799 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3800 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3801 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3802 i256mem, SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3803 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3804 i256mem, SSE_INTALUQ_ITINS_P, 0, 0>, VEX_4V;
3805 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3806 VR256, memopv4i64, i256mem,
3807 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3810 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3811 VR256, memopv4i64, i256mem,
3812 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3813 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3814 VR256, memopv4i64, i256mem,
3815 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3816 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3817 VR256, memopv4i64, i256mem,
3818 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3819 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3820 VR256, memopv4i64, i256mem,
3821 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3822 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3823 VR256, memopv4i64, i256mem,
3824 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3825 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3826 VR256, memopv4i64, i256mem,
3827 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3828 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3829 VR256, memopv4i64, i256mem,
3830 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3831 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3832 VR256, memopv4i64, i256mem,
3833 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3834 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3835 VR256, memopv4i64, i256mem,
3836 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3837 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3838 VR256, memopv4i64, i256mem,
3839 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V;
3840 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3841 VR256, memopv4i64, i256mem,
3842 SSE_PMADD, 1, 0>, VEX_4V;
3843 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3844 VR256, memopv4i64, i256mem,
3845 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3846 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3847 VR256, memopv4i64, i256mem,
3848 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3849 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3850 VR256, memopv4i64, i256mem,
3851 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3852 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3853 VR256, memopv4i64, i256mem,
3854 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3855 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3856 VR256, memopv4i64, i256mem,
3857 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3858 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3859 VR256, memopv4i64, i256mem,
3860 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3861 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3862 VR256, memopv4i64, i256mem,
3863 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
3866 let Constraints = "$src1 = $dst" in {
3867 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3868 i128mem, SSE_INTALU_ITINS_P, 1>;
3869 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3870 i128mem, SSE_INTALU_ITINS_P, 1>;
3871 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3872 i128mem, SSE_INTALU_ITINS_P, 1>;
3873 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3874 i128mem, SSE_INTALUQ_ITINS_P, 1>;
3875 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3876 i128mem, SSE_INTMUL_ITINS_P, 1>;
3877 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3878 i128mem, SSE_INTALU_ITINS_P>;
3879 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3880 i128mem, SSE_INTALU_ITINS_P>;
3881 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3882 i128mem, SSE_INTALU_ITINS_P>;
3883 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3884 i128mem, SSE_INTALUQ_ITINS_P>;
3885 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3886 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3889 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3890 VR128, memopv2i64, i128mem,
3891 SSE_INTALU_ITINS_P>;
3892 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3893 VR128, memopv2i64, i128mem,
3894 SSE_INTALU_ITINS_P>;
3895 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3896 VR128, memopv2i64, i128mem,
3897 SSE_INTALU_ITINS_P>;
3898 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3899 VR128, memopv2i64, i128mem,
3900 SSE_INTALU_ITINS_P>;
3901 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3902 VR128, memopv2i64, i128mem,
3903 SSE_INTALU_ITINS_P, 1>;
3904 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3905 VR128, memopv2i64, i128mem,
3906 SSE_INTALU_ITINS_P, 1>;
3907 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3908 VR128, memopv2i64, i128mem,
3909 SSE_INTALU_ITINS_P, 1>;
3910 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3911 VR128, memopv2i64, i128mem,
3912 SSE_INTALU_ITINS_P, 1>;
3913 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3914 VR128, memopv2i64, i128mem,
3915 SSE_INTMUL_ITINS_P, 1>;
3916 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3917 VR128, memopv2i64, i128mem,
3918 SSE_INTMUL_ITINS_P, 1>;
3919 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3920 VR128, memopv2i64, i128mem,
3922 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3923 VR128, memopv2i64, i128mem,
3924 SSE_INTALU_ITINS_P, 1>;
3925 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3926 VR128, memopv2i64, i128mem,
3927 SSE_INTALU_ITINS_P, 1>;
3928 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3929 VR128, memopv2i64, i128mem,
3930 SSE_INTALU_ITINS_P, 1>;
3931 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3932 VR128, memopv2i64, i128mem,
3933 SSE_INTALU_ITINS_P, 1>;
3934 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3935 VR128, memopv2i64, i128mem,
3936 SSE_INTALU_ITINS_P, 1>;
3937 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3938 VR128, memopv2i64, i128mem,
3939 SSE_INTALU_ITINS_P, 1>;
3940 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3941 VR128, memopv2i64, i128mem,
3942 SSE_INTALU_ITINS_P, 1>;
3944 } // Constraints = "$src1 = $dst"
3946 //===---------------------------------------------------------------------===//
3947 // SSE2 - Packed Integer Logical Instructions
3948 //===---------------------------------------------------------------------===//
3950 let Predicates = [HasAVX] in {
3951 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3952 VR128, v8i16, v8i16, bc_v8i16,
3953 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3954 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3955 VR128, v4i32, v4i32, bc_v4i32,
3956 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3957 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3958 VR128, v2i64, v2i64, bc_v2i64,
3959 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3961 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3962 VR128, v8i16, v8i16, bc_v8i16,
3963 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3964 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3965 VR128, v4i32, v4i32, bc_v4i32,
3966 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3967 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3968 VR128, v2i64, v2i64, bc_v2i64,
3969 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3971 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3972 VR128, v8i16, v8i16, bc_v8i16,
3973 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3974 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3975 VR128, v4i32, v4i32, bc_v4i32,
3976 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3978 let ExeDomain = SSEPackedInt in {
3979 // 128-bit logical shifts.
3980 def VPSLLDQri : PDIi8<0x73, MRM7r,
3981 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3982 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3984 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3986 def VPSRLDQri : PDIi8<0x73, MRM3r,
3987 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3988 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3990 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3992 // PSRADQri doesn't exist in SSE[1-3].
3994 } // Predicates = [HasAVX]
3996 let Predicates = [HasAVX2] in {
3997 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3998 VR256, v16i16, v8i16, bc_v8i16,
3999 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4000 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
4001 VR256, v8i32, v4i32, bc_v4i32,
4002 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4003 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
4004 VR256, v4i64, v2i64, bc_v2i64,
4005 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4007 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
4008 VR256, v16i16, v8i16, bc_v8i16,
4009 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4010 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
4011 VR256, v8i32, v4i32, bc_v4i32,
4012 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4013 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
4014 VR256, v4i64, v2i64, bc_v2i64,
4015 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4017 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
4018 VR256, v16i16, v8i16, bc_v8i16,
4019 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4020 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
4021 VR256, v8i32, v4i32, bc_v4i32,
4022 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
4024 let ExeDomain = SSEPackedInt in {
4025 // 256-bit logical shifts.
4026 def VPSLLDQYri : PDIi8<0x73, MRM7r,
4027 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4028 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4030 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
4032 def VPSRLDQYri : PDIi8<0x73, MRM3r,
4033 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
4034 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4036 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
4038 // PSRADQYri doesn't exist in SSE[1-3].
4040 } // Predicates = [HasAVX2]
4042 let Constraints = "$src1 = $dst" in {
4043 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
4044 VR128, v8i16, v8i16, bc_v8i16,
4045 SSE_INTSHIFT_ITINS_P>;
4046 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
4047 VR128, v4i32, v4i32, bc_v4i32,
4048 SSE_INTSHIFT_ITINS_P>;
4049 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
4050 VR128, v2i64, v2i64, bc_v2i64,
4051 SSE_INTSHIFT_ITINS_P>;
4053 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
4054 VR128, v8i16, v8i16, bc_v8i16,
4055 SSE_INTSHIFT_ITINS_P>;
4056 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
4057 VR128, v4i32, v4i32, bc_v4i32,
4058 SSE_INTSHIFT_ITINS_P>;
4059 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
4060 VR128, v2i64, v2i64, bc_v2i64,
4061 SSE_INTSHIFT_ITINS_P>;
4063 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
4064 VR128, v8i16, v8i16, bc_v8i16,
4065 SSE_INTSHIFT_ITINS_P>;
4066 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
4067 VR128, v4i32, v4i32, bc_v4i32,
4068 SSE_INTSHIFT_ITINS_P>;
4070 let ExeDomain = SSEPackedInt in {
4071 // 128-bit logical shifts.
4072 def PSLLDQri : PDIi8<0x73, MRM7r,
4073 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4074 "pslldq\t{$src2, $dst|$dst, $src2}",
4076 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
4077 def PSRLDQri : PDIi8<0x73, MRM3r,
4078 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4079 "psrldq\t{$src2, $dst|$dst, $src2}",
4081 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
4082 // PSRADQri doesn't exist in SSE[1-3].
4084 } // Constraints = "$src1 = $dst"
4086 let Predicates = [HasAVX] in {
4087 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4088 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4089 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4090 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4091 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4092 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4094 // Shift up / down and insert zero's.
4095 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4096 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4097 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4098 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4101 let Predicates = [HasAVX2] in {
4102 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
4103 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4104 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
4105 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
4108 let Predicates = [HasSSE2] in {
4109 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
4110 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4111 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
4112 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4113 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
4114 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
4116 // Shift up / down and insert zero's.
4117 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
4118 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4119 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
4120 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
4123 //===---------------------------------------------------------------------===//
4124 // SSE2 - Packed Integer Comparison Instructions
4125 //===---------------------------------------------------------------------===//
4127 let Predicates = [HasAVX] in {
4128 defm VPCMPEQB : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v16i8,
4129 VR128, memopv2i64, i128mem,
4130 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4131 defm VPCMPEQW : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v8i16,
4132 VR128, memopv2i64, i128mem,
4133 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4134 defm VPCMPEQD : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v4i32,
4135 VR128, memopv2i64, i128mem,
4136 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4137 defm VPCMPGTB : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v16i8,
4138 VR128, memopv2i64, i128mem,
4139 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4140 defm VPCMPGTW : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v8i16,
4141 VR128, memopv2i64, i128mem,
4142 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4143 defm VPCMPGTD : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v4i32,
4144 VR128, memopv2i64, i128mem,
4145 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4148 let Predicates = [HasAVX2] in {
4149 defm VPCMPEQBY : PDI_binop_rm<0x74, "vpcmpeqb", X86pcmpeq, v32i8,
4150 VR256, memopv4i64, i256mem,
4151 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4152 defm VPCMPEQWY : PDI_binop_rm<0x75, "vpcmpeqw", X86pcmpeq, v16i16,
4153 VR256, memopv4i64, i256mem,
4154 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4155 defm VPCMPEQDY : PDI_binop_rm<0x76, "vpcmpeqd", X86pcmpeq, v8i32,
4156 VR256, memopv4i64, i256mem,
4157 SSE_INTALU_ITINS_P, 1, 0>, VEX_4V;
4158 defm VPCMPGTBY : PDI_binop_rm<0x64, "vpcmpgtb", X86pcmpgt, v32i8,
4159 VR256, memopv4i64, i256mem,
4160 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4161 defm VPCMPGTWY : PDI_binop_rm<0x65, "vpcmpgtw", X86pcmpgt, v16i16,
4162 VR256, memopv4i64, i256mem,
4163 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4164 defm VPCMPGTDY : PDI_binop_rm<0x66, "vpcmpgtd", X86pcmpgt, v8i32,
4165 VR256, memopv4i64, i256mem,
4166 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4169 let Constraints = "$src1 = $dst" in {
4170 defm PCMPEQB : PDI_binop_rm<0x74, "pcmpeqb", X86pcmpeq, v16i8,
4171 VR128, memopv2i64, i128mem,
4172 SSE_INTALU_ITINS_P, 1>;
4173 defm PCMPEQW : PDI_binop_rm<0x75, "pcmpeqw", X86pcmpeq, v8i16,
4174 VR128, memopv2i64, i128mem,
4175 SSE_INTALU_ITINS_P, 1>;
4176 defm PCMPEQD : PDI_binop_rm<0x76, "pcmpeqd", X86pcmpeq, v4i32,
4177 VR128, memopv2i64, i128mem,
4178 SSE_INTALU_ITINS_P, 1>;
4179 defm PCMPGTB : PDI_binop_rm<0x64, "pcmpgtb", X86pcmpgt, v16i8,
4180 VR128, memopv2i64, i128mem,
4181 SSE_INTALU_ITINS_P>;
4182 defm PCMPGTW : PDI_binop_rm<0x65, "pcmpgtw", X86pcmpgt, v8i16,
4183 VR128, memopv2i64, i128mem,
4184 SSE_INTALU_ITINS_P>;
4185 defm PCMPGTD : PDI_binop_rm<0x66, "pcmpgtd", X86pcmpgt, v4i32,
4186 VR128, memopv2i64, i128mem,
4187 SSE_INTALU_ITINS_P>;
4188 } // Constraints = "$src1 = $dst"
4190 //===---------------------------------------------------------------------===//
4191 // SSE2 - Packed Integer Pack Instructions
4192 //===---------------------------------------------------------------------===//
4194 let Predicates = [HasAVX] in {
4195 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4196 VR128, memopv2i64, i128mem,
4197 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4198 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4199 VR128, memopv2i64, i128mem,
4200 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4201 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4202 VR128, memopv2i64, i128mem,
4203 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4206 let Predicates = [HasAVX2] in {
4207 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4208 VR256, memopv4i64, i256mem,
4209 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4210 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4211 VR256, memopv4i64, i256mem,
4212 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4213 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4214 VR256, memopv4i64, i256mem,
4215 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
4218 let Constraints = "$src1 = $dst" in {
4219 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4220 VR128, memopv2i64, i128mem,
4221 SSE_INTALU_ITINS_P>;
4222 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4223 VR128, memopv2i64, i128mem,
4224 SSE_INTALU_ITINS_P>;
4225 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4226 VR128, memopv2i64, i128mem,
4227 SSE_INTALU_ITINS_P>;
4228 } // Constraints = "$src1 = $dst"
4230 //===---------------------------------------------------------------------===//
4231 // SSE2 - Packed Integer Shuffle Instructions
4232 //===---------------------------------------------------------------------===//
4234 let ExeDomain = SSEPackedInt in {
4235 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4236 def ri : Ii8<0x70, MRMSrcReg,
4237 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4238 !strconcat(OpcodeStr,
4239 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4240 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4242 def mi : Ii8<0x70, MRMSrcMem,
4243 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4244 !strconcat(OpcodeStr,
4245 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4247 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4252 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4253 def Yri : Ii8<0x70, MRMSrcReg,
4254 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4255 !strconcat(OpcodeStr,
4256 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4257 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4258 def Ymi : Ii8<0x70, MRMSrcMem,
4259 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4260 !strconcat(OpcodeStr,
4261 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4263 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4264 (i8 imm:$src2))))]>;
4266 } // ExeDomain = SSEPackedInt
4268 let Predicates = [HasAVX] in {
4269 let AddedComplexity = 5 in
4270 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4272 // SSE2 with ImmT == Imm8 and XS prefix.
4273 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4275 // SSE2 with ImmT == Imm8 and XD prefix.
4276 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4278 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4279 (VPSHUFDmi addr:$src1, imm:$imm)>;
4280 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4281 (VPSHUFDri VR128:$src1, imm:$imm)>;
4284 let Predicates = [HasAVX2] in {
4285 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>, TB, OpSize, VEX;
4286 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>, XS, VEX;
4287 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>, XD, VEX;
4290 let Predicates = [HasSSE2] in {
4291 let AddedComplexity = 5 in
4292 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4294 // SSE2 with ImmT == Imm8 and XS prefix.
4295 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4297 // SSE2 with ImmT == Imm8 and XD prefix.
4298 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4300 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4301 (PSHUFDmi addr:$src1, imm:$imm)>;
4302 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4303 (PSHUFDri VR128:$src1, imm:$imm)>;
4306 //===---------------------------------------------------------------------===//
4307 // SSE2 - Packed Integer Unpack Instructions
4308 //===---------------------------------------------------------------------===//
4310 let ExeDomain = SSEPackedInt in {
4311 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4312 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4313 def rr : PDI<opc, MRMSrcReg,
4314 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4316 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4317 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4318 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4320 def rm : PDI<opc, MRMSrcMem,
4321 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4323 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4324 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4325 [(set VR128:$dst, (OpNode VR128:$src1,
4326 (bc_frag (memopv2i64
4331 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4332 SDNode OpNode, PatFrag bc_frag> {
4333 def Yrr : PDI<opc, MRMSrcReg,
4334 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4335 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4336 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4337 def Yrm : PDI<opc, MRMSrcMem,
4338 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4339 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4340 [(set VR256:$dst, (OpNode VR256:$src1,
4341 (bc_frag (memopv4i64 addr:$src2))))]>;
4344 let Predicates = [HasAVX] in {
4345 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4346 bc_v16i8, 0>, VEX_4V;
4347 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4348 bc_v8i16, 0>, VEX_4V;
4349 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4350 bc_v4i32, 0>, VEX_4V;
4351 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4352 bc_v2i64, 0>, VEX_4V;
4354 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4355 bc_v16i8, 0>, VEX_4V;
4356 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4357 bc_v8i16, 0>, VEX_4V;
4358 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4359 bc_v4i32, 0>, VEX_4V;
4360 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4361 bc_v2i64, 0>, VEX_4V;
4364 let Predicates = [HasAVX2] in {
4365 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4367 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4369 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4371 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4374 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4376 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4378 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4380 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4384 let Constraints = "$src1 = $dst" in {
4385 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4387 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4389 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4391 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4394 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4396 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4398 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4400 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4403 } // ExeDomain = SSEPackedInt
4405 // Patterns for using AVX1 instructions with integer vectors
4406 // Here to give AVX2 priority
4407 let Predicates = [HasAVX] in {
4408 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4409 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4410 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4411 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4412 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4413 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4414 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4415 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4417 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4418 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4419 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4420 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4421 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4422 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4423 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4424 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4427 //===---------------------------------------------------------------------===//
4428 // SSE2 - Packed Integer Extract and Insert
4429 //===---------------------------------------------------------------------===//
4431 let ExeDomain = SSEPackedInt in {
4432 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4433 def rri : Ii8<0xC4, MRMSrcReg,
4434 (outs VR128:$dst), (ins VR128:$src1,
4435 GR32:$src2, i32i8imm:$src3),
4437 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4438 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4440 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4441 def rmi : Ii8<0xC4, MRMSrcMem,
4442 (outs VR128:$dst), (ins VR128:$src1,
4443 i16mem:$src2, i32i8imm:$src3),
4445 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4446 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4448 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4449 imm:$src3))], IIC_SSE_PINSRW>;
4453 let Predicates = [HasAVX] in
4454 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4455 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4456 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4457 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4458 imm:$src2))]>, TB, OpSize, VEX;
4459 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4460 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4461 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4462 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4463 imm:$src2))], IIC_SSE_PEXTRW>;
4466 let Predicates = [HasAVX] in {
4467 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4468 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4469 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4470 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4471 []>, TB, OpSize, VEX_4V;
4474 let Constraints = "$src1 = $dst" in
4475 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4477 } // ExeDomain = SSEPackedInt
4479 //===---------------------------------------------------------------------===//
4480 // SSE2 - Packed Mask Creation
4481 //===---------------------------------------------------------------------===//
4483 let ExeDomain = SSEPackedInt in {
4485 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4486 "pmovmskb\t{$src, $dst|$dst, $src}",
4487 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4488 IIC_SSE_MOVMSK>, VEX;
4489 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4490 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4492 let Predicates = [HasAVX2] in {
4493 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4494 "pmovmskb\t{$src, $dst|$dst, $src}",
4495 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4496 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4497 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4500 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4501 "pmovmskb\t{$src, $dst|$dst, $src}",
4502 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4505 } // ExeDomain = SSEPackedInt
4507 //===---------------------------------------------------------------------===//
4508 // SSE2 - Conditional Store
4509 //===---------------------------------------------------------------------===//
4511 let ExeDomain = SSEPackedInt in {
4514 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4515 (ins VR128:$src, VR128:$mask),
4516 "maskmovdqu\t{$mask, $src|$src, $mask}",
4517 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4518 IIC_SSE_MASKMOV>, VEX;
4520 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4521 (ins VR128:$src, VR128:$mask),
4522 "maskmovdqu\t{$mask, $src|$src, $mask}",
4523 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4524 IIC_SSE_MASKMOV>, VEX;
4527 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4528 "maskmovdqu\t{$mask, $src|$src, $mask}",
4529 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4532 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4533 "maskmovdqu\t{$mask, $src|$src, $mask}",
4534 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4537 } // ExeDomain = SSEPackedInt
4539 //===---------------------------------------------------------------------===//
4540 // SSE2 - Move Doubleword
4541 //===---------------------------------------------------------------------===//
4543 //===---------------------------------------------------------------------===//
4544 // Move Int Doubleword to Packed Double Int
4546 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4547 "movd\t{$src, $dst|$dst, $src}",
4549 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4551 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4552 "movd\t{$src, $dst|$dst, $src}",
4554 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4557 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4558 "mov{d|q}\t{$src, $dst|$dst, $src}",
4560 (v2i64 (scalar_to_vector GR64:$src)))],
4561 IIC_SSE_MOVDQ>, VEX;
4562 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4563 "mov{d|q}\t{$src, $dst|$dst, $src}",
4564 [(set FR64:$dst, (bitconvert GR64:$src))],
4565 IIC_SSE_MOVDQ>, VEX;
4567 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4568 "movd\t{$src, $dst|$dst, $src}",
4570 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4571 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4572 "movd\t{$src, $dst|$dst, $src}",
4574 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4576 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4577 "mov{d|q}\t{$src, $dst|$dst, $src}",
4579 (v2i64 (scalar_to_vector GR64:$src)))],
4581 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4582 "mov{d|q}\t{$src, $dst|$dst, $src}",
4583 [(set FR64:$dst, (bitconvert GR64:$src))],
4586 //===---------------------------------------------------------------------===//
4587 // Move Int Doubleword to Single Scalar
4589 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4590 "movd\t{$src, $dst|$dst, $src}",
4591 [(set FR32:$dst, (bitconvert GR32:$src))],
4592 IIC_SSE_MOVDQ>, VEX;
4594 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4595 "movd\t{$src, $dst|$dst, $src}",
4596 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4599 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4600 "movd\t{$src, $dst|$dst, $src}",
4601 [(set FR32:$dst, (bitconvert GR32:$src))],
4604 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4605 "movd\t{$src, $dst|$dst, $src}",
4606 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4609 //===---------------------------------------------------------------------===//
4610 // Move Packed Doubleword Int to Packed Double Int
4612 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4613 "movd\t{$src, $dst|$dst, $src}",
4614 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4615 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4616 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4617 (ins i32mem:$dst, VR128:$src),
4618 "movd\t{$src, $dst|$dst, $src}",
4619 [(store (i32 (vector_extract (v4i32 VR128:$src),
4620 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4622 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4623 "movd\t{$src, $dst|$dst, $src}",
4624 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4625 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4626 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4627 "movd\t{$src, $dst|$dst, $src}",
4628 [(store (i32 (vector_extract (v4i32 VR128:$src),
4629 (iPTR 0))), addr:$dst)],
4632 //===---------------------------------------------------------------------===//
4633 // Move Packed Doubleword Int first element to Doubleword Int
4635 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4636 "mov{d|q}\t{$src, $dst|$dst, $src}",
4637 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4640 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4642 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4643 "mov{d|q}\t{$src, $dst|$dst, $src}",
4644 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4648 //===---------------------------------------------------------------------===//
4649 // Bitcast FR64 <-> GR64
4651 let Predicates = [HasAVX] in
4652 def VMOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4653 "vmovq\t{$src, $dst|$dst, $src}",
4654 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4656 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4657 "mov{d|q}\t{$src, $dst|$dst, $src}",
4658 [(set GR64:$dst, (bitconvert FR64:$src))],
4659 IIC_SSE_MOVDQ>, VEX;
4660 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4661 "movq\t{$src, $dst|$dst, $src}",
4662 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4663 IIC_SSE_MOVDQ>, VEX;
4665 def MOV64toSDrm : SSDI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4666 "movq\t{$src, $dst|$dst, $src}",
4667 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4669 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4670 "mov{d|q}\t{$src, $dst|$dst, $src}",
4671 [(set GR64:$dst, (bitconvert FR64:$src))],
4673 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4674 "movq\t{$src, $dst|$dst, $src}",
4675 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4678 //===---------------------------------------------------------------------===//
4679 // Move Scalar Single to Double Int
4681 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4682 "movd\t{$src, $dst|$dst, $src}",
4683 [(set GR32:$dst, (bitconvert FR32:$src))],
4684 IIC_SSE_MOVD_ToGP>, VEX;
4685 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4686 "movd\t{$src, $dst|$dst, $src}",
4687 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4688 IIC_SSE_MOVDQ>, VEX;
4689 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4690 "movd\t{$src, $dst|$dst, $src}",
4691 [(set GR32:$dst, (bitconvert FR32:$src))],
4693 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4694 "movd\t{$src, $dst|$dst, $src}",
4695 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4698 //===---------------------------------------------------------------------===//
4699 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4701 let AddedComplexity = 15 in {
4702 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4703 "movd\t{$src, $dst|$dst, $src}",
4704 [(set VR128:$dst, (v4i32 (X86vzmovl
4705 (v4i32 (scalar_to_vector GR32:$src)))))],
4706 IIC_SSE_MOVDQ>, VEX;
4707 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4708 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4709 [(set VR128:$dst, (v2i64 (X86vzmovl
4710 (v2i64 (scalar_to_vector GR64:$src)))))],
4714 let AddedComplexity = 15 in {
4715 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4716 "movd\t{$src, $dst|$dst, $src}",
4717 [(set VR128:$dst, (v4i32 (X86vzmovl
4718 (v4i32 (scalar_to_vector GR32:$src)))))],
4720 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4721 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4722 [(set VR128:$dst, (v2i64 (X86vzmovl
4723 (v2i64 (scalar_to_vector GR64:$src)))))],
4727 let AddedComplexity = 20 in {
4728 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4729 "movd\t{$src, $dst|$dst, $src}",
4731 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4732 (loadi32 addr:$src))))))],
4733 IIC_SSE_MOVDQ>, VEX;
4734 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4735 "movd\t{$src, $dst|$dst, $src}",
4737 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4738 (loadi32 addr:$src))))))],
4742 let Predicates = [HasAVX] in {
4743 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4744 let AddedComplexity = 20 in {
4745 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4746 (VMOVZDI2PDIrm addr:$src)>;
4747 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4748 (VMOVZDI2PDIrm addr:$src)>;
4750 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4751 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4752 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4753 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4754 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4755 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4756 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4759 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4760 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4761 (MOVZDI2PDIrm addr:$src)>;
4762 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4763 (MOVZDI2PDIrm addr:$src)>;
4766 // These are the correct encodings of the instructions so that we know how to
4767 // read correct assembly, even though we continue to emit the wrong ones for
4768 // compatibility with Darwin's buggy assembler.
4769 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4770 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4771 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4772 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4773 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4774 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4775 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4776 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4777 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4778 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4779 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4780 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4782 //===---------------------------------------------------------------------===//
4783 // SSE2 - Move Quadword
4784 //===---------------------------------------------------------------------===//
4786 //===---------------------------------------------------------------------===//
4787 // Move Quadword Int to Packed Quadword Int
4789 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4790 "vmovq\t{$src, $dst|$dst, $src}",
4792 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4793 VEX, Requires<[HasAVX]>;
4794 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4795 "movq\t{$src, $dst|$dst, $src}",
4797 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4799 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4801 //===---------------------------------------------------------------------===//
4802 // Move Packed Quadword Int to Quadword Int
4804 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4805 "movq\t{$src, $dst|$dst, $src}",
4806 [(store (i64 (vector_extract (v2i64 VR128:$src),
4807 (iPTR 0))), addr:$dst)],
4808 IIC_SSE_MOVDQ>, VEX;
4809 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4810 "movq\t{$src, $dst|$dst, $src}",
4811 [(store (i64 (vector_extract (v2i64 VR128:$src),
4812 (iPTR 0))), addr:$dst)],
4815 //===---------------------------------------------------------------------===//
4816 // Store / copy lower 64-bits of a XMM register.
4818 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4819 "movq\t{$src, $dst|$dst, $src}",
4820 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4821 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4822 "movq\t{$src, $dst|$dst, $src}",
4823 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4826 let AddedComplexity = 20 in
4827 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4828 "vmovq\t{$src, $dst|$dst, $src}",
4830 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4831 (loadi64 addr:$src))))))],
4833 XS, VEX, Requires<[HasAVX]>;
4835 let AddedComplexity = 20 in
4836 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4837 "movq\t{$src, $dst|$dst, $src}",
4839 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4840 (loadi64 addr:$src))))))],
4842 XS, Requires<[HasSSE2]>;
4844 let Predicates = [HasAVX], AddedComplexity = 20 in {
4845 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4846 (VMOVZQI2PQIrm addr:$src)>;
4847 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4848 (VMOVZQI2PQIrm addr:$src)>;
4849 def : Pat<(v2i64 (X86vzload addr:$src)),
4850 (VMOVZQI2PQIrm addr:$src)>;
4853 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4854 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4855 (MOVZQI2PQIrm addr:$src)>;
4856 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4857 (MOVZQI2PQIrm addr:$src)>;
4858 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4861 let Predicates = [HasAVX] in {
4862 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4863 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4864 def : Pat<(v4i64 (X86vzload addr:$src)),
4865 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4868 //===---------------------------------------------------------------------===//
4869 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4870 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4872 let AddedComplexity = 15 in
4873 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4874 "vmovq\t{$src, $dst|$dst, $src}",
4875 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4877 XS, VEX, Requires<[HasAVX]>;
4878 let AddedComplexity = 15 in
4879 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4880 "movq\t{$src, $dst|$dst, $src}",
4881 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4883 XS, Requires<[HasSSE2]>;
4885 let AddedComplexity = 20 in
4886 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4887 "vmovq\t{$src, $dst|$dst, $src}",
4888 [(set VR128:$dst, (v2i64 (X86vzmovl
4889 (loadv2i64 addr:$src))))],
4891 XS, VEX, Requires<[HasAVX]>;
4892 let AddedComplexity = 20 in {
4893 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4894 "movq\t{$src, $dst|$dst, $src}",
4895 [(set VR128:$dst, (v2i64 (X86vzmovl
4896 (loadv2i64 addr:$src))))],
4898 XS, Requires<[HasSSE2]>;
4901 let AddedComplexity = 20 in {
4902 let Predicates = [HasAVX] in {
4903 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4904 (VMOVZPQILo2PQIrm addr:$src)>;
4905 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4906 (VMOVZPQILo2PQIrr VR128:$src)>;
4908 let Predicates = [HasSSE2] in {
4909 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4910 (MOVZPQILo2PQIrm addr:$src)>;
4911 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4912 (MOVZPQILo2PQIrr VR128:$src)>;
4916 // Instructions to match in the assembler
4917 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4918 "movq\t{$src, $dst|$dst, $src}", [],
4919 IIC_SSE_MOVDQ>, VEX, VEX_W;
4920 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4921 "movq\t{$src, $dst|$dst, $src}", [],
4922 IIC_SSE_MOVDQ>, VEX, VEX_W;
4923 // Recognize "movd" with GR64 destination, but encode as a "movq"
4924 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4925 "movd\t{$src, $dst|$dst, $src}", [],
4926 IIC_SSE_MOVDQ>, VEX, VEX_W;
4928 // Instructions for the disassembler
4929 // xr = XMM register
4932 let Predicates = [HasAVX] in
4933 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4934 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4935 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4936 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4938 //===---------------------------------------------------------------------===//
4939 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4940 //===---------------------------------------------------------------------===//
4941 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4942 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4943 X86MemOperand x86memop> {
4944 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4945 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4946 [(set RC:$dst, (vt (OpNode RC:$src)))],
4948 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4949 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4950 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4954 let Predicates = [HasAVX] in {
4955 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4956 v4f32, VR128, memopv4f32, f128mem>, VEX;
4957 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4958 v4f32, VR128, memopv4f32, f128mem>, VEX;
4959 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4960 v8f32, VR256, memopv8f32, f256mem>, VEX;
4961 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4962 v8f32, VR256, memopv8f32, f256mem>, VEX;
4964 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4965 memopv4f32, f128mem>;
4966 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4967 memopv4f32, f128mem>;
4969 let Predicates = [HasAVX] in {
4970 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4971 (VMOVSHDUPrr VR128:$src)>;
4972 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4973 (VMOVSHDUPrm addr:$src)>;
4974 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4975 (VMOVSLDUPrr VR128:$src)>;
4976 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4977 (VMOVSLDUPrm addr:$src)>;
4978 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4979 (VMOVSHDUPYrr VR256:$src)>;
4980 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4981 (VMOVSHDUPYrm addr:$src)>;
4982 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4983 (VMOVSLDUPYrr VR256:$src)>;
4984 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4985 (VMOVSLDUPYrm addr:$src)>;
4988 let Predicates = [HasSSE3] in {
4989 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4990 (MOVSHDUPrr VR128:$src)>;
4991 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4992 (MOVSHDUPrm addr:$src)>;
4993 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4994 (MOVSLDUPrr VR128:$src)>;
4995 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4996 (MOVSLDUPrm addr:$src)>;
4999 //===---------------------------------------------------------------------===//
5000 // SSE3 - Replicate Double FP - MOVDDUP
5001 //===---------------------------------------------------------------------===//
5003 multiclass sse3_replicate_dfp<string OpcodeStr> {
5004 let neverHasSideEffects = 1 in
5005 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5006 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5007 [], IIC_SSE_MOV_LH>;
5008 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
5009 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5012 (scalar_to_vector (loadf64 addr:$src)))))],
5016 // FIXME: Merge with above classe when there're patterns for the ymm version
5017 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
5018 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
5019 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5020 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
5021 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
5022 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5025 (scalar_to_vector (loadf64 addr:$src)))))]>;
5028 let Predicates = [HasAVX] in {
5029 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
5030 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
5033 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
5035 let Predicates = [HasAVX] in {
5036 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5037 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5038 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5039 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5040 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5041 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5042 def : Pat<(X86Movddup (bc_v2f64
5043 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5044 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5047 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5048 (VMOVDDUPYrm addr:$src)>;
5049 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5050 (VMOVDDUPYrm addr:$src)>;
5051 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5052 (VMOVDDUPYrm addr:$src)>;
5053 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5054 (VMOVDDUPYrr VR256:$src)>;
5057 let Predicates = [HasSSE3] in {
5058 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5059 (MOVDDUPrm addr:$src)>;
5060 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5061 (MOVDDUPrm addr:$src)>;
5062 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5063 (MOVDDUPrm addr:$src)>;
5064 def : Pat<(X86Movddup (bc_v2f64
5065 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5066 (MOVDDUPrm addr:$src)>;
5069 //===---------------------------------------------------------------------===//
5070 // SSE3 - Move Unaligned Integer
5071 //===---------------------------------------------------------------------===//
5073 let Predicates = [HasAVX] in {
5074 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5075 "vlddqu\t{$src, $dst|$dst, $src}",
5076 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5077 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5078 "vlddqu\t{$src, $dst|$dst, $src}",
5079 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5081 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5082 "lddqu\t{$src, $dst|$dst, $src}",
5083 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
5086 //===---------------------------------------------------------------------===//
5087 // SSE3 - Arithmetic
5088 //===---------------------------------------------------------------------===//
5090 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5091 X86MemOperand x86memop, OpndItins itins,
5093 def rr : I<0xD0, MRMSrcReg,
5094 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5096 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5097 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5098 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
5099 def rm : I<0xD0, MRMSrcMem,
5100 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5102 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5103 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5104 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
5107 let Predicates = [HasAVX] in {
5108 let ExeDomain = SSEPackedSingle in {
5109 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5110 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5111 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5112 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
5114 let ExeDomain = SSEPackedDouble in {
5115 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5116 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5117 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5118 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
5121 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5122 let ExeDomain = SSEPackedSingle in
5123 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5124 f128mem, SSE_ALU_F32P>, TB, XD;
5125 let ExeDomain = SSEPackedDouble in
5126 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5127 f128mem, SSE_ALU_F64P>, TB, OpSize;
5130 //===---------------------------------------------------------------------===//
5131 // SSE3 Instructions
5132 //===---------------------------------------------------------------------===//
5135 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5136 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5137 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5139 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5140 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5141 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5143 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5145 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5146 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5147 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5148 IIC_SSE_HADDSUB_RM>;
5150 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5151 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5152 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5154 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5155 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5156 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
5158 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5160 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5161 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5162 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
5163 IIC_SSE_HADDSUB_RM>;
5166 let Predicates = [HasAVX] in {
5167 let ExeDomain = SSEPackedSingle in {
5168 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5169 X86fhadd, 0>, VEX_4V;
5170 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5171 X86fhsub, 0>, VEX_4V;
5172 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5173 X86fhadd, 0>, VEX_4V;
5174 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5175 X86fhsub, 0>, VEX_4V;
5177 let ExeDomain = SSEPackedDouble in {
5178 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5179 X86fhadd, 0>, VEX_4V;
5180 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5181 X86fhsub, 0>, VEX_4V;
5182 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5183 X86fhadd, 0>, VEX_4V;
5184 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5185 X86fhsub, 0>, VEX_4V;
5189 let Constraints = "$src1 = $dst" in {
5190 let ExeDomain = SSEPackedSingle in {
5191 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5192 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5194 let ExeDomain = SSEPackedDouble in {
5195 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5196 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5200 //===---------------------------------------------------------------------===//
5201 // SSSE3 - Packed Absolute Instructions
5202 //===---------------------------------------------------------------------===//
5205 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5206 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5207 Intrinsic IntId128> {
5208 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5210 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5211 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
5214 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5216 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5219 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
5223 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5224 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5225 Intrinsic IntId256> {
5226 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5228 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5229 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5232 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5234 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5237 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5240 let Predicates = [HasAVX] in {
5241 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5242 int_x86_ssse3_pabs_b_128>, VEX;
5243 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5244 int_x86_ssse3_pabs_w_128>, VEX;
5245 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5246 int_x86_ssse3_pabs_d_128>, VEX;
5249 let Predicates = [HasAVX2] in {
5250 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5251 int_x86_avx2_pabs_b>, VEX;
5252 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5253 int_x86_avx2_pabs_w>, VEX;
5254 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5255 int_x86_avx2_pabs_d>, VEX;
5258 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5259 int_x86_ssse3_pabs_b_128>;
5260 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5261 int_x86_ssse3_pabs_w_128>;
5262 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5263 int_x86_ssse3_pabs_d_128>;
5265 //===---------------------------------------------------------------------===//
5266 // SSSE3 - Packed Binary Operator Instructions
5267 //===---------------------------------------------------------------------===//
5269 def SSE_PHADDSUBD : OpndItins<
5270 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5272 def SSE_PHADDSUBSW : OpndItins<
5273 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5275 def SSE_PHADDSUBW : OpndItins<
5276 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5278 def SSE_PSHUFB : OpndItins<
5279 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5281 def SSE_PSIGN : OpndItins<
5282 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5284 def SSE_PMULHRSW : OpndItins<
5285 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5288 /// SS3I_binop_rm - Simple SSSE3 bin op
5289 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5290 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5291 X86MemOperand x86memop, OpndItins itins,
5293 let isCommutable = 1 in
5294 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5295 (ins RC:$src1, RC:$src2),
5297 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5298 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5299 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5301 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5302 (ins RC:$src1, x86memop:$src2),
5304 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5305 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5307 (OpVT (OpNode RC:$src1,
5308 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5311 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5312 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5313 Intrinsic IntId128, OpndItins itins,
5315 let isCommutable = 1 in
5316 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5317 (ins VR128:$src1, VR128:$src2),
5319 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5321 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5323 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5324 (ins VR128:$src1, i128mem:$src2),
5326 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5327 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5329 (IntId128 VR128:$src1,
5330 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5333 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5334 Intrinsic IntId256> {
5335 let isCommutable = 1 in
5336 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5337 (ins VR256:$src1, VR256:$src2),
5338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5339 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5341 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5342 (ins VR256:$src1, i256mem:$src2),
5343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5345 (IntId256 VR256:$src1,
5346 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5349 let ImmT = NoImm, Predicates = [HasAVX] in {
5350 let isCommutable = 0 in {
5351 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5352 memopv2i64, i128mem,
5353 SSE_PHADDSUBW, 0>, VEX_4V;
5354 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5355 memopv2i64, i128mem,
5356 SSE_PHADDSUBD, 0>, VEX_4V;
5357 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5358 memopv2i64, i128mem,
5359 SSE_PHADDSUBW, 0>, VEX_4V;
5360 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5361 memopv2i64, i128mem,
5362 SSE_PHADDSUBD, 0>, VEX_4V;
5363 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5364 memopv2i64, i128mem,
5365 SSE_PSIGN, 0>, VEX_4V;
5366 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5367 memopv2i64, i128mem,
5368 SSE_PSIGN, 0>, VEX_4V;
5369 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5370 memopv2i64, i128mem,
5371 SSE_PSIGN, 0>, VEX_4V;
5372 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5373 memopv2i64, i128mem,
5374 SSE_PSHUFB, 0>, VEX_4V;
5375 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5376 int_x86_ssse3_phadd_sw_128,
5377 SSE_PHADDSUBSW, 0>, VEX_4V;
5378 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5379 int_x86_ssse3_phsub_sw_128,
5380 SSE_PHADDSUBSW, 0>, VEX_4V;
5381 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5382 int_x86_ssse3_pmadd_ub_sw_128,
5383 SSE_PMADD, 0>, VEX_4V;
5385 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5386 int_x86_ssse3_pmul_hr_sw_128,
5387 SSE_PMULHRSW, 0>, VEX_4V;
5390 let ImmT = NoImm, Predicates = [HasAVX2] in {
5391 let isCommutable = 0 in {
5392 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5393 memopv4i64, i256mem,
5394 SSE_PHADDSUBW, 0>, VEX_4V;
5395 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5396 memopv4i64, i256mem,
5397 SSE_PHADDSUBW, 0>, VEX_4V;
5398 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5399 memopv4i64, i256mem,
5400 SSE_PHADDSUBW, 0>, VEX_4V;
5401 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5402 memopv4i64, i256mem,
5403 SSE_PHADDSUBW, 0>, VEX_4V;
5404 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5405 memopv4i64, i256mem,
5406 SSE_PHADDSUBW, 0>, VEX_4V;
5407 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5408 memopv4i64, i256mem,
5409 SSE_PHADDSUBW, 0>, VEX_4V;
5410 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5411 memopv4i64, i256mem,
5412 SSE_PHADDSUBW, 0>, VEX_4V;
5413 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5414 memopv4i64, i256mem,
5415 SSE_PHADDSUBW, 0>, VEX_4V;
5416 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5417 int_x86_avx2_phadd_sw>, VEX_4V;
5418 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5419 int_x86_avx2_phsub_sw>, VEX_4V;
5420 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5421 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5423 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5424 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5427 // None of these have i8 immediate fields.
5428 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5429 let isCommutable = 0 in {
5430 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5431 memopv2i64, i128mem, SSE_PHADDSUBW>;
5432 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5433 memopv2i64, i128mem, SSE_PHADDSUBD>;
5434 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5435 memopv2i64, i128mem, SSE_PHADDSUBW>;
5436 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5437 memopv2i64, i128mem, SSE_PHADDSUBD>;
5438 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5439 memopv2i64, i128mem, SSE_PSIGN>;
5440 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5441 memopv2i64, i128mem, SSE_PSIGN>;
5442 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5443 memopv2i64, i128mem, SSE_PSIGN>;
5444 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5445 memopv2i64, i128mem, SSE_PSHUFB>;
5446 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5447 int_x86_ssse3_phadd_sw_128,
5449 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5450 int_x86_ssse3_phsub_sw_128,
5452 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5453 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5455 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5456 int_x86_ssse3_pmul_hr_sw_128,
5460 //===---------------------------------------------------------------------===//
5461 // SSSE3 - Packed Align Instruction Patterns
5462 //===---------------------------------------------------------------------===//
5464 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5465 let neverHasSideEffects = 1 in {
5466 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5467 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5469 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5471 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5472 [], IIC_SSE_PALIGNR>, OpSize;
5474 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5475 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5477 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5479 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5480 [], IIC_SSE_PALIGNR>, OpSize;
5484 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5485 let neverHasSideEffects = 1 in {
5486 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5487 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5489 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5492 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5493 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5495 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5500 let Predicates = [HasAVX] in
5501 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5502 let Predicates = [HasAVX2] in
5503 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5504 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5505 defm PALIGN : ssse3_palign<"palignr">;
5507 let Predicates = [HasAVX2] in {
5508 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5509 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5510 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5511 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5512 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5513 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5514 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5515 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5518 let Predicates = [HasAVX] in {
5519 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5520 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5521 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5522 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5523 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5524 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5525 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5526 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5529 let Predicates = [HasSSSE3] in {
5530 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5531 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5532 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5533 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5534 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5535 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5536 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5537 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5540 //===---------------------------------------------------------------------===//
5541 // SSSE3 - Thread synchronization
5542 //===---------------------------------------------------------------------===//
5544 let usesCustomInserter = 1 in {
5545 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5546 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5547 Requires<[HasSSE3]>;
5548 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5549 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5550 Requires<[HasSSE3]>;
5553 let Uses = [EAX, ECX, EDX] in
5554 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5555 TB, Requires<[HasSSE3]>;
5556 let Uses = [ECX, EAX] in
5557 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", [], IIC_SSE_MWAIT>,
5558 TB, Requires<[HasSSE3]>;
5560 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5561 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5563 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5564 Requires<[In32BitMode]>;
5565 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5566 Requires<[In64BitMode]>;
5568 //===----------------------------------------------------------------------===//
5569 // SSE4.1 - Packed Move with Sign/Zero Extend
5570 //===----------------------------------------------------------------------===//
5572 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5573 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5574 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5575 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5577 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5578 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5580 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5584 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5586 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5587 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5588 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5590 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5591 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5592 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5595 let Predicates = [HasAVX] in {
5596 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5598 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5600 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5602 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5604 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5606 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5610 let Predicates = [HasAVX2] in {
5611 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5612 int_x86_avx2_pmovsxbw>, VEX;
5613 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5614 int_x86_avx2_pmovsxwd>, VEX;
5615 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5616 int_x86_avx2_pmovsxdq>, VEX;
5617 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5618 int_x86_avx2_pmovzxbw>, VEX;
5619 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5620 int_x86_avx2_pmovzxwd>, VEX;
5621 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5622 int_x86_avx2_pmovzxdq>, VEX;
5625 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5626 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5627 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5628 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5629 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5630 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5632 let Predicates = [HasAVX] in {
5633 // Common patterns involving scalar load.
5634 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5635 (VPMOVSXBWrm addr:$src)>;
5636 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5637 (VPMOVSXBWrm addr:$src)>;
5639 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5640 (VPMOVSXWDrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5642 (VPMOVSXWDrm addr:$src)>;
5644 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5645 (VPMOVSXDQrm addr:$src)>;
5646 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5647 (VPMOVSXDQrm addr:$src)>;
5649 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5650 (VPMOVZXBWrm addr:$src)>;
5651 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5652 (VPMOVZXBWrm addr:$src)>;
5654 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5655 (VPMOVZXWDrm addr:$src)>;
5656 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5657 (VPMOVZXWDrm addr:$src)>;
5659 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5660 (VPMOVZXDQrm addr:$src)>;
5661 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5662 (VPMOVZXDQrm addr:$src)>;
5665 let Predicates = [HasSSE41] in {
5666 // Common patterns involving scalar load.
5667 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5668 (PMOVSXBWrm addr:$src)>;
5669 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5670 (PMOVSXBWrm addr:$src)>;
5672 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5673 (PMOVSXWDrm addr:$src)>;
5674 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5675 (PMOVSXWDrm addr:$src)>;
5677 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5678 (PMOVSXDQrm addr:$src)>;
5679 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5680 (PMOVSXDQrm addr:$src)>;
5682 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5683 (PMOVZXBWrm addr:$src)>;
5684 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5685 (PMOVZXBWrm addr:$src)>;
5687 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5688 (PMOVZXWDrm addr:$src)>;
5689 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5690 (PMOVZXWDrm addr:$src)>;
5692 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5693 (PMOVZXDQrm addr:$src)>;
5694 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5695 (PMOVZXDQrm addr:$src)>;
5698 let Predicates = [HasAVX2] in {
5699 let AddedComplexity = 15 in {
5700 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5701 (VPMOVZXDQYrr VR128:$src)>;
5702 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5703 (VPMOVZXWDYrr VR128:$src)>;
5706 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5707 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5710 let Predicates = [HasAVX] in {
5711 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5712 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5715 let Predicates = [HasSSE41] in {
5716 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5717 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5721 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5722 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5723 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5724 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5726 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5727 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5729 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5733 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5735 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5736 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5737 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5739 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5740 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5742 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5746 let Predicates = [HasAVX] in {
5747 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5749 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5751 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5753 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5757 let Predicates = [HasAVX2] in {
5758 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5759 int_x86_avx2_pmovsxbd>, VEX;
5760 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5761 int_x86_avx2_pmovsxwq>, VEX;
5762 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5763 int_x86_avx2_pmovzxbd>, VEX;
5764 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5765 int_x86_avx2_pmovzxwq>, VEX;
5768 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5769 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5770 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5771 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5773 let Predicates = [HasAVX] in {
5774 // Common patterns involving scalar load
5775 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5776 (VPMOVSXBDrm addr:$src)>;
5777 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5778 (VPMOVSXWQrm addr:$src)>;
5780 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5781 (VPMOVZXBDrm addr:$src)>;
5782 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5783 (VPMOVZXWQrm addr:$src)>;
5786 let Predicates = [HasSSE41] in {
5787 // Common patterns involving scalar load
5788 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5789 (PMOVSXBDrm addr:$src)>;
5790 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5791 (PMOVSXWQrm addr:$src)>;
5793 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5794 (PMOVZXBDrm addr:$src)>;
5795 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5796 (PMOVZXWQrm addr:$src)>;
5799 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5800 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5801 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5802 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5804 // Expecting a i16 load any extended to i32 value.
5805 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5806 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5807 [(set VR128:$dst, (IntId (bitconvert
5808 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5812 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5814 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5815 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5816 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5818 // Expecting a i16 load any extended to i32 value.
5819 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5820 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5821 [(set VR256:$dst, (IntId (bitconvert
5822 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5826 let Predicates = [HasAVX] in {
5827 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5829 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5832 let Predicates = [HasAVX2] in {
5833 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5834 int_x86_avx2_pmovsxbq>, VEX;
5835 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5836 int_x86_avx2_pmovzxbq>, VEX;
5838 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5839 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5841 let Predicates = [HasAVX] in {
5842 // Common patterns involving scalar load
5843 def : Pat<(int_x86_sse41_pmovsxbq
5844 (bitconvert (v4i32 (X86vzmovl
5845 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5846 (VPMOVSXBQrm addr:$src)>;
5848 def : Pat<(int_x86_sse41_pmovzxbq
5849 (bitconvert (v4i32 (X86vzmovl
5850 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5851 (VPMOVZXBQrm addr:$src)>;
5854 let Predicates = [HasSSE41] in {
5855 // Common patterns involving scalar load
5856 def : Pat<(int_x86_sse41_pmovsxbq
5857 (bitconvert (v4i32 (X86vzmovl
5858 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5859 (PMOVSXBQrm addr:$src)>;
5861 def : Pat<(int_x86_sse41_pmovzxbq
5862 (bitconvert (v4i32 (X86vzmovl
5863 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5864 (PMOVZXBQrm addr:$src)>;
5867 //===----------------------------------------------------------------------===//
5868 // SSE4.1 - Extract Instructions
5869 //===----------------------------------------------------------------------===//
5871 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5872 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5873 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5874 (ins VR128:$src1, i32i8imm:$src2),
5875 !strconcat(OpcodeStr,
5876 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5877 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5879 let neverHasSideEffects = 1, mayStore = 1 in
5880 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5881 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5882 !strconcat(OpcodeStr,
5883 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5886 // There's an AssertZext in the way of writing the store pattern
5887 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5890 let Predicates = [HasAVX] in {
5891 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5892 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5893 (ins VR128:$src1, i32i8imm:$src2),
5894 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5897 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5900 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5901 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5902 let neverHasSideEffects = 1, mayStore = 1 in
5903 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5904 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5905 !strconcat(OpcodeStr,
5906 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5909 // There's an AssertZext in the way of writing the store pattern
5910 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5913 let Predicates = [HasAVX] in
5914 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5916 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5919 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5920 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5921 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5922 (ins VR128:$src1, i32i8imm:$src2),
5923 !strconcat(OpcodeStr,
5924 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5926 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5927 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5928 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5929 !strconcat(OpcodeStr,
5930 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5931 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5932 addr:$dst)]>, OpSize;
5935 let Predicates = [HasAVX] in
5936 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5938 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5940 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5941 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5942 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5943 (ins VR128:$src1, i32i8imm:$src2),
5944 !strconcat(OpcodeStr,
5945 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5947 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5948 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5949 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5950 !strconcat(OpcodeStr,
5951 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5952 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5953 addr:$dst)]>, OpSize, REX_W;
5956 let Predicates = [HasAVX] in
5957 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5959 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5961 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5963 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5964 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5965 (ins VR128:$src1, i32i8imm:$src2),
5966 !strconcat(OpcodeStr,
5967 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5969 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5971 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5972 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5973 !strconcat(OpcodeStr,
5974 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5975 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5976 addr:$dst)]>, OpSize;
5979 let ExeDomain = SSEPackedSingle in {
5980 let Predicates = [HasAVX] in {
5981 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5982 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5983 (ins VR128:$src1, i32i8imm:$src2),
5984 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5987 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5990 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5991 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5994 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5996 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5999 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
6000 Requires<[HasSSE41]>;
6002 //===----------------------------------------------------------------------===//
6003 // SSE4.1 - Insert Instructions
6004 //===----------------------------------------------------------------------===//
6006 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
6007 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6008 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6010 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6012 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6014 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
6015 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6016 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
6018 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6020 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6022 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
6023 imm:$src3))]>, OpSize;
6026 let Predicates = [HasAVX] in
6027 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
6028 let Constraints = "$src1 = $dst" in
6029 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6031 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6032 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6033 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6035 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6037 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6039 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6041 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6042 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6044 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6046 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6048 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6049 imm:$src3)))]>, OpSize;
6052 let Predicates = [HasAVX] in
6053 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6054 let Constraints = "$src1 = $dst" in
6055 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6057 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6058 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6059 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6061 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6063 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6065 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6067 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6068 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6070 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6072 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6074 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6075 imm:$src3)))]>, OpSize;
6078 let Predicates = [HasAVX] in
6079 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6080 let Constraints = "$src1 = $dst" in
6081 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6083 // insertps has a few different modes, there's the first two here below which
6084 // are optimized inserts that won't zero arbitrary elements in the destination
6085 // vector. The next one matches the intrinsic and could zero arbitrary elements
6086 // in the target vector.
6087 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6088 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6089 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6091 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6093 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6095 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6097 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6098 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6100 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6102 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6104 (X86insrtps VR128:$src1,
6105 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6106 imm:$src3))]>, OpSize;
6109 let ExeDomain = SSEPackedSingle in {
6110 let Predicates = [HasAVX] in
6111 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6112 let Constraints = "$src1 = $dst" in
6113 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6116 //===----------------------------------------------------------------------===//
6117 // SSE4.1 - Round Instructions
6118 //===----------------------------------------------------------------------===//
6120 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6121 X86MemOperand x86memop, RegisterClass RC,
6122 PatFrag mem_frag32, PatFrag mem_frag64,
6123 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6124 let ExeDomain = SSEPackedSingle in {
6125 // Intrinsic operation, reg.
6126 // Vector intrinsic operation, reg
6127 def PSr : SS4AIi8<opcps, MRMSrcReg,
6128 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6129 !strconcat(OpcodeStr,
6130 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6131 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6134 // Vector intrinsic operation, mem
6135 def PSm : SS4AIi8<opcps, MRMSrcMem,
6136 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6137 !strconcat(OpcodeStr,
6138 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6140 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6142 } // ExeDomain = SSEPackedSingle
6144 let ExeDomain = SSEPackedDouble in {
6145 // Vector intrinsic operation, reg
6146 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6147 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6148 !strconcat(OpcodeStr,
6149 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6150 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6153 // Vector intrinsic operation, mem
6154 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6155 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6156 !strconcat(OpcodeStr,
6157 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6159 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6161 } // ExeDomain = SSEPackedDouble
6164 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6167 Intrinsic F64Int, bit Is2Addr = 1> {
6168 let ExeDomain = GenericDomain in {
6170 def SSr : SS4AIi8<opcss, MRMSrcReg,
6171 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6173 !strconcat(OpcodeStr,
6174 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6175 !strconcat(OpcodeStr,
6176 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6179 // Intrinsic operation, reg.
6180 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6181 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6183 !strconcat(OpcodeStr,
6184 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6185 !strconcat(OpcodeStr,
6186 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6187 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6190 // Intrinsic operation, mem.
6191 def SSm : SS4AIi8<opcss, MRMSrcMem,
6192 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6194 !strconcat(OpcodeStr,
6195 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6196 !strconcat(OpcodeStr,
6197 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6199 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6203 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6204 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6206 !strconcat(OpcodeStr,
6207 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6208 !strconcat(OpcodeStr,
6209 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6212 // Intrinsic operation, reg.
6213 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6214 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6216 !strconcat(OpcodeStr,
6217 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6218 !strconcat(OpcodeStr,
6219 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6220 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6223 // Intrinsic operation, mem.
6224 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6225 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6227 !strconcat(OpcodeStr,
6228 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6229 !strconcat(OpcodeStr,
6230 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6232 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6234 } // ExeDomain = GenericDomain
6237 // FP round - roundss, roundps, roundsd, roundpd
6238 let Predicates = [HasAVX] in {
6240 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6241 memopv4f32, memopv2f64,
6242 int_x86_sse41_round_ps,
6243 int_x86_sse41_round_pd>, VEX;
6244 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6245 memopv8f32, memopv4f64,
6246 int_x86_avx_round_ps_256,
6247 int_x86_avx_round_pd_256>, VEX;
6248 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6249 int_x86_sse41_round_ss,
6250 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6252 def : Pat<(ffloor FR32:$src),
6253 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6254 def : Pat<(f64 (ffloor FR64:$src)),
6255 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6256 def : Pat<(f32 (fnearbyint FR32:$src)),
6257 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6258 def : Pat<(f64 (fnearbyint FR64:$src)),
6259 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6260 def : Pat<(f32 (fceil FR32:$src)),
6261 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6262 def : Pat<(f64 (fceil FR64:$src)),
6263 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6264 def : Pat<(f32 (frint FR32:$src)),
6265 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6266 def : Pat<(f64 (frint FR64:$src)),
6267 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6268 def : Pat<(f32 (ftrunc FR32:$src)),
6269 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6270 def : Pat<(f64 (ftrunc FR64:$src)),
6271 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6274 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6275 memopv4f32, memopv2f64,
6276 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6277 let Constraints = "$src1 = $dst" in
6278 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6279 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6281 def : Pat<(ffloor FR32:$src),
6282 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6283 def : Pat<(f64 (ffloor FR64:$src)),
6284 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6285 def : Pat<(f32 (fnearbyint FR32:$src)),
6286 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6287 def : Pat<(f64 (fnearbyint FR64:$src)),
6288 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6289 def : Pat<(f32 (fceil FR32:$src)),
6290 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6291 def : Pat<(f64 (fceil FR64:$src)),
6292 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6293 def : Pat<(f32 (frint FR32:$src)),
6294 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6295 def : Pat<(f64 (frint FR64:$src)),
6296 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6297 def : Pat<(f32 (ftrunc FR32:$src)),
6298 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6299 def : Pat<(f64 (ftrunc FR64:$src)),
6300 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6302 //===----------------------------------------------------------------------===//
6303 // SSE4.1 - Packed Bit Test
6304 //===----------------------------------------------------------------------===//
6306 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6307 // the intel intrinsic that corresponds to this.
6308 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6309 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6310 "vptest\t{$src2, $src1|$src1, $src2}",
6311 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6313 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6314 "vptest\t{$src2, $src1|$src1, $src2}",
6315 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6318 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6319 "vptest\t{$src2, $src1|$src1, $src2}",
6320 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6322 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6323 "vptest\t{$src2, $src1|$src1, $src2}",
6324 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6328 let Defs = [EFLAGS] in {
6329 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6330 "ptest\t{$src2, $src1|$src1, $src2}",
6331 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6333 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6334 "ptest\t{$src2, $src1|$src1, $src2}",
6335 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6339 // The bit test instructions below are AVX only
6340 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6341 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6342 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6343 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6344 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6345 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6346 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6347 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6351 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6352 let ExeDomain = SSEPackedSingle in {
6353 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6354 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6356 let ExeDomain = SSEPackedDouble in {
6357 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6358 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6362 //===----------------------------------------------------------------------===//
6363 // SSE4.1 - Misc Instructions
6364 //===----------------------------------------------------------------------===//
6366 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6367 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6368 "popcnt{w}\t{$src, $dst|$dst, $src}",
6369 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6371 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6372 "popcnt{w}\t{$src, $dst|$dst, $src}",
6373 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6374 (implicit EFLAGS)]>, OpSize, XS;
6376 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6377 "popcnt{l}\t{$src, $dst|$dst, $src}",
6378 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6380 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6381 "popcnt{l}\t{$src, $dst|$dst, $src}",
6382 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6383 (implicit EFLAGS)]>, XS;
6385 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6386 "popcnt{q}\t{$src, $dst|$dst, $src}",
6387 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6389 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6390 "popcnt{q}\t{$src, $dst|$dst, $src}",
6391 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6392 (implicit EFLAGS)]>, XS;
6397 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6398 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6399 Intrinsic IntId128> {
6400 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6402 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6403 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6404 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6406 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6409 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6412 let Predicates = [HasAVX] in
6413 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6414 int_x86_sse41_phminposuw>, VEX;
6415 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6416 int_x86_sse41_phminposuw>;
6418 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6419 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6420 Intrinsic IntId128, bit Is2Addr = 1> {
6421 let isCommutable = 1 in
6422 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6423 (ins VR128:$src1, VR128:$src2),
6425 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6426 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6427 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6428 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6429 (ins VR128:$src1, i128mem:$src2),
6431 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6432 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6434 (IntId128 VR128:$src1,
6435 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6438 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6439 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6440 Intrinsic IntId256> {
6441 let isCommutable = 1 in
6442 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6443 (ins VR256:$src1, VR256:$src2),
6444 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6445 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6446 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6447 (ins VR256:$src1, i256mem:$src2),
6448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6450 (IntId256 VR256:$src1,
6451 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6454 let Predicates = [HasAVX] in {
6455 let isCommutable = 0 in
6456 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6458 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6460 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6462 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6464 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6466 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6468 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6470 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6472 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6474 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6478 let Predicates = [HasAVX2] in {
6479 let isCommutable = 0 in
6480 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6481 int_x86_avx2_packusdw>, VEX_4V;
6482 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6483 int_x86_avx2_pmins_b>, VEX_4V;
6484 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6485 int_x86_avx2_pmins_d>, VEX_4V;
6486 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6487 int_x86_avx2_pminu_d>, VEX_4V;
6488 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6489 int_x86_avx2_pminu_w>, VEX_4V;
6490 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6491 int_x86_avx2_pmaxs_b>, VEX_4V;
6492 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6493 int_x86_avx2_pmaxs_d>, VEX_4V;
6494 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6495 int_x86_avx2_pmaxu_d>, VEX_4V;
6496 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6497 int_x86_avx2_pmaxu_w>, VEX_4V;
6498 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6499 int_x86_avx2_pmul_dq>, VEX_4V;
6502 let Constraints = "$src1 = $dst" in {
6503 let isCommutable = 0 in
6504 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6505 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6506 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6507 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6508 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6509 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6510 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6511 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6512 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6513 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6516 /// SS48I_binop_rm - Simple SSE41 binary operator.
6517 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6518 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6519 X86MemOperand x86memop, bit Is2Addr = 1> {
6520 let isCommutable = 1 in
6521 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6522 (ins RC:$src1, RC:$src2),
6524 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6525 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6526 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6527 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6528 (ins RC:$src1, x86memop:$src2),
6530 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6531 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6533 (OpVT (OpNode RC:$src1,
6534 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6537 let Predicates = [HasAVX] in {
6538 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6539 memopv2i64, i128mem, 0>, VEX_4V;
6540 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6541 memopv2i64, i128mem, 0>, VEX_4V;
6543 let Predicates = [HasAVX2] in {
6544 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6545 memopv4i64, i256mem, 0>, VEX_4V;
6546 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6547 memopv4i64, i256mem, 0>, VEX_4V;
6550 let Constraints = "$src1 = $dst" in {
6551 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6552 memopv2i64, i128mem>;
6553 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6554 memopv2i64, i128mem>;
6557 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6558 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6559 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6560 X86MemOperand x86memop, bit Is2Addr = 1> {
6561 let isCommutable = 1 in
6562 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6563 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6565 !strconcat(OpcodeStr,
6566 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6567 !strconcat(OpcodeStr,
6568 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6569 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6571 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6572 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6574 !strconcat(OpcodeStr,
6575 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6576 !strconcat(OpcodeStr,
6577 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6580 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6584 let Predicates = [HasAVX] in {
6585 let isCommutable = 0 in {
6586 let ExeDomain = SSEPackedSingle in {
6587 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6588 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6589 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6590 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6592 let ExeDomain = SSEPackedDouble in {
6593 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6594 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6595 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6596 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6598 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6599 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6600 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6601 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6603 let ExeDomain = SSEPackedSingle in
6604 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6605 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6606 let ExeDomain = SSEPackedDouble in
6607 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6608 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6609 let ExeDomain = SSEPackedSingle in
6610 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6611 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6614 let Predicates = [HasAVX2] in {
6615 let isCommutable = 0 in {
6616 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6617 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6618 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6619 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6623 let Constraints = "$src1 = $dst" in {
6624 let isCommutable = 0 in {
6625 let ExeDomain = SSEPackedSingle in
6626 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6627 VR128, memopv4f32, i128mem>;
6628 let ExeDomain = SSEPackedDouble in
6629 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6630 VR128, memopv2f64, i128mem>;
6631 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6632 VR128, memopv2i64, i128mem>;
6633 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6634 VR128, memopv2i64, i128mem>;
6636 let ExeDomain = SSEPackedSingle in
6637 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6638 VR128, memopv4f32, i128mem>;
6639 let ExeDomain = SSEPackedDouble in
6640 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6641 VR128, memopv2f64, i128mem>;
6644 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6645 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6646 RegisterClass RC, X86MemOperand x86memop,
6647 PatFrag mem_frag, Intrinsic IntId> {
6648 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6649 (ins RC:$src1, RC:$src2, RC:$src3),
6650 !strconcat(OpcodeStr,
6651 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6652 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6653 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6655 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6656 (ins RC:$src1, x86memop:$src2, RC:$src3),
6657 !strconcat(OpcodeStr,
6658 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6660 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6662 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6665 let Predicates = [HasAVX] in {
6666 let ExeDomain = SSEPackedDouble in {
6667 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6668 memopv2f64, int_x86_sse41_blendvpd>;
6669 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6670 memopv4f64, int_x86_avx_blendv_pd_256>;
6671 } // ExeDomain = SSEPackedDouble
6672 let ExeDomain = SSEPackedSingle in {
6673 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6674 memopv4f32, int_x86_sse41_blendvps>;
6675 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6676 memopv8f32, int_x86_avx_blendv_ps_256>;
6677 } // ExeDomain = SSEPackedSingle
6678 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6679 memopv2i64, int_x86_sse41_pblendvb>;
6682 let Predicates = [HasAVX2] in {
6683 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6684 memopv4i64, int_x86_avx2_pblendvb>;
6687 let Predicates = [HasAVX] in {
6688 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6689 (v16i8 VR128:$src2))),
6690 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6691 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6692 (v4i32 VR128:$src2))),
6693 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6694 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6695 (v4f32 VR128:$src2))),
6696 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6697 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6698 (v2i64 VR128:$src2))),
6699 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6700 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6701 (v2f64 VR128:$src2))),
6702 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6703 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6704 (v8i32 VR256:$src2))),
6705 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6706 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6707 (v8f32 VR256:$src2))),
6708 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6709 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6710 (v4i64 VR256:$src2))),
6711 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6712 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6713 (v4f64 VR256:$src2))),
6714 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6716 def : Pat<(v8f32 (X86Blendps (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6718 (VBLENDPSYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6719 def : Pat<(v4f64 (X86Blendpd (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6721 (VBLENDPDYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6723 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6725 (VPBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6726 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6728 (VBLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6729 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6731 (VBLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6734 let Predicates = [HasAVX2] in {
6735 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6736 (v32i8 VR256:$src2))),
6737 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6738 def : Pat<(v16i16 (X86Blendpw (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6740 (VPBLENDWYrri VR256:$src2, VR256:$src1, imm:$mask)>;
6743 /// SS41I_ternary_int - SSE 4.1 ternary operator
6744 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6745 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6747 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6748 (ins VR128:$src1, VR128:$src2),
6749 !strconcat(OpcodeStr,
6750 "\t{$src2, $dst|$dst, $src2}"),
6751 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6754 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6755 (ins VR128:$src1, i128mem:$src2),
6756 !strconcat(OpcodeStr,
6757 "\t{$src2, $dst|$dst, $src2}"),
6760 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6764 let ExeDomain = SSEPackedDouble in
6765 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6766 int_x86_sse41_blendvpd>;
6767 let ExeDomain = SSEPackedSingle in
6768 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6769 int_x86_sse41_blendvps>;
6770 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6771 int_x86_sse41_pblendvb>;
6773 let Predicates = [HasSSE41] in {
6774 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6775 (v16i8 VR128:$src2))),
6776 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6777 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6778 (v4i32 VR128:$src2))),
6779 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6780 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6781 (v4f32 VR128:$src2))),
6782 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6783 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6784 (v2i64 VR128:$src2))),
6785 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6786 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6787 (v2f64 VR128:$src2))),
6788 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6790 def : Pat<(v8i16 (X86Blendpw (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6792 (PBLENDWrri VR128:$src2, VR128:$src1, imm:$mask)>;
6793 def : Pat<(v4f32 (X86Blendps (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6795 (BLENDPSrri VR128:$src2, VR128:$src1, imm:$mask)>;
6796 def : Pat<(v2f64 (X86Blendpd (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6798 (BLENDPDrri VR128:$src2, VR128:$src1, imm:$mask)>;
6802 let Predicates = [HasAVX] in
6803 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6804 "vmovntdqa\t{$src, $dst|$dst, $src}",
6805 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6807 let Predicates = [HasAVX2] in
6808 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6809 "vmovntdqa\t{$src, $dst|$dst, $src}",
6810 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6812 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6813 "movntdqa\t{$src, $dst|$dst, $src}",
6814 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6817 //===----------------------------------------------------------------------===//
6818 // SSE4.2 - Compare Instructions
6819 //===----------------------------------------------------------------------===//
6821 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6822 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6823 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6824 X86MemOperand x86memop, bit Is2Addr = 1> {
6825 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6826 (ins RC:$src1, RC:$src2),
6828 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6829 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6830 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6832 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6833 (ins RC:$src1, x86memop:$src2),
6835 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6836 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6838 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6841 let Predicates = [HasAVX] in
6842 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6843 memopv2i64, i128mem, 0>, VEX_4V;
6845 let Predicates = [HasAVX2] in
6846 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6847 memopv4i64, i256mem, 0>, VEX_4V;
6849 let Constraints = "$src1 = $dst" in
6850 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6851 memopv2i64, i128mem>;
6853 //===----------------------------------------------------------------------===//
6854 // SSE4.2 - String/text Processing Instructions
6855 //===----------------------------------------------------------------------===//
6857 // Packed Compare Implicit Length Strings, Return Mask
6858 multiclass pseudo_pcmpistrm<string asm> {
6859 def REG : PseudoI<(outs VR128:$dst),
6860 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6861 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6863 def MEM : PseudoI<(outs VR128:$dst),
6864 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6865 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6866 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6869 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6870 let AddedComplexity = 1 in
6871 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6872 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6875 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6876 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6877 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6878 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6880 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6881 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6882 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6885 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6886 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6887 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6888 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6890 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6891 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6892 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6895 // Packed Compare Explicit Length Strings, Return Mask
6896 multiclass pseudo_pcmpestrm<string asm> {
6897 def REG : PseudoI<(outs VR128:$dst),
6898 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6899 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6900 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6901 def MEM : PseudoI<(outs VR128:$dst),
6902 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6903 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6904 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6907 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6908 let AddedComplexity = 1 in
6909 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6910 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6913 let Predicates = [HasAVX],
6914 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6915 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6916 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6917 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6919 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6920 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6921 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6924 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6925 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6926 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6927 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6929 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6930 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6931 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6934 // Packed Compare Implicit Length Strings, Return Index
6935 let Defs = [ECX, EFLAGS] in {
6936 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6937 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6938 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6939 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6940 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6941 (implicit EFLAGS)]>, OpSize;
6942 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6943 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6944 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6945 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6946 (implicit EFLAGS)]>, OpSize;
6950 let Predicates = [HasAVX] in {
6951 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6953 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6955 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6957 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6959 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6961 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6965 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6966 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6967 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6968 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6969 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6970 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6972 // Packed Compare Explicit Length Strings, Return Index
6973 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6974 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6975 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6976 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6977 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6978 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6979 (implicit EFLAGS)]>, OpSize;
6980 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6981 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6982 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6984 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6985 (implicit EFLAGS)]>, OpSize;
6989 let Predicates = [HasAVX] in {
6990 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6992 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6994 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6996 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6998 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
7000 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
7004 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
7005 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
7006 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
7007 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
7008 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
7009 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
7011 //===----------------------------------------------------------------------===//
7012 // SSE4.2 - CRC Instructions
7013 //===----------------------------------------------------------------------===//
7015 // No CRC instructions have AVX equivalents
7017 // crc intrinsic instruction
7018 // This set of instructions are only rm, the only difference is the size
7020 let Constraints = "$src1 = $dst" in {
7021 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7022 (ins GR32:$src1, i8mem:$src2),
7023 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7025 (int_x86_sse42_crc32_32_8 GR32:$src1,
7026 (load addr:$src2)))]>;
7027 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7028 (ins GR32:$src1, GR8:$src2),
7029 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7031 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7032 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7033 (ins GR32:$src1, i16mem:$src2),
7034 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7036 (int_x86_sse42_crc32_32_16 GR32:$src1,
7037 (load addr:$src2)))]>,
7039 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7040 (ins GR32:$src1, GR16:$src2),
7041 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7043 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7045 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7046 (ins GR32:$src1, i32mem:$src2),
7047 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7049 (int_x86_sse42_crc32_32_32 GR32:$src1,
7050 (load addr:$src2)))]>;
7051 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7052 (ins GR32:$src1, GR32:$src2),
7053 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7055 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7056 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7057 (ins GR64:$src1, i8mem:$src2),
7058 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7060 (int_x86_sse42_crc32_64_8 GR64:$src1,
7061 (load addr:$src2)))]>,
7063 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7064 (ins GR64:$src1, GR8:$src2),
7065 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7067 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7069 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7070 (ins GR64:$src1, i64mem:$src2),
7071 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7073 (int_x86_sse42_crc32_64_64 GR64:$src1,
7074 (load addr:$src2)))]>,
7076 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7077 (ins GR64:$src1, GR64:$src2),
7078 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7080 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7084 //===----------------------------------------------------------------------===//
7085 // AES-NI Instructions
7086 //===----------------------------------------------------------------------===//
7088 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7089 Intrinsic IntId128, bit Is2Addr = 1> {
7090 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7091 (ins VR128:$src1, VR128:$src2),
7093 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7094 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7095 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7097 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7098 (ins VR128:$src1, i128mem:$src2),
7100 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7101 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7103 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7106 // Perform One Round of an AES Encryption/Decryption Flow
7107 let Predicates = [HasAVX, HasAES] in {
7108 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7109 int_x86_aesni_aesenc, 0>, VEX_4V;
7110 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7111 int_x86_aesni_aesenclast, 0>, VEX_4V;
7112 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7113 int_x86_aesni_aesdec, 0>, VEX_4V;
7114 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7115 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7118 let Constraints = "$src1 = $dst" in {
7119 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7120 int_x86_aesni_aesenc>;
7121 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7122 int_x86_aesni_aesenclast>;
7123 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7124 int_x86_aesni_aesdec>;
7125 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7126 int_x86_aesni_aesdeclast>;
7129 // Perform the AES InvMixColumn Transformation
7130 let Predicates = [HasAVX, HasAES] in {
7131 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7133 "vaesimc\t{$src1, $dst|$dst, $src1}",
7135 (int_x86_aesni_aesimc VR128:$src1))]>,
7137 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7138 (ins i128mem:$src1),
7139 "vaesimc\t{$src1, $dst|$dst, $src1}",
7140 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7143 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7145 "aesimc\t{$src1, $dst|$dst, $src1}",
7147 (int_x86_aesni_aesimc VR128:$src1))]>,
7149 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7150 (ins i128mem:$src1),
7151 "aesimc\t{$src1, $dst|$dst, $src1}",
7152 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7155 // AES Round Key Generation Assist
7156 let Predicates = [HasAVX, HasAES] in {
7157 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7158 (ins VR128:$src1, i8imm:$src2),
7159 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7161 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7163 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7164 (ins i128mem:$src1, i8imm:$src2),
7165 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7167 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7170 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7171 (ins VR128:$src1, i8imm:$src2),
7172 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7174 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7176 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7177 (ins i128mem:$src1, i8imm:$src2),
7178 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7180 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7183 //===----------------------------------------------------------------------===//
7184 // PCLMUL Instructions
7185 //===----------------------------------------------------------------------===//
7187 // AVX carry-less Multiplication instructions
7188 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7189 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7190 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7192 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7194 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7195 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7196 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7197 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7198 (memopv2i64 addr:$src2), imm:$src3))]>;
7200 // Carry-less Multiplication instructions
7201 let Constraints = "$src1 = $dst" in {
7202 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7203 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7204 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7206 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7208 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7209 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7210 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7211 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7212 (memopv2i64 addr:$src2), imm:$src3))]>;
7213 } // Constraints = "$src1 = $dst"
7216 multiclass pclmul_alias<string asm, int immop> {
7217 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7218 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7220 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7221 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7223 def : InstAlias<!strconcat("vpclmul", asm,
7224 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7225 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7227 def : InstAlias<!strconcat("vpclmul", asm,
7228 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7229 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7231 defm : pclmul_alias<"hqhq", 0x11>;
7232 defm : pclmul_alias<"hqlq", 0x01>;
7233 defm : pclmul_alias<"lqhq", 0x10>;
7234 defm : pclmul_alias<"lqlq", 0x00>;
7236 //===----------------------------------------------------------------------===//
7237 // SSE4A Instructions
7238 //===----------------------------------------------------------------------===//
7240 let Predicates = [HasSSE4A] in {
7242 let Constraints = "$src = $dst" in {
7243 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7244 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7245 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7246 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7247 imm:$idx))]>, TB, OpSize;
7248 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7249 (ins VR128:$src, VR128:$mask),
7250 "extrq\t{$mask, $src|$src, $mask}",
7251 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7252 VR128:$mask))]>, TB, OpSize;
7254 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7255 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7256 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7257 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7258 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7259 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7260 (ins VR128:$src, VR128:$mask),
7261 "insertq\t{$mask, $src|$src, $mask}",
7262 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7263 VR128:$mask))]>, XD;
7266 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7267 "movntss\t{$src, $dst|$dst, $src}",
7268 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7270 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7271 "movntsd\t{$src, $dst|$dst, $src}",
7272 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7275 //===----------------------------------------------------------------------===//
7277 //===----------------------------------------------------------------------===//
7279 //===----------------------------------------------------------------------===//
7280 // VBROADCAST - Load from memory and broadcast to all elements of the
7281 // destination operand
7283 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7284 X86MemOperand x86memop, Intrinsic Int> :
7285 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7286 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7287 [(set RC:$dst, (Int addr:$src))]>, VEX;
7289 // AVX2 adds register forms
7290 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7292 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7293 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7294 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7296 let ExeDomain = SSEPackedSingle in {
7297 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7298 int_x86_avx_vbroadcast_ss>;
7299 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7300 int_x86_avx_vbroadcast_ss_256>;
7302 let ExeDomain = SSEPackedDouble in
7303 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7304 int_x86_avx_vbroadcast_sd_256>;
7305 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7306 int_x86_avx_vbroadcastf128_pd_256>;
7308 let ExeDomain = SSEPackedSingle in {
7309 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7310 int_x86_avx2_vbroadcast_ss_ps>;
7311 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7312 int_x86_avx2_vbroadcast_ss_ps_256>;
7314 let ExeDomain = SSEPackedDouble in
7315 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7316 int_x86_avx2_vbroadcast_sd_pd_256>;
7318 let Predicates = [HasAVX2] in
7319 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7320 int_x86_avx2_vbroadcasti128>;
7322 let Predicates = [HasAVX] in
7323 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7324 (VBROADCASTF128 addr:$src)>;
7327 //===----------------------------------------------------------------------===//
7328 // VINSERTF128 - Insert packed floating-point values
7330 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7331 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7332 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7333 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7336 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7337 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7338 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7342 let Predicates = [HasAVX] in {
7343 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7345 (VINSERTF128rr VR256:$src1, VR128:$src2,
7346 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7347 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7349 (VINSERTF128rr VR256:$src1, VR128:$src2,
7350 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7351 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7353 (VINSERTF128rr VR256:$src1, VR128:$src2,
7354 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7355 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7357 (VINSERTF128rr VR256:$src1, VR128:$src2,
7358 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7359 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7361 (VINSERTF128rr VR256:$src1, VR128:$src2,
7362 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7363 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7365 (VINSERTF128rr VR256:$src1, VR128:$src2,
7366 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7368 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (loadv4f32 addr:$src2),
7370 (VINSERTF128rm VR256:$src1, addr:$src2,
7371 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7372 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (loadv2f64 addr:$src2),
7374 (VINSERTF128rm VR256:$src1, addr:$src2,
7375 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7376 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (loadv2i64 addr:$src2),
7378 (VINSERTF128rm VR256:$src1, addr:$src2,
7379 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7382 //===----------------------------------------------------------------------===//
7383 // VEXTRACTF128 - Extract packed floating-point values
7385 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7386 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7387 (ins VR256:$src1, i8imm:$src2),
7388 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7391 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7392 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7393 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7397 // Extract and store.
7398 let Predicates = [HasAVX] in {
7399 def : Pat<(alignedstore (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2), addr:$dst),
7400 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7401 def : Pat<(alignedstore (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2), addr:$dst),
7402 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7403 def : Pat<(alignedstore (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2), addr:$dst),
7404 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7406 def : Pat<(int_x86_sse_storeu_ps addr:$dst, (int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2)),
7407 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7408 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, (int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2)),
7409 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7410 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, (bc_v16i8 (int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2))),
7411 (VEXTRACTF128mr addr:$dst, VR256:$src1, imm:$src2)>;
7415 let Predicates = [HasAVX] in {
7416 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7417 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7418 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7419 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7420 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7421 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7423 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7424 (v4f32 (VEXTRACTF128rr
7425 (v8f32 VR256:$src1),
7426 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7427 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7428 (v2f64 (VEXTRACTF128rr
7429 (v4f64 VR256:$src1),
7430 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7431 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7432 (v2i64 (VEXTRACTF128rr
7433 (v4i64 VR256:$src1),
7434 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7435 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7436 (v4i32 (VEXTRACTF128rr
7437 (v8i32 VR256:$src1),
7438 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7439 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7440 (v8i16 (VEXTRACTF128rr
7441 (v16i16 VR256:$src1),
7442 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7443 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7444 (v16i8 (VEXTRACTF128rr
7445 (v32i8 VR256:$src1),
7446 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7449 //===----------------------------------------------------------------------===//
7450 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7452 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7453 Intrinsic IntLd, Intrinsic IntLd256,
7454 Intrinsic IntSt, Intrinsic IntSt256> {
7455 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7456 (ins VR128:$src1, f128mem:$src2),
7457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7458 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7460 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7461 (ins VR256:$src1, f256mem:$src2),
7462 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7463 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7465 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7466 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7467 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7468 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7469 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7470 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7471 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7472 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7475 let ExeDomain = SSEPackedSingle in
7476 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7477 int_x86_avx_maskload_ps,
7478 int_x86_avx_maskload_ps_256,
7479 int_x86_avx_maskstore_ps,
7480 int_x86_avx_maskstore_ps_256>;
7481 let ExeDomain = SSEPackedDouble in
7482 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7483 int_x86_avx_maskload_pd,
7484 int_x86_avx_maskload_pd_256,
7485 int_x86_avx_maskstore_pd,
7486 int_x86_avx_maskstore_pd_256>;
7488 //===----------------------------------------------------------------------===//
7489 // VPERMIL - Permute Single and Double Floating-Point Values
7491 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7492 RegisterClass RC, X86MemOperand x86memop_f,
7493 X86MemOperand x86memop_i, PatFrag i_frag,
7494 Intrinsic IntVar, ValueType vt> {
7495 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7496 (ins RC:$src1, RC:$src2),
7497 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7498 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7499 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7500 (ins RC:$src1, x86memop_i:$src2),
7501 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7502 [(set RC:$dst, (IntVar RC:$src1,
7503 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7505 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7506 (ins RC:$src1, i8imm:$src2),
7507 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7508 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7509 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7510 (ins x86memop_f:$src1, i8imm:$src2),
7511 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7513 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7516 let ExeDomain = SSEPackedSingle in {
7517 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7518 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7519 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7520 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>;
7522 let ExeDomain = SSEPackedDouble in {
7523 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7524 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7525 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7526 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>;
7529 let Predicates = [HasAVX] in {
7530 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7531 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7532 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7533 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7534 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7536 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7537 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7538 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7540 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7541 (VPERMILPDri VR128:$src1, imm:$imm)>;
7542 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7543 (VPERMILPDmi addr:$src1, imm:$imm)>;
7546 //===----------------------------------------------------------------------===//
7547 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7549 let ExeDomain = SSEPackedSingle in {
7550 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7551 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7552 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7553 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7554 (i8 imm:$src3))))]>, VEX_4V;
7555 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7556 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7557 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7558 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7559 (i8 imm:$src3)))]>, VEX_4V;
7562 let Predicates = [HasAVX] in {
7563 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7564 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7565 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7566 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7567 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7568 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7569 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7570 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7571 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7572 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7574 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7575 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7576 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7577 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7578 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7579 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7580 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7581 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7582 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7583 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7584 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7585 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7586 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7587 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7588 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7589 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7590 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7591 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7594 //===----------------------------------------------------------------------===//
7595 // VZERO - Zero YMM registers
7597 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7598 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7599 // Zero All YMM registers
7600 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7601 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7603 // Zero Upper bits of YMM registers
7604 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7605 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7608 //===----------------------------------------------------------------------===//
7609 // Half precision conversion instructions
7610 //===----------------------------------------------------------------------===//
7611 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7612 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7613 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7614 [(set RC:$dst, (Int VR128:$src))]>,
7616 let neverHasSideEffects = 1, mayLoad = 1 in
7617 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7618 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7621 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7622 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7623 (ins RC:$src1, i32i8imm:$src2),
7624 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7625 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7627 let neverHasSideEffects = 1, mayStore = 1 in
7628 def mr : Ii8<0x1D, MRMDestMem, (outs),
7629 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7630 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7634 let Predicates = [HasAVX, HasF16C] in {
7635 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7636 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7637 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7638 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7641 //===----------------------------------------------------------------------===//
7642 // AVX2 Instructions
7643 //===----------------------------------------------------------------------===//
7645 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7646 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7647 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7648 X86MemOperand x86memop> {
7649 let isCommutable = 1 in
7650 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7651 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7652 !strconcat(OpcodeStr,
7653 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7654 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7656 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7657 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7658 !strconcat(OpcodeStr,
7659 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7662 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7666 let isCommutable = 0 in {
7667 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7668 VR128, memopv2i64, i128mem>;
7669 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7670 VR256, memopv4i64, i256mem>;
7673 //===----------------------------------------------------------------------===//
7674 // VPBROADCAST - Load from memory and broadcast to all elements of the
7675 // destination operand
7677 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7678 X86MemOperand x86memop, PatFrag ld_frag,
7679 Intrinsic Int128, Intrinsic Int256> {
7680 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7681 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7682 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7683 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7684 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7686 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7687 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7688 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7689 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7690 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7691 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7693 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7696 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7697 int_x86_avx2_pbroadcastb_128,
7698 int_x86_avx2_pbroadcastb_256>;
7699 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7700 int_x86_avx2_pbroadcastw_128,
7701 int_x86_avx2_pbroadcastw_256>;
7702 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7703 int_x86_avx2_pbroadcastd_128,
7704 int_x86_avx2_pbroadcastd_256>;
7705 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7706 int_x86_avx2_pbroadcastq_128,
7707 int_x86_avx2_pbroadcastq_256>;
7709 let Predicates = [HasAVX2] in {
7710 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7711 (VPBROADCASTBrm addr:$src)>;
7712 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7713 (VPBROADCASTBYrm addr:$src)>;
7714 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7715 (VPBROADCASTWrm addr:$src)>;
7716 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7717 (VPBROADCASTWYrm addr:$src)>;
7718 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7719 (VPBROADCASTDrm addr:$src)>;
7720 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7721 (VPBROADCASTDYrm addr:$src)>;
7722 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7723 (VPBROADCASTQrm addr:$src)>;
7724 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7725 (VPBROADCASTQYrm addr:$src)>;
7727 // Provide fallback in case the load node that is used in the patterns above
7728 // is used by additional users, which prevents the pattern selection.
7729 let AddedComplexity = 20 in {
7730 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7732 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7733 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7735 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss))>;
7736 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7738 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd))>;
7740 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7742 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7743 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7745 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss))>;
7746 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7748 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd))>;
7752 // AVX1 broadcast patterns
7753 let Predicates = [HasAVX] in {
7754 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7755 (VBROADCASTSSYrm addr:$src)>;
7756 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7757 (VBROADCASTSDrm addr:$src)>;
7758 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7759 (VBROADCASTSSYrm addr:$src)>;
7760 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7761 (VBROADCASTSDrm addr:$src)>;
7762 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7763 (VBROADCASTSSrm addr:$src)>;
7764 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7765 (VBROADCASTSSrm addr:$src)>;
7767 // Provide fallback in case the load node that is used in the patterns above
7768 // is used by additional users, which prevents the pattern selection.
7769 let AddedComplexity = 20 in {
7770 // 128bit broadcasts:
7771 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7773 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0)>;
7774 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7775 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7777 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss), 0),
7780 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss),
7782 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7783 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7785 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd), 0),
7788 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd),
7791 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7793 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0)>;
7794 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7795 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7797 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss), 0),
7800 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)), GR32:$src, sub_ss),
7802 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7803 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7805 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd), 0),
7808 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)), GR64:$src, sub_sd),
7813 //===----------------------------------------------------------------------===//
7814 // VPERM - Permute instructions
7817 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7819 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7820 (ins VR256:$src1, VR256:$src2),
7821 !strconcat(OpcodeStr,
7822 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7824 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>, VEX_4V;
7825 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7826 (ins VR256:$src1, i256mem:$src2),
7827 !strconcat(OpcodeStr,
7828 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7830 (OpVT (X86VPermv VR256:$src1,
7831 (bitconvert (mem_frag addr:$src2)))))]>,
7835 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7836 let ExeDomain = SSEPackedSingle in
7837 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7839 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7841 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7842 (ins VR256:$src1, i8imm:$src2),
7843 !strconcat(OpcodeStr,
7844 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7846 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>, VEX;
7847 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7848 (ins i256mem:$src1, i8imm:$src2),
7849 !strconcat(OpcodeStr,
7850 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7852 (OpVT (X86VPermi (mem_frag addr:$src1),
7853 (i8 imm:$src2))))]>, VEX;
7856 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7857 let ExeDomain = SSEPackedDouble in
7858 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7860 //===----------------------------------------------------------------------===//
7861 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7863 let AddedComplexity = 1 in {
7864 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7865 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7866 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7867 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7868 (i8 imm:$src3))))]>, VEX_4V;
7869 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7870 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7871 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7872 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7873 (i8 imm:$src3)))]>, VEX_4V;
7876 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7877 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7878 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7879 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7880 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7881 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7882 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7884 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7886 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7887 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7888 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7889 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7890 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7892 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7896 //===----------------------------------------------------------------------===//
7897 // VINSERTI128 - Insert packed integer values
7899 let neverHasSideEffects = 1 in {
7900 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7901 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7902 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7905 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7906 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7907 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7911 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7912 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7914 (VINSERTI128rr VR256:$src1, VR128:$src2,
7915 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7916 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7918 (VINSERTI128rr VR256:$src1, VR128:$src2,
7919 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7920 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7922 (VINSERTI128rr VR256:$src1, VR128:$src2,
7923 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7924 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7926 (VINSERTI128rr VR256:$src1, VR128:$src2,
7927 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7930 //===----------------------------------------------------------------------===//
7931 // VEXTRACTI128 - Extract packed integer values
7933 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7934 (ins VR256:$src1, i8imm:$src2),
7935 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7937 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7939 let neverHasSideEffects = 1, mayStore = 1 in
7940 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7941 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7942 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7944 let Predicates = [HasAVX2], AddedComplexity = 1 in {
7945 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7946 (v2i64 (VEXTRACTI128rr
7947 (v4i64 VR256:$src1),
7948 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7949 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7950 (v4i32 (VEXTRACTI128rr
7951 (v8i32 VR256:$src1),
7952 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7953 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7954 (v8i16 (VEXTRACTI128rr
7955 (v16i16 VR256:$src1),
7956 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7957 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7958 (v16i8 (VEXTRACTI128rr
7959 (v32i8 VR256:$src1),
7960 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7963 //===----------------------------------------------------------------------===//
7964 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7966 multiclass avx2_pmovmask<string OpcodeStr,
7967 Intrinsic IntLd128, Intrinsic IntLd256,
7968 Intrinsic IntSt128, Intrinsic IntSt256> {
7969 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7970 (ins VR128:$src1, i128mem:$src2),
7971 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7972 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7973 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7974 (ins VR256:$src1, i256mem:$src2),
7975 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7976 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7977 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7978 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7980 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7981 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7982 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7983 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7984 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7987 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7988 int_x86_avx2_maskload_d,
7989 int_x86_avx2_maskload_d_256,
7990 int_x86_avx2_maskstore_d,
7991 int_x86_avx2_maskstore_d_256>;
7992 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7993 int_x86_avx2_maskload_q,
7994 int_x86_avx2_maskload_q_256,
7995 int_x86_avx2_maskstore_q,
7996 int_x86_avx2_maskstore_q_256>, VEX_W;
7999 //===----------------------------------------------------------------------===//
8000 // Variable Bit Shifts
8002 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8003 ValueType vt128, ValueType vt256> {
8004 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8005 (ins VR128:$src1, VR128:$src2),
8006 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8008 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8010 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8011 (ins VR128:$src1, i128mem:$src2),
8012 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8014 (vt128 (OpNode VR128:$src1,
8015 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8017 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8018 (ins VR256:$src1, VR256:$src2),
8019 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8021 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8023 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8024 (ins VR256:$src1, i256mem:$src2),
8025 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8027 (vt256 (OpNode VR256:$src1,
8028 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8032 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8033 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8034 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8035 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8036 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;