1 //===-- X86InstrSSE.td - SSE Instruction Set ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
16 class OpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm> {
17 InstrItinClass rr = arg_rr;
18 InstrItinClass rm = arg_rm;
21 class SizeItins<OpndItins arg_s, OpndItins arg_d> {
27 class ShiftOpndItins<InstrItinClass arg_rr, InstrItinClass arg_rm,
28 InstrItinClass arg_ri> {
29 InstrItinClass rr = arg_rr;
30 InstrItinClass rm = arg_rm;
31 InstrItinClass ri = arg_ri;
36 def SSE_ALU_F32S : OpndItins<
37 IIC_SSE_ALU_F32S_RR, IIC_SSE_ALU_F32S_RM
40 def SSE_ALU_F64S : OpndItins<
41 IIC_SSE_ALU_F64S_RR, IIC_SSE_ALU_F64S_RM
44 def SSE_ALU_ITINS_S : SizeItins<
45 SSE_ALU_F32S, SSE_ALU_F64S
48 def SSE_MUL_F32S : OpndItins<
49 IIC_SSE_MUL_F32S_RR, IIC_SSE_MUL_F64S_RM
52 def SSE_MUL_F64S : OpndItins<
53 IIC_SSE_MUL_F64S_RR, IIC_SSE_MUL_F64S_RM
56 def SSE_MUL_ITINS_S : SizeItins<
57 SSE_MUL_F32S, SSE_MUL_F64S
60 def SSE_DIV_F32S : OpndItins<
61 IIC_SSE_DIV_F32S_RR, IIC_SSE_DIV_F64S_RM
64 def SSE_DIV_F64S : OpndItins<
65 IIC_SSE_DIV_F64S_RR, IIC_SSE_DIV_F64S_RM
68 def SSE_DIV_ITINS_S : SizeItins<
69 SSE_DIV_F32S, SSE_DIV_F64S
73 def SSE_ALU_F32P : OpndItins<
74 IIC_SSE_ALU_F32P_RR, IIC_SSE_ALU_F32P_RM
77 def SSE_ALU_F64P : OpndItins<
78 IIC_SSE_ALU_F64P_RR, IIC_SSE_ALU_F64P_RM
81 def SSE_ALU_ITINS_P : SizeItins<
82 SSE_ALU_F32P, SSE_ALU_F64P
85 def SSE_MUL_F32P : OpndItins<
86 IIC_SSE_MUL_F32P_RR, IIC_SSE_MUL_F64P_RM
89 def SSE_MUL_F64P : OpndItins<
90 IIC_SSE_MUL_F64P_RR, IIC_SSE_MUL_F64P_RM
93 def SSE_MUL_ITINS_P : SizeItins<
94 SSE_MUL_F32P, SSE_MUL_F64P
97 def SSE_DIV_F32P : OpndItins<
98 IIC_SSE_DIV_F32P_RR, IIC_SSE_DIV_F64P_RM
101 def SSE_DIV_F64P : OpndItins<
102 IIC_SSE_DIV_F64P_RR, IIC_SSE_DIV_F64P_RM
105 def SSE_DIV_ITINS_P : SizeItins<
106 SSE_DIV_F32P, SSE_DIV_F64P
109 def SSE_BIT_ITINS_P : OpndItins<
110 IIC_SSE_BIT_P_RR, IIC_SSE_BIT_P_RM
113 def SSE_INTALU_ITINS_P : OpndItins<
114 IIC_SSE_INTALU_P_RR, IIC_SSE_INTALU_P_RM
117 def SSE_INTALUQ_ITINS_P : OpndItins<
118 IIC_SSE_INTALUQ_P_RR, IIC_SSE_INTALUQ_P_RM
121 def SSE_INTMUL_ITINS_P : OpndItins<
122 IIC_SSE_INTMUL_P_RR, IIC_SSE_INTMUL_P_RM
125 def SSE_INTSHIFT_ITINS_P : ShiftOpndItins<
126 IIC_SSE_INTSH_P_RR, IIC_SSE_INTSH_P_RM, IIC_SSE_INTSH_P_RI
129 def SSE_MOVA_ITINS : OpndItins<
130 IIC_SSE_MOVA_P_RR, IIC_SSE_MOVA_P_RM
133 def SSE_MOVU_ITINS : OpndItins<
134 IIC_SSE_MOVU_P_RR, IIC_SSE_MOVU_P_RM
137 //===----------------------------------------------------------------------===//
138 // SSE 1 & 2 Instructions Classes
139 //===----------------------------------------------------------------------===//
141 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
142 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
143 RegisterClass RC, X86MemOperand x86memop,
146 let isCommutable = 1 in {
147 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
149 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
150 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
151 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))], itins.rr>;
153 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
155 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
156 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
157 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))], itins.rm>;
160 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
161 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
162 string asm, string SSEVer, string FPSizeStr,
163 Operand memopr, ComplexPattern mem_cpat,
166 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
168 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
169 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
170 [(set RC:$dst, (!cast<Intrinsic>(
171 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
172 RC:$src1, RC:$src2))], itins.rr>;
173 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
175 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
176 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
177 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
178 SSEVer, "_", OpcodeStr, FPSizeStr))
179 RC:$src1, mem_cpat:$src2))], itins.rm>;
182 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
183 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
184 RegisterClass RC, ValueType vt,
185 X86MemOperand x86memop, PatFrag mem_frag,
186 Domain d, OpndItins itins, bit Is2Addr = 1> {
187 let isCommutable = 1 in
188 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
190 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
191 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
192 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], itins.rr, d>;
194 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
196 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
197 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
198 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))],
202 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
203 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
204 string OpcodeStr, X86MemOperand x86memop,
205 list<dag> pat_rr, list<dag> pat_rm,
207 bit rr_hasSideEffects = 0> {
208 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
209 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
211 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
212 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
213 pat_rr, IIC_DEFAULT, d>;
214 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
218 pat_rm, IIC_DEFAULT, d>;
221 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
222 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
223 string asm, string SSEVer, string FPSizeStr,
224 X86MemOperand x86memop, PatFrag mem_frag,
225 Domain d, OpndItins itins, bit Is2Addr = 1> {
226 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
228 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
229 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
230 [(set RC:$dst, (!cast<Intrinsic>(
231 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
232 RC:$src1, RC:$src2))], IIC_DEFAULT, d>;
233 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
235 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
236 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
237 [(set RC:$dst, (!cast<Intrinsic>(
238 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
239 RC:$src1, (mem_frag addr:$src2)))], IIC_DEFAULT, d>;
242 //===----------------------------------------------------------------------===//
243 // Non-instruction patterns
244 //===----------------------------------------------------------------------===//
246 // A vector extract of the first f32/f64 position is a subregister copy
247 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
248 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
249 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
250 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
252 // A 128-bit subvector extract from the first 256-bit vector position
253 // is a subregister copy that needs no instruction.
254 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
255 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
256 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
257 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
259 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
260 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
261 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
262 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
264 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
265 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
266 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
267 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$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 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
272 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
273 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
274 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
275 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
276 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
277 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
278 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
279 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
280 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
281 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
282 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
283 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
286 // Implicitly promote a 32-bit scalar to a vector.
287 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
288 (COPY_TO_REGCLASS FR32:$src, VR128)>;
289 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
290 (COPY_TO_REGCLASS FR32:$src, VR128)>;
291 // Implicitly promote a 64-bit scalar to a vector.
292 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
293 (COPY_TO_REGCLASS FR64:$src, VR128)>;
294 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
295 (COPY_TO_REGCLASS FR64:$src, VR128)>;
297 // Bitcasts between 128-bit vector types. Return the original type since
298 // no instruction is needed for the conversion
299 let Predicates = [HasSSE2] in {
300 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
301 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
302 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
303 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
304 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
305 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
306 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
307 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
308 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
309 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
310 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
311 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
312 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
313 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
314 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
315 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
316 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
317 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
318 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
319 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
320 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
321 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
322 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
323 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
324 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
325 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
326 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
327 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
328 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
329 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
332 // Bitcasts between 256-bit vector types. Return the original type since
333 // no instruction is needed for the conversion
334 let Predicates = [HasAVX] in {
335 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
336 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
337 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
338 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
339 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
340 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
341 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
342 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
343 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
344 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
345 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
346 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
347 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
348 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
349 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
350 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
351 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
352 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
353 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
354 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
355 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
356 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
357 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
358 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
359 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
360 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
361 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
362 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
363 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
364 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
367 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
368 // This is expanded by ExpandPostRAPseudos.
369 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
371 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
372 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
373 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
374 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
377 //===----------------------------------------------------------------------===//
378 // AVX & SSE - Zero/One Vectors
379 //===----------------------------------------------------------------------===//
381 // Alias instruction that maps zero vector to pxor / xorp* for sse.
382 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
383 // swizzled by ExecutionDepsFix to pxor.
384 // We set canFoldAsLoad because this can be converted to a constant-pool
385 // load of an all-zeros value if folding it would be beneficial.
386 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
388 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
389 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
392 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
393 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
394 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
395 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
396 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
399 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
400 // and doesn't need it because on sandy bridge the register is set to zero
401 // at the rename stage without using any execution unit, so SET0PSY
402 // and SET0PDY can be used for vector int instructions without penalty
403 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
404 isPseudo = 1, Predicates = [HasAVX] in {
405 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
406 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
409 let Predicates = [HasAVX] in
410 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
412 let Predicates = [HasAVX2] in {
413 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
414 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
415 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
416 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
419 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
420 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
421 let Predicates = [HasAVX1Only] in {
422 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
423 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
424 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
426 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
427 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
428 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
430 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
431 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
432 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
434 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
435 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
436 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
439 // We set canFoldAsLoad because this can be converted to a constant-pool
440 // load of an all-ones value if folding it would be beneficial.
441 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
443 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
444 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
445 let Predicates = [HasAVX2] in
446 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
447 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
451 //===----------------------------------------------------------------------===//
452 // SSE 1 & 2 - Move FP Scalar Instructions
454 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
455 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
456 // is used instead. Register-to-register movss/movsd is not modeled as an
457 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
458 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
459 //===----------------------------------------------------------------------===//
461 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
462 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
463 [(set VR128:$dst, (vt (OpNode VR128:$src1,
464 (scalar_to_vector RC:$src2))))],
467 // Loading from memory automatically zeroing upper bits.
468 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
469 PatFrag mem_pat, string OpcodeStr> :
470 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
471 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
472 [(set RC:$dst, (mem_pat addr:$src))],
476 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
477 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
479 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
480 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
483 // For the disassembler
484 let isCodeGenOnly = 1 in {
485 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
486 (ins VR128:$src1, FR32:$src2),
487 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
490 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
491 (ins VR128:$src1, FR64:$src2),
492 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
497 let canFoldAsLoad = 1, isReMaterializable = 1 in {
498 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
500 let AddedComplexity = 20 in
501 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
505 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
506 "movss\t{$src, $dst|$dst, $src}",
507 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
509 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
510 "movsd\t{$src, $dst|$dst, $src}",
511 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
515 let Constraints = "$src1 = $dst" in {
516 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
517 "movss\t{$src2, $dst|$dst, $src2}">, XS;
518 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
519 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
521 // For the disassembler
522 let isCodeGenOnly = 1 in {
523 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
524 (ins VR128:$src1, FR32:$src2),
525 "movss\t{$src2, $dst|$dst, $src2}", [],
526 IIC_SSE_MOV_S_RR>, XS;
527 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
528 (ins VR128:$src1, FR64:$src2),
529 "movsd\t{$src2, $dst|$dst, $src2}", [],
530 IIC_SSE_MOV_S_RR>, XD;
534 let canFoldAsLoad = 1, isReMaterializable = 1 in {
535 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
537 let AddedComplexity = 20 in
538 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
541 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
542 "movss\t{$src, $dst|$dst, $src}",
543 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
544 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
545 "movsd\t{$src, $dst|$dst, $src}",
546 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
549 let Predicates = [HasAVX] in {
550 let AddedComplexity = 15 in {
551 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
552 // MOVS{S,D} to the lower bits.
553 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
554 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
555 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
556 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
557 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
558 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
559 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
560 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
562 // Move low f32 and clear high bits.
563 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
564 (SUBREG_TO_REG (i32 0),
565 (VMOVSSrr (v4f32 (V_SET0)),
566 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
567 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
568 (SUBREG_TO_REG (i32 0),
569 (VMOVSSrr (v4i32 (V_SET0)),
570 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
573 let AddedComplexity = 20 in {
574 // MOVSSrm zeros the high parts of the register; represent this
575 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
576 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
577 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
578 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
579 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
580 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
581 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
583 // MOVSDrm zeros the high parts of the register; represent this
584 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
585 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
586 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
587 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
588 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
589 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
590 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
591 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
592 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
593 def : Pat<(v2f64 (X86vzload addr:$src)),
594 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
596 // Represent the same patterns above but in the form they appear for
598 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
599 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
600 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
601 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
602 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
603 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
604 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
605 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
606 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
608 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
609 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
610 (SUBREG_TO_REG (i32 0),
611 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
613 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
614 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
615 (SUBREG_TO_REG (i64 0),
616 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
618 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
619 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
620 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
622 // Move low f64 and clear high bits.
623 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
624 (SUBREG_TO_REG (i32 0),
625 (VMOVSDrr (v2f64 (V_SET0)),
626 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
628 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
629 (SUBREG_TO_REG (i32 0),
630 (VMOVSDrr (v2i64 (V_SET0)),
631 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
633 // Extract and store.
634 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
636 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
637 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
639 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
641 // Shuffle with VMOVSS
642 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
643 (VMOVSSrr (v4i32 VR128:$src1),
644 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
645 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
646 (VMOVSSrr (v4f32 VR128:$src1),
647 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
650 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
651 (SUBREG_TO_REG (i32 0),
652 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
653 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
655 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
656 (SUBREG_TO_REG (i32 0),
657 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
658 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
661 // Shuffle with VMOVSD
662 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
663 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
664 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
665 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
666 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
667 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
668 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
669 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
673 (SUBREG_TO_REG (i32 0),
674 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
675 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
677 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
678 (SUBREG_TO_REG (i32 0),
679 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
680 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
684 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
685 // is during lowering, where it's not possible to recognize the fold cause
686 // it has two uses through a bitcast. One use disappears at isel time and the
687 // fold opportunity reappears.
688 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
689 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
690 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
691 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
692 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
693 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
694 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
695 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
698 let Predicates = [UseSSE1] in {
699 let AddedComplexity = 15 in {
700 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
701 // MOVSS to the lower bits.
702 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
703 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
704 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
705 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
706 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
707 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
710 let AddedComplexity = 20 in {
711 // MOVSSrm already zeros the high parts of the register.
712 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
713 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
714 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
715 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
716 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
717 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
720 // Extract and store.
721 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
723 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
725 // Shuffle with MOVSS
726 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
727 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
728 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
729 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
732 let Predicates = [UseSSE2] in {
733 let AddedComplexity = 15 in {
734 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
735 // MOVSD to the lower bits.
736 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
737 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
740 let AddedComplexity = 20 in {
741 // MOVSDrm already zeros the high parts of the register.
742 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
743 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
744 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
745 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
746 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
747 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
748 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
749 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
750 def : Pat<(v2f64 (X86vzload addr:$src)),
751 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
754 // Extract and store.
755 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
757 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
759 // Shuffle with MOVSD
760 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
761 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
762 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
763 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
764 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
765 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
766 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
767 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
769 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
770 // is during lowering, where it's not possible to recognize the fold cause
771 // it has two uses through a bitcast. One use disappears at isel time and the
772 // fold opportunity reappears.
773 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
774 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
775 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
776 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
777 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
778 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
779 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
780 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
783 //===----------------------------------------------------------------------===//
784 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
785 //===----------------------------------------------------------------------===//
787 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
788 X86MemOperand x86memop, PatFrag ld_frag,
789 string asm, Domain d,
791 bit IsReMaterializable = 1> {
792 let neverHasSideEffects = 1 in
793 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
794 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
795 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
796 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
797 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
798 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
801 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
802 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
804 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
805 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
807 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
808 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
810 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
811 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
814 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
815 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
817 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
818 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
819 TB, OpSize, VEX, VEX_L;
820 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
821 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
823 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
824 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
825 TB, OpSize, VEX, VEX_L;
826 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
827 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
829 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
830 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
832 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
833 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
835 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
836 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
839 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
840 "movaps\t{$src, $dst|$dst, $src}",
841 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
842 IIC_SSE_MOVA_P_MR>, VEX;
843 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
844 "movapd\t{$src, $dst|$dst, $src}",
845 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
846 IIC_SSE_MOVA_P_MR>, VEX;
847 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
848 "movups\t{$src, $dst|$dst, $src}",
849 [(store (v4f32 VR128:$src), addr:$dst)],
850 IIC_SSE_MOVU_P_MR>, VEX;
851 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
852 "movupd\t{$src, $dst|$dst, $src}",
853 [(store (v2f64 VR128:$src), addr:$dst)],
854 IIC_SSE_MOVU_P_MR>, VEX;
855 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
856 "movaps\t{$src, $dst|$dst, $src}",
857 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
858 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
859 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
860 "movapd\t{$src, $dst|$dst, $src}",
861 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
862 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
863 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
864 "movups\t{$src, $dst|$dst, $src}",
865 [(store (v8f32 VR256:$src), addr:$dst)],
866 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
867 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
868 "movupd\t{$src, $dst|$dst, $src}",
869 [(store (v4f64 VR256:$src), addr:$dst)],
870 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
873 let isCodeGenOnly = 1 in {
874 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
876 "movaps\t{$src, $dst|$dst, $src}", [],
877 IIC_SSE_MOVA_P_RR>, VEX;
878 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
880 "movapd\t{$src, $dst|$dst, $src}", [],
881 IIC_SSE_MOVA_P_RR>, VEX;
882 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
884 "movups\t{$src, $dst|$dst, $src}", [],
885 IIC_SSE_MOVU_P_RR>, VEX;
886 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
888 "movupd\t{$src, $dst|$dst, $src}", [],
889 IIC_SSE_MOVU_P_RR>, VEX;
890 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
892 "movaps\t{$src, $dst|$dst, $src}", [],
893 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
894 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
896 "movapd\t{$src, $dst|$dst, $src}", [],
897 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
898 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
900 "movups\t{$src, $dst|$dst, $src}", [],
901 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
902 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
904 "movupd\t{$src, $dst|$dst, $src}", [],
905 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
908 let Predicates = [HasAVX] in {
909 def : Pat<(v8i32 (X86vzmovl
910 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
911 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
912 def : Pat<(v4i64 (X86vzmovl
913 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
914 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
915 def : Pat<(v8f32 (X86vzmovl
916 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
917 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
918 def : Pat<(v4f64 (X86vzmovl
919 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
920 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
924 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
925 (VMOVUPSYmr addr:$dst, VR256:$src)>;
926 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
927 (VMOVUPDYmr addr:$dst, VR256:$src)>;
929 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
930 "movaps\t{$src, $dst|$dst, $src}",
931 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
933 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
934 "movapd\t{$src, $dst|$dst, $src}",
935 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
937 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
938 "movups\t{$src, $dst|$dst, $src}",
939 [(store (v4f32 VR128:$src), addr:$dst)],
941 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
942 "movupd\t{$src, $dst|$dst, $src}",
943 [(store (v2f64 VR128:$src), addr:$dst)],
947 let isCodeGenOnly = 1 in {
948 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
949 "movaps\t{$src, $dst|$dst, $src}", [],
951 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
952 "movapd\t{$src, $dst|$dst, $src}", [],
954 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
955 "movups\t{$src, $dst|$dst, $src}", [],
957 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
958 "movupd\t{$src, $dst|$dst, $src}", [],
962 let Predicates = [HasAVX] in {
963 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
964 (VMOVUPSmr addr:$dst, VR128:$src)>;
965 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
966 (VMOVUPDmr addr:$dst, VR128:$src)>;
969 let Predicates = [UseSSE1] in
970 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
971 (MOVUPSmr addr:$dst, VR128:$src)>;
972 let Predicates = [UseSSE2] in
973 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
974 (MOVUPDmr addr:$dst, VR128:$src)>;
976 // Use vmovaps/vmovups for AVX integer load/store.
977 let Predicates = [HasAVX] in {
978 // 128-bit load/store
979 def : Pat<(alignedloadv2i64 addr:$src),
980 (VMOVAPSrm addr:$src)>;
981 def : Pat<(loadv2i64 addr:$src),
982 (VMOVUPSrm addr:$src)>;
984 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
985 (VMOVAPSmr addr:$dst, VR128:$src)>;
986 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
987 (VMOVAPSmr addr:$dst, VR128:$src)>;
988 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
989 (VMOVAPSmr addr:$dst, VR128:$src)>;
990 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
991 (VMOVAPSmr addr:$dst, VR128:$src)>;
992 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
993 (VMOVUPSmr addr:$dst, VR128:$src)>;
994 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
995 (VMOVUPSmr addr:$dst, VR128:$src)>;
996 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
997 (VMOVUPSmr addr:$dst, VR128:$src)>;
998 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
999 (VMOVUPSmr addr:$dst, VR128:$src)>;
1001 // 256-bit load/store
1002 def : Pat<(alignedloadv4i64 addr:$src),
1003 (VMOVAPSYrm addr:$src)>;
1004 def : Pat<(loadv4i64 addr:$src),
1005 (VMOVUPSYrm addr:$src)>;
1006 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
1007 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1008 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
1009 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1010 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
1011 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1012 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
1013 (VMOVAPSYmr addr:$dst, VR256:$src)>;
1014 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
1015 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1016 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
1017 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1018 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
1019 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1020 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
1021 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1023 // Special patterns for storing subvector extracts of lower 128-bits
1024 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1025 def : Pat<(alignedstore (v2f64 (extract_subvector
1026 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1027 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1028 def : Pat<(alignedstore (v4f32 (extract_subvector
1029 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1030 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1031 def : Pat<(alignedstore (v2i64 (extract_subvector
1032 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1033 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1034 def : Pat<(alignedstore (v4i32 (extract_subvector
1035 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1036 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1037 def : Pat<(alignedstore (v8i16 (extract_subvector
1038 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1039 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1040 def : Pat<(alignedstore (v16i8 (extract_subvector
1041 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1042 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1044 def : Pat<(store (v2f64 (extract_subvector
1045 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1046 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1047 def : Pat<(store (v4f32 (extract_subvector
1048 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1049 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1050 def : Pat<(store (v2i64 (extract_subvector
1051 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1052 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1053 def : Pat<(store (v4i32 (extract_subvector
1054 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1055 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1056 def : Pat<(store (v8i16 (extract_subvector
1057 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1058 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1059 def : Pat<(store (v16i8 (extract_subvector
1060 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1061 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1064 // Use movaps / movups for SSE integer load / store (one byte shorter).
1065 // The instructions selected below are then converted to MOVDQA/MOVDQU
1066 // during the SSE domain pass.
1067 let Predicates = [UseSSE1] in {
1068 def : Pat<(alignedloadv2i64 addr:$src),
1069 (MOVAPSrm addr:$src)>;
1070 def : Pat<(loadv2i64 addr:$src),
1071 (MOVUPSrm addr:$src)>;
1073 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1074 (MOVAPSmr addr:$dst, VR128:$src)>;
1075 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1076 (MOVAPSmr addr:$dst, VR128:$src)>;
1077 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1078 (MOVAPSmr addr:$dst, VR128:$src)>;
1079 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1080 (MOVAPSmr addr:$dst, VR128:$src)>;
1081 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1082 (MOVUPSmr addr:$dst, VR128:$src)>;
1083 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1084 (MOVUPSmr addr:$dst, VR128:$src)>;
1085 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1086 (MOVUPSmr addr:$dst, VR128:$src)>;
1087 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1088 (MOVUPSmr addr:$dst, VR128:$src)>;
1091 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1092 // bits are disregarded. FIXME: Set encoding to pseudo!
1093 let neverHasSideEffects = 1 in {
1094 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1095 "movaps\t{$src, $dst|$dst, $src}", [],
1096 IIC_SSE_MOVA_P_RR>, VEX;
1097 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1098 "movapd\t{$src, $dst|$dst, $src}", [],
1099 IIC_SSE_MOVA_P_RR>, VEX;
1100 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1101 "movaps\t{$src, $dst|$dst, $src}", [],
1103 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1104 "movapd\t{$src, $dst|$dst, $src}", [],
1108 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1109 // bits are disregarded. FIXME: Set encoding to pseudo!
1110 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1111 let isCodeGenOnly = 1 in {
1112 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1113 "movaps\t{$src, $dst|$dst, $src}",
1114 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1115 IIC_SSE_MOVA_P_RM>, VEX;
1116 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1117 "movapd\t{$src, $dst|$dst, $src}",
1118 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1119 IIC_SSE_MOVA_P_RM>, VEX;
1121 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1122 "movaps\t{$src, $dst|$dst, $src}",
1123 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1125 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1126 "movapd\t{$src, $dst|$dst, $src}",
1127 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1131 //===----------------------------------------------------------------------===//
1132 // SSE 1 & 2 - Move Low packed FP Instructions
1133 //===----------------------------------------------------------------------===//
1135 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1136 SDNode psnode, SDNode pdnode, string base_opc,
1137 string asm_opr, InstrItinClass itin> {
1138 def PSrm : PI<opc, MRMSrcMem,
1139 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1140 !strconcat(base_opc, "s", asm_opr),
1143 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1144 itin, SSEPackedSingle>, TB;
1146 def PDrm : PI<opc, MRMSrcMem,
1147 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1148 !strconcat(base_opc, "d", asm_opr),
1149 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1150 (scalar_to_vector (loadf64 addr:$src2)))))],
1151 itin, SSEPackedDouble>, TB, OpSize;
1154 let AddedComplexity = 20 in {
1155 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1156 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1157 IIC_SSE_MOV_LH>, VEX_4V;
1159 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1160 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1161 "\t{$src2, $dst|$dst, $src2}",
1165 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1166 "movlps\t{$src, $dst|$dst, $src}",
1167 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1168 (iPTR 0))), addr:$dst)],
1169 IIC_SSE_MOV_LH>, VEX;
1170 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1171 "movlpd\t{$src, $dst|$dst, $src}",
1172 [(store (f64 (vector_extract (v2f64 VR128:$src),
1173 (iPTR 0))), addr:$dst)],
1174 IIC_SSE_MOV_LH>, VEX;
1175 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1176 "movlps\t{$src, $dst|$dst, $src}",
1177 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1178 (iPTR 0))), addr:$dst)],
1180 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1181 "movlpd\t{$src, $dst|$dst, $src}",
1182 [(store (f64 (vector_extract (v2f64 VR128:$src),
1183 (iPTR 0))), addr:$dst)],
1186 let Predicates = [HasAVX] in {
1187 // Shuffle with VMOVLPS
1188 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1189 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1190 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1191 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1193 // Shuffle with VMOVLPD
1194 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1195 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1196 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1197 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1200 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1202 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1203 def : Pat<(store (v4i32 (X86Movlps
1204 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1205 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1206 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1208 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1209 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1211 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1214 let Predicates = [UseSSE1] in {
1215 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1216 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1217 (iPTR 0))), addr:$src1),
1218 (MOVLPSmr addr:$src1, VR128:$src2)>;
1220 // Shuffle with MOVLPS
1221 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1222 (MOVLPSrm VR128:$src1, addr:$src2)>;
1223 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1224 (MOVLPSrm VR128:$src1, addr:$src2)>;
1225 def : Pat<(X86Movlps VR128:$src1,
1226 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1227 (MOVLPSrm VR128:$src1, addr:$src2)>;
1230 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1232 (MOVLPSmr addr:$src1, VR128:$src2)>;
1233 def : Pat<(store (v4i32 (X86Movlps
1234 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1236 (MOVLPSmr addr:$src1, VR128:$src2)>;
1239 let Predicates = [UseSSE2] in {
1240 // Shuffle with MOVLPD
1241 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1242 (MOVLPDrm VR128:$src1, addr:$src2)>;
1243 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1244 (MOVLPDrm VR128:$src1, addr:$src2)>;
1247 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1249 (MOVLPDmr addr:$src1, VR128:$src2)>;
1250 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1252 (MOVLPDmr addr:$src1, VR128:$src2)>;
1255 //===----------------------------------------------------------------------===//
1256 // SSE 1 & 2 - Move Hi packed FP Instructions
1257 //===----------------------------------------------------------------------===//
1259 let AddedComplexity = 20 in {
1260 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1261 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1262 IIC_SSE_MOV_LH>, VEX_4V;
1264 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1265 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1266 "\t{$src2, $dst|$dst, $src2}",
1270 // v2f64 extract element 1 is always custom lowered to unpack high to low
1271 // and extract element 0 so the non-store version isn't too horrible.
1272 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1273 "movhps\t{$src, $dst|$dst, $src}",
1274 [(store (f64 (vector_extract
1275 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1276 (bc_v2f64 (v4f32 VR128:$src))),
1277 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1278 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1279 "movhpd\t{$src, $dst|$dst, $src}",
1280 [(store (f64 (vector_extract
1281 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1282 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1283 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1284 "movhps\t{$src, $dst|$dst, $src}",
1285 [(store (f64 (vector_extract
1286 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1287 (bc_v2f64 (v4f32 VR128:$src))),
1288 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1289 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1290 "movhpd\t{$src, $dst|$dst, $src}",
1291 [(store (f64 (vector_extract
1292 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1293 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1295 let Predicates = [HasAVX] in {
1297 def : Pat<(X86Movlhps VR128:$src1,
1298 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1299 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1300 def : Pat<(X86Movlhps VR128:$src1,
1301 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1302 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1304 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1305 // is during lowering, where it's not possible to recognize the load fold
1306 // cause it has two uses through a bitcast. One use disappears at isel time
1307 // and the fold opportunity reappears.
1308 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1309 (scalar_to_vector (loadf64 addr:$src2)))),
1310 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1313 let Predicates = [UseSSE1] in {
1315 def : Pat<(X86Movlhps VR128:$src1,
1316 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1317 (MOVHPSrm VR128:$src1, addr:$src2)>;
1318 def : Pat<(X86Movlhps VR128:$src1,
1319 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1320 (MOVHPSrm VR128:$src1, addr:$src2)>;
1323 let Predicates = [UseSSE2] in {
1324 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1325 // is during lowering, where it's not possible to recognize the load fold
1326 // cause it has two uses through a bitcast. One use disappears at isel time
1327 // and the fold opportunity reappears.
1328 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1329 (scalar_to_vector (loadf64 addr:$src2)))),
1330 (MOVHPDrm VR128:$src1, addr:$src2)>;
1333 //===----------------------------------------------------------------------===//
1334 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1335 //===----------------------------------------------------------------------===//
1337 let AddedComplexity = 20 in {
1338 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1339 (ins VR128:$src1, VR128:$src2),
1340 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1342 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1345 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1346 (ins VR128:$src1, VR128:$src2),
1347 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1349 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1353 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1354 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1355 (ins VR128:$src1, VR128:$src2),
1356 "movlhps\t{$src2, $dst|$dst, $src2}",
1358 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1360 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1361 (ins VR128:$src1, VR128:$src2),
1362 "movhlps\t{$src2, $dst|$dst, $src2}",
1364 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1368 let Predicates = [HasAVX] in {
1370 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1371 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1372 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1373 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1376 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1377 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1380 let Predicates = [UseSSE1] in {
1382 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1383 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1384 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1385 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1388 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1389 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1392 //===----------------------------------------------------------------------===//
1393 // SSE 1 & 2 - Conversion Instructions
1394 //===----------------------------------------------------------------------===//
1396 def SSE_CVT_PD : OpndItins<
1397 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1400 def SSE_CVT_PS : OpndItins<
1401 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1404 def SSE_CVT_Scalar : OpndItins<
1405 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1408 def SSE_CVT_SS2SI_32 : OpndItins<
1409 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1412 def SSE_CVT_SS2SI_64 : OpndItins<
1413 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1416 def SSE_CVT_SD2SI : OpndItins<
1417 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1420 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1421 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1422 string asm, OpndItins itins> {
1423 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1424 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1426 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1427 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1431 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1432 X86MemOperand x86memop, string asm, Domain d,
1434 let neverHasSideEffects = 1 in {
1435 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1438 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1443 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1444 X86MemOperand x86memop, string asm> {
1445 let neverHasSideEffects = 1 in {
1446 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1447 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1449 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1450 (ins DstRC:$src1, x86memop:$src),
1451 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1452 } // neverHasSideEffects = 1
1455 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1456 "cvttss2si\t{$src, $dst|$dst, $src}",
1459 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1460 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1462 XS, VEX, VEX_W, VEX_LIG;
1463 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1464 "cvttsd2si\t{$src, $dst|$dst, $src}",
1467 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1468 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1470 XD, VEX, VEX_W, VEX_LIG;
1472 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1473 // register, but the same isn't true when only using memory operands,
1474 // provide other assembly "l" and "q" forms to address this explicitly
1475 // where appropriate to do so.
1476 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1477 XS, VEX_4V, VEX_LIG;
1478 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1479 XS, VEX_4V, VEX_W, VEX_LIG;
1480 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1481 XD, VEX_4V, VEX_LIG;
1482 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1483 XD, VEX_4V, VEX_W, VEX_LIG;
1485 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1486 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1487 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1488 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1490 let Predicates = [HasAVX] in {
1491 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1492 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1493 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1494 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1495 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1496 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1497 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1498 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1500 def : Pat<(f32 (sint_to_fp GR32:$src)),
1501 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1502 def : Pat<(f32 (sint_to_fp GR64:$src)),
1503 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1504 def : Pat<(f64 (sint_to_fp GR32:$src)),
1505 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1506 def : Pat<(f64 (sint_to_fp GR64:$src)),
1507 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1510 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1511 "cvttss2si\t{$src, $dst|$dst, $src}",
1512 SSE_CVT_SS2SI_32>, XS;
1513 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1514 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1515 SSE_CVT_SS2SI_64>, XS, REX_W;
1516 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1517 "cvttsd2si\t{$src, $dst|$dst, $src}",
1519 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1520 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1521 SSE_CVT_SD2SI>, XD, REX_W;
1522 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1523 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1524 SSE_CVT_Scalar>, XS;
1525 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1526 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1527 SSE_CVT_Scalar>, XS, REX_W;
1528 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1529 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1530 SSE_CVT_Scalar>, XD;
1531 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1532 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1533 SSE_CVT_Scalar>, XD, REX_W;
1535 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1536 // and/or XMM operand(s).
1538 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1539 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1540 string asm, OpndItins itins> {
1541 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1542 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1543 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1544 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1545 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1546 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1549 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1550 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1551 PatFrag ld_frag, string asm, OpndItins itins,
1553 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1555 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1556 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1557 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1559 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1560 (ins DstRC:$src1, x86memop:$src2),
1562 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1563 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1564 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1568 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1569 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
1570 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1571 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1572 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
1573 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1575 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1576 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1577 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1578 sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1581 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1582 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1583 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1584 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1585 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1586 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1588 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1589 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1590 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1591 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1592 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1593 SSE_CVT_Scalar, 0>, XD,
1596 let Constraints = "$src1 = $dst" in {
1597 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1598 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1599 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1600 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1601 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1602 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1603 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1604 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1605 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1606 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1607 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1608 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1613 // Aliases for intrinsics
1614 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1615 ssmem, sse_load_f32, "cvttss2si",
1616 SSE_CVT_SS2SI_32>, XS, VEX;
1617 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1618 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1619 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1621 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1622 sdmem, sse_load_f64, "cvttsd2si",
1623 SSE_CVT_SD2SI>, XD, VEX;
1624 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1625 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1626 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1628 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1629 ssmem, sse_load_f32, "cvttss2si",
1630 SSE_CVT_SS2SI_32>, XS;
1631 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1632 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1633 "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
1634 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1635 sdmem, sse_load_f64, "cvttsd2si",
1637 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1638 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1639 "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1641 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1642 ssmem, sse_load_f32, "cvtss2si{l}",
1643 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1644 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1645 ssmem, sse_load_f32, "cvtss2si{q}",
1646 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1648 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1649 ssmem, sse_load_f32, "cvtss2si{l}",
1650 SSE_CVT_SS2SI_32>, XS;
1651 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1652 ssmem, sse_load_f32, "cvtss2si{q}",
1653 SSE_CVT_SS2SI_64>, XS, REX_W;
1655 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1656 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1657 SSEPackedSingle, SSE_CVT_PS>,
1658 TB, VEX, Requires<[HasAVX]>;
1659 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1660 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1661 SSEPackedSingle, SSE_CVT_PS>,
1662 TB, VEX, VEX_L, Requires<[HasAVX]>;
1664 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1665 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1666 SSEPackedSingle, SSE_CVT_PS>,
1667 TB, Requires<[UseSSE2]>;
1671 // Convert scalar double to scalar single
1672 let neverHasSideEffects = 1 in {
1673 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1674 (ins FR64:$src1, FR64:$src2),
1675 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1676 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1678 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1679 (ins FR64:$src1, f64mem:$src2),
1680 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1681 [], IIC_SSE_CVT_Scalar_RM>,
1682 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1685 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1688 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1689 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1690 [(set FR32:$dst, (fround FR64:$src))],
1691 IIC_SSE_CVT_Scalar_RR>;
1692 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1693 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1694 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1695 IIC_SSE_CVT_Scalar_RM>,
1697 Requires<[UseSSE2, OptForSize]>;
1699 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1700 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1701 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1703 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1704 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
1705 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1706 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1707 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1708 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1709 VR128:$src1, sse_load_f64:$src2))],
1710 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
1712 let Constraints = "$src1 = $dst" in {
1713 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1714 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1715 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1717 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1718 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
1719 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1720 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1721 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1722 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1723 VR128:$src1, sse_load_f64:$src2))],
1724 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
1727 // Convert scalar single to scalar double
1728 // SSE2 instructions with XS prefix
1729 let neverHasSideEffects = 1 in {
1730 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1731 (ins FR32:$src1, FR32:$src2),
1732 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1733 [], IIC_SSE_CVT_Scalar_RR>,
1734 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1736 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1737 (ins FR32:$src1, f32mem:$src2),
1738 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1739 [], IIC_SSE_CVT_Scalar_RM>,
1740 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1743 def : Pat<(f64 (fextend FR32:$src)),
1744 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1745 def : Pat<(fextend (loadf32 addr:$src)),
1746 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1748 def : Pat<(extloadf32 addr:$src),
1749 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1750 Requires<[HasAVX, OptForSize]>;
1751 def : Pat<(extloadf32 addr:$src),
1752 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1753 Requires<[HasAVX, OptForSpeed]>;
1755 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1756 "cvtss2sd\t{$src, $dst|$dst, $src}",
1757 [(set FR64:$dst, (fextend FR32:$src))],
1758 IIC_SSE_CVT_Scalar_RR>, XS,
1759 Requires<[UseSSE2]>;
1760 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1761 "cvtss2sd\t{$src, $dst|$dst, $src}",
1762 [(set FR64:$dst, (extloadf32 addr:$src))],
1763 IIC_SSE_CVT_Scalar_RM>, XS,
1764 Requires<[UseSSE2, OptForSize]>;
1766 // extload f32 -> f64. This matches load+fextend because we have a hack in
1767 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1769 // Since these loads aren't folded into the fextend, we have to match it
1771 def : Pat<(fextend (loadf32 addr:$src)),
1772 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1773 def : Pat<(extloadf32 addr:$src),
1774 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1776 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1777 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1778 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1780 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1781 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
1782 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1783 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1784 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1786 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1787 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
1788 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1789 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1790 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1791 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1793 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1794 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
1795 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1796 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1797 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1799 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1800 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
1803 // Convert packed single/double fp to doubleword
1804 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1805 "cvtps2dq\t{$src, $dst|$dst, $src}",
1806 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1807 IIC_SSE_CVT_PS_RR>, VEX;
1808 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1809 "cvtps2dq\t{$src, $dst|$dst, $src}",
1811 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1812 IIC_SSE_CVT_PS_RM>, VEX;
1813 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1814 "cvtps2dq\t{$src, $dst|$dst, $src}",
1816 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1817 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1818 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1819 "cvtps2dq\t{$src, $dst|$dst, $src}",
1821 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1822 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1823 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 "cvtps2dq\t{$src, $dst|$dst, $src}",
1825 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1827 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1828 "cvtps2dq\t{$src, $dst|$dst, $src}",
1830 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$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, $src}",
1841 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1845 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1846 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1847 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1848 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1850 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1853 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1854 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1856 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L;
1857 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1858 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1860 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1862 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1863 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1866 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1867 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1869 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1871 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1872 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1873 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1876 // Convert with truncation packed single/double fp to doubleword
1877 // SSE2 packed instructions with XS prefix
1878 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1879 "cvttps2dq\t{$src, $dst|$dst, $src}",
1881 (int_x86_sse2_cvttps2dq VR128:$src))],
1882 IIC_SSE_CVT_PS_RR>, VEX;
1883 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1884 "cvttps2dq\t{$src, $dst|$dst, $src}",
1885 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1886 (memopv4f32 addr:$src)))],
1887 IIC_SSE_CVT_PS_RM>, VEX;
1888 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1889 "cvttps2dq\t{$src, $dst|$dst, $src}",
1891 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1892 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1893 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1894 "cvttps2dq\t{$src, $dst|$dst, $src}",
1895 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1896 (memopv8f32 addr:$src)))],
1897 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1899 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1900 "cvttps2dq\t{$src, $dst|$dst, $src}",
1901 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1903 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1904 "cvttps2dq\t{$src, $dst|$dst, $src}",
1906 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1909 let Predicates = [HasAVX] in {
1910 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1911 (VCVTDQ2PSrr VR128:$src)>;
1912 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1913 (VCVTDQ2PSrm addr:$src)>;
1915 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1916 (VCVTDQ2PSrr VR128:$src)>;
1917 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1918 (VCVTDQ2PSrm addr:$src)>;
1920 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1921 (VCVTTPS2DQrr VR128:$src)>;
1922 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1923 (VCVTTPS2DQrm addr:$src)>;
1925 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1926 (VCVTDQ2PSYrr VR256:$src)>;
1927 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1928 (VCVTDQ2PSYrm addr:$src)>;
1930 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1931 (VCVTTPS2DQYrr VR256:$src)>;
1932 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1933 (VCVTTPS2DQYrm addr:$src)>;
1936 let Predicates = [UseSSE2] in {
1937 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1938 (CVTDQ2PSrr VR128:$src)>;
1939 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1940 (CVTDQ2PSrm addr:$src)>;
1942 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1943 (CVTDQ2PSrr VR128:$src)>;
1944 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1945 (CVTDQ2PSrm addr:$src)>;
1947 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1948 (CVTTPS2DQrr VR128:$src)>;
1949 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1950 (CVTTPS2DQrm addr:$src)>;
1953 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1954 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1956 (int_x86_sse2_cvttpd2dq VR128:$src))],
1957 IIC_SSE_CVT_PD_RR>, VEX;
1959 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1960 // register, but the same isn't true when using memory operands instead.
1961 // Provide other assembly rr and rm forms to address this explicitly.
1964 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1965 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1966 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1967 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1968 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1969 (memopv2f64 addr:$src)))],
1970 IIC_SSE_CVT_PD_RM>, VEX;
1973 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1974 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1976 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
1977 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
1978 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1979 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1981 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
1982 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1983 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1984 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1986 let Predicates = [HasAVX] in {
1987 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1988 (VCVTTPD2DQYrr VR256:$src)>;
1989 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1990 (VCVTTPD2DQYrm addr:$src)>;
1991 } // Predicates = [HasAVX]
1993 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1994 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1995 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1997 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1998 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1999 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2000 (memopv2f64 addr:$src)))],
2003 // Convert packed single to packed double
2004 let Predicates = [HasAVX] in {
2005 // SSE2 instructions without OpSize prefix
2006 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2007 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2008 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2009 IIC_SSE_CVT_PD_RR>, TB, VEX;
2010 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2011 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2012 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2013 IIC_SSE_CVT_PD_RM>, TB, VEX;
2014 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2015 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2017 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2018 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L;
2019 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2020 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2022 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2023 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L;
2026 let Predicates = [UseSSE2] in {
2027 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2028 "cvtps2pd\t{$src, $dst|$dst, $src}",
2029 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2030 IIC_SSE_CVT_PD_RR>, TB;
2031 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2032 "cvtps2pd\t{$src, $dst|$dst, $src}",
2033 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2034 IIC_SSE_CVT_PD_RM>, TB;
2037 // Convert Packed DW Integers to Packed Double FP
2038 let Predicates = [HasAVX] in {
2039 let neverHasSideEffects = 1, mayLoad = 1 in
2040 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2041 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2043 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2044 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2046 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2047 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2048 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2050 (int_x86_avx_cvtdq2_pd_256
2051 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L;
2052 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2053 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2055 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L;
2058 let neverHasSideEffects = 1, mayLoad = 1 in
2059 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2060 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2062 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2063 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2064 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2067 // AVX 256-bit register conversion intrinsics
2068 let Predicates = [HasAVX] in {
2069 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2070 (VCVTDQ2PDYrr VR128:$src)>;
2071 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2072 (VCVTDQ2PDYrm addr:$src)>;
2073 } // Predicates = [HasAVX]
2075 // Convert packed double to packed single
2076 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2077 // register, but the same isn't true when using memory operands instead.
2078 // Provide other assembly rr and rm forms to address this explicitly.
2079 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2080 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2081 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2082 IIC_SSE_CVT_PD_RR>, VEX;
2085 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2086 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2087 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2088 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2090 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2091 IIC_SSE_CVT_PD_RM>, VEX;
2094 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2095 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2097 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2098 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
2099 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2100 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2102 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2103 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2104 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2105 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2107 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2108 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2109 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2111 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2112 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2114 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2118 // AVX 256-bit register conversion intrinsics
2119 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2120 // whenever possible to avoid declaring two versions of each one.
2121 let Predicates = [HasAVX] in {
2122 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2123 (VCVTDQ2PSYrr VR256:$src)>;
2124 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2125 (VCVTDQ2PSYrm addr:$src)>;
2127 // Match fround and fextend for 128/256-bit conversions
2128 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2129 (VCVTPD2PSrr VR128:$src)>;
2130 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2131 (VCVTPD2PSXrm addr:$src)>;
2132 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2133 (VCVTPD2PSYrr VR256:$src)>;
2134 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2135 (VCVTPD2PSYrm addr:$src)>;
2137 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2138 (VCVTPS2PDrr VR128:$src)>;
2139 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2140 (VCVTPS2PDYrr VR128:$src)>;
2141 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2142 (VCVTPS2PDYrm addr:$src)>;
2145 let Predicates = [UseSSE2] in {
2146 // Match fround and fextend for 128 conversions
2147 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2148 (CVTPD2PSrr VR128:$src)>;
2149 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2150 (CVTPD2PSrm addr:$src)>;
2152 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2153 (CVTPS2PDrr VR128:$src)>;
2156 //===----------------------------------------------------------------------===//
2157 // SSE 1 & 2 - Compare Instructions
2158 //===----------------------------------------------------------------------===//
2160 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2161 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2162 Operand CC, SDNode OpNode, ValueType VT,
2163 PatFrag ld_frag, string asm, string asm_alt,
2165 def rr : SIi8<0xC2, MRMSrcReg,
2166 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2167 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2169 def rm : SIi8<0xC2, MRMSrcMem,
2170 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2171 [(set RC:$dst, (OpNode (VT RC:$src1),
2172 (ld_frag addr:$src2), imm:$cc))],
2175 // Accept explicit immediate argument form instead of comparison code.
2176 let neverHasSideEffects = 1 in {
2177 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2178 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2179 IIC_SSE_ALU_F32S_RR>;
2181 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2182 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2183 IIC_SSE_ALU_F32S_RM>;
2187 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2188 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2189 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2191 XS, VEX_4V, VEX_LIG;
2192 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2193 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2194 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2195 SSE_ALU_F32S>, // same latency as 32 bit compare
2196 XD, VEX_4V, VEX_LIG;
2198 let Constraints = "$src1 = $dst" in {
2199 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2200 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2201 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2203 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2204 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2205 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2206 SSE_ALU_F32S>, // same latency as 32 bit compare
2210 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2211 Intrinsic Int, string asm, OpndItins itins> {
2212 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2213 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2214 [(set VR128:$dst, (Int VR128:$src1,
2215 VR128:$src, imm:$cc))],
2217 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2218 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2219 [(set VR128:$dst, (Int VR128:$src1,
2220 (load addr:$src), imm:$cc))],
2224 // Aliases to match intrinsics which expect XMM operand(s).
2225 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2226 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2229 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2230 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2231 SSE_ALU_F32S>, // same latency as f32
2233 let Constraints = "$src1 = $dst" in {
2234 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2235 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2237 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2238 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2239 SSE_ALU_F32S>, // same latency as f32
2244 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2245 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2246 ValueType vt, X86MemOperand x86memop,
2247 PatFrag ld_frag, string OpcodeStr, Domain d> {
2248 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2249 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2250 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2251 IIC_SSE_COMIS_RR, d>;
2252 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2253 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2254 [(set EFLAGS, (OpNode (vt RC:$src1),
2255 (ld_frag addr:$src2)))],
2256 IIC_SSE_COMIS_RM, d>;
2259 let Defs = [EFLAGS] in {
2260 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2261 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2262 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2263 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2265 let Pattern = []<dag> in {
2266 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2267 "comiss", SSEPackedSingle>, TB, VEX,
2269 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2270 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2274 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2275 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2276 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2277 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2279 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2280 load, "comiss", SSEPackedSingle>, TB, VEX;
2281 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2282 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2283 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2284 "ucomiss", SSEPackedSingle>, TB;
2285 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2286 "ucomisd", SSEPackedDouble>, TB, OpSize;
2288 let Pattern = []<dag> in {
2289 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2290 "comiss", SSEPackedSingle>, TB;
2291 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2292 "comisd", SSEPackedDouble>, TB, OpSize;
2295 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2296 load, "ucomiss", SSEPackedSingle>, TB;
2297 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2298 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2300 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2301 "comiss", SSEPackedSingle>, TB;
2302 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2303 "comisd", SSEPackedDouble>, TB, OpSize;
2304 } // Defs = [EFLAGS]
2306 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2307 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2308 Operand CC, Intrinsic Int, string asm,
2309 string asm_alt, Domain d> {
2310 def rri : PIi8<0xC2, MRMSrcReg,
2311 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2312 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2313 IIC_SSE_CMPP_RR, d>;
2314 def rmi : PIi8<0xC2, MRMSrcMem,
2315 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2316 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2317 IIC_SSE_CMPP_RM, d>;
2319 // Accept explicit immediate argument form instead of comparison code.
2320 let neverHasSideEffects = 1 in {
2321 def rri_alt : PIi8<0xC2, MRMSrcReg,
2322 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2323 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2324 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2325 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2326 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2330 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2331 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2332 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2333 SSEPackedSingle>, TB, VEX_4V;
2334 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2335 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2336 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2337 SSEPackedDouble>, TB, OpSize, VEX_4V;
2338 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2339 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2340 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2341 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2342 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2343 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2344 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2345 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2346 let Constraints = "$src1 = $dst" in {
2347 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2348 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2349 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2350 SSEPackedSingle>, TB;
2351 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2352 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2353 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2354 SSEPackedDouble>, TB, OpSize;
2357 let Predicates = [HasAVX] in {
2358 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2359 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2360 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2361 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2362 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2363 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2364 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2365 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2367 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2368 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2369 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2370 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2371 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2372 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2373 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2374 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2377 let Predicates = [UseSSE1] in {
2378 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2379 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2380 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2381 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2384 let Predicates = [UseSSE2] in {
2385 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2386 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2387 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2388 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2391 //===----------------------------------------------------------------------===//
2392 // SSE 1 & 2 - Shuffle Instructions
2393 //===----------------------------------------------------------------------===//
2395 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2396 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2397 ValueType vt, string asm, PatFrag mem_frag,
2398 Domain d, bit IsConvertibleToThreeAddress = 0> {
2399 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2400 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2401 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2402 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2403 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2404 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2405 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2406 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2407 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2410 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2411 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2412 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2413 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2414 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2415 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2416 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2417 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2418 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2419 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2420 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2421 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2423 let Constraints = "$src1 = $dst" in {
2424 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2425 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2426 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2428 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2429 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2430 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2434 let Predicates = [HasAVX] in {
2435 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2436 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2437 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2438 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2439 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2441 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2442 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2443 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2444 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2445 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2448 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2449 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2450 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2451 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2452 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2454 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2455 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2456 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2457 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2458 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2461 let Predicates = [UseSSE1] in {
2462 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2463 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2464 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2465 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2466 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2469 let Predicates = [UseSSE2] in {
2470 // Generic SHUFPD patterns
2471 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2472 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2473 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2474 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2475 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2478 //===----------------------------------------------------------------------===//
2479 // SSE 1 & 2 - Unpack Instructions
2480 //===----------------------------------------------------------------------===//
2482 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2483 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2484 PatFrag mem_frag, RegisterClass RC,
2485 X86MemOperand x86memop, string asm,
2487 def rr : PI<opc, MRMSrcReg,
2488 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2490 (vt (OpNode RC:$src1, RC:$src2)))],
2492 def rm : PI<opc, MRMSrcMem,
2493 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2495 (vt (OpNode RC:$src1,
2496 (mem_frag addr:$src2))))],
2500 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2501 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2502 SSEPackedSingle>, TB, VEX_4V;
2503 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2504 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2505 SSEPackedDouble>, TB, OpSize, VEX_4V;
2506 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2507 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2508 SSEPackedSingle>, TB, VEX_4V;
2509 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2510 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2511 SSEPackedDouble>, TB, OpSize, VEX_4V;
2513 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2514 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2515 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2516 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2517 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2518 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2519 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2520 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2521 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2522 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2523 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2524 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2526 let Constraints = "$src1 = $dst" in {
2527 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2528 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2529 SSEPackedSingle>, TB;
2530 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2531 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2532 SSEPackedDouble>, TB, OpSize;
2533 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2534 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2535 SSEPackedSingle>, TB;
2536 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2537 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2538 SSEPackedDouble>, TB, OpSize;
2539 } // Constraints = "$src1 = $dst"
2541 let Predicates = [HasAVX1Only] in {
2542 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2543 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2544 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2545 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2546 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2547 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2548 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2549 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2551 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2552 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2553 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2554 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2555 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2556 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2557 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2558 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2561 let Predicates = [HasAVX] in {
2562 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2563 // problem is during lowering, where it's not possible to recognize the load
2564 // fold cause it has two uses through a bitcast. One use disappears at isel
2565 // time and the fold opportunity reappears.
2566 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2567 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2570 let Predicates = [UseSSE2] in {
2571 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2572 // problem is during lowering, where it's not possible to recognize the load
2573 // fold cause it has two uses through a bitcast. One use disappears at isel
2574 // time and the fold opportunity reappears.
2575 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2576 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2579 //===----------------------------------------------------------------------===//
2580 // SSE 1 & 2 - Extract Floating-Point Sign mask
2581 //===----------------------------------------------------------------------===//
2583 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2584 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2586 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2587 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2588 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2589 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2590 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2591 IIC_SSE_MOVMSK, d>, REX_W;
2594 let Predicates = [HasAVX] in {
2595 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2596 "movmskps", SSEPackedSingle>, TB, VEX;
2597 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2598 "movmskpd", SSEPackedDouble>, TB,
2600 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2601 "movmskps", SSEPackedSingle>, TB,
2603 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2604 "movmskpd", SSEPackedDouble>, TB,
2607 def : Pat<(i32 (X86fgetsign FR32:$src)),
2608 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2609 def : Pat<(i64 (X86fgetsign FR32:$src)),
2610 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2611 def : Pat<(i32 (X86fgetsign FR64:$src)),
2612 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2613 def : Pat<(i64 (X86fgetsign FR64:$src)),
2614 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2617 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2618 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2619 SSEPackedSingle>, TB, VEX;
2620 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2621 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2622 SSEPackedDouble>, TB,
2624 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2625 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2626 SSEPackedSingle>, TB, VEX, VEX_L;
2627 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2628 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2629 SSEPackedDouble>, TB,
2633 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2634 SSEPackedSingle>, TB;
2635 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2636 SSEPackedDouble>, TB, OpSize;
2638 def : Pat<(i32 (X86fgetsign FR32:$src)),
2639 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2640 Requires<[UseSSE1]>;
2641 def : Pat<(i64 (X86fgetsign FR32:$src)),
2642 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2643 Requires<[UseSSE1]>;
2644 def : Pat<(i32 (X86fgetsign FR64:$src)),
2645 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2646 Requires<[UseSSE2]>;
2647 def : Pat<(i64 (X86fgetsign FR64:$src)),
2648 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2649 Requires<[UseSSE2]>;
2651 //===---------------------------------------------------------------------===//
2652 // SSE2 - Packed Integer Logical Instructions
2653 //===---------------------------------------------------------------------===//
2655 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2657 /// PDI_binop_rm - Simple SSE2 binary operator.
2658 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2659 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2660 X86MemOperand x86memop, OpndItins itins,
2661 bit IsCommutable, bit Is2Addr> {
2662 let isCommutable = IsCommutable in
2663 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2664 (ins RC:$src1, RC:$src2),
2666 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2667 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2668 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2669 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2670 (ins RC:$src1, x86memop:$src2),
2672 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2673 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2674 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2675 (bitconvert (memop_frag addr:$src2)))))],
2678 } // ExeDomain = SSEPackedInt
2680 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2681 ValueType OpVT128, ValueType OpVT256,
2682 OpndItins itins, bit IsCommutable = 0> {
2683 let Predicates = [HasAVX] in
2684 defm VP#NAME# : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2685 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2687 let Constraints = "$src1 = $dst" in
2688 defm P#NAME# : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2689 memopv2i64, i128mem, itins, IsCommutable, 1>;
2691 let Predicates = [HasAVX2] in
2692 defm VP#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2693 OpVT256, VR256, memopv4i64, i256mem, itins,
2694 IsCommutable, 0>, VEX_4V, VEX_L;
2697 // These are ordered here for pattern ordering requirements with the fp versions
2699 defm AND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2700 defm OR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2701 defm XOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2702 defm ANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2703 SSE_BIT_ITINS_P, 0>;
2705 //===----------------------------------------------------------------------===//
2706 // SSE 1 & 2 - Logical Instructions
2707 //===----------------------------------------------------------------------===//
2709 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2711 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2712 SDNode OpNode, OpndItins itins> {
2713 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2714 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2717 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2718 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2721 let Constraints = "$src1 = $dst" in {
2722 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2723 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2726 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2727 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2732 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2733 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2735 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2737 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2740 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2741 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2744 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2746 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2748 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2749 !strconcat(OpcodeStr, "ps"), f256mem,
2750 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2751 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2752 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2754 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2755 !strconcat(OpcodeStr, "pd"), f256mem,
2756 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2757 (bc_v4i64 (v4f64 VR256:$src2))))],
2758 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2759 (memopv4i64 addr:$src2)))], 0>,
2760 TB, OpSize, VEX_4V, VEX_L;
2762 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2763 // are all promoted to v2i64, and the patterns are covered by the int
2764 // version. This is needed in SSE only, because v2i64 isn't supported on
2765 // SSE1, but only on SSE2.
2766 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2767 !strconcat(OpcodeStr, "ps"), f128mem, [],
2768 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2769 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2771 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2772 !strconcat(OpcodeStr, "pd"), f128mem,
2773 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2774 (bc_v2i64 (v2f64 VR128:$src2))))],
2775 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2776 (memopv2i64 addr:$src2)))], 0>,
2779 let Constraints = "$src1 = $dst" in {
2780 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2781 !strconcat(OpcodeStr, "ps"), f128mem,
2782 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2783 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2784 (memopv2i64 addr:$src2)))]>, TB;
2786 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2787 !strconcat(OpcodeStr, "pd"), f128mem,
2788 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2789 (bc_v2i64 (v2f64 VR128:$src2))))],
2790 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2791 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2795 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2796 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2797 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2798 let isCommutable = 0 in
2799 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2801 //===----------------------------------------------------------------------===//
2802 // SSE 1 & 2 - Arithmetic Instructions
2803 //===----------------------------------------------------------------------===//
2805 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2808 /// In addition, we also have a special variant of the scalar form here to
2809 /// represent the associated intrinsic operation. This form is unlike the
2810 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2811 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2813 /// These three forms can each be reg+reg or reg+mem.
2816 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2818 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2821 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2822 OpNode, FR32, f32mem,
2823 itins.s, Is2Addr>, XS;
2824 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2825 OpNode, FR64, f64mem,
2826 itins.d, Is2Addr>, XD;
2829 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2832 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2833 v4f32, f128mem, memopv4f32, SSEPackedSingle, itins.s, Is2Addr>,
2835 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2836 v2f64, f128mem, memopv2f64, SSEPackedDouble, itins.d, Is2Addr>,
2840 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2843 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2844 v8f32, f256mem, memopv8f32, SSEPackedSingle, itins.s, 0>,
2846 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2847 v4f64, f256mem, memopv4f64, SSEPackedDouble, itins.d, 0>,
2851 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2854 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2855 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2856 itins.s, Is2Addr>, XS;
2857 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2858 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2859 itins.d, Is2Addr>, XD;
2862 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2865 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2866 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2867 SSEPackedSingle, itins.s, Is2Addr>,
2870 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2871 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2872 SSEPackedDouble, itins.d, Is2Addr>,
2876 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr,
2878 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2879 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2880 SSEPackedSingle, itins.s, 0>, TB, VEX_L;
2882 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2883 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2884 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_L;
2887 // Binary Arithmetic instructions
2888 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2889 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2891 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P, 0>,
2892 basic_sse12_fp_binop_p_y<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2894 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2895 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2897 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P, 0>,
2898 basic_sse12_fp_binop_p_y<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2901 let isCommutable = 0 in {
2902 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2903 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2905 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P, 0>,
2906 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2908 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2909 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2911 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_ALU_ITINS_P, 0>,
2912 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2914 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2915 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2917 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P, 0>,
2918 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P, 0>,
2919 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2920 basic_sse12_fp_binop_p_y_int<0x5F, "max", SSE_ALU_ITINS_P>,
2922 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2923 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2925 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P, 0>,
2926 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P, 0>,
2927 basic_sse12_fp_binop_p_y_int<0x5D, "min", SSE_ALU_ITINS_P>,
2928 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2932 let Constraints = "$src1 = $dst" in {
2933 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2934 basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>,
2935 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2936 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2937 basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>,
2938 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2940 let isCommutable = 0 in {
2941 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2942 basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>,
2943 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2944 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2945 basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>,
2946 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2947 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2948 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>,
2949 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>,
2950 basic_sse12_fp_binop_p_int<0x5F, "max", SSE_ALU_ITINS_P>;
2951 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2952 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>,
2953 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>,
2954 basic_sse12_fp_binop_p_int<0x5D, "min", SSE_ALU_ITINS_P>;
2958 let isCodeGenOnly = 1 in {
2959 defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
2961 defm VMAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P, 0>,
2962 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>, VEX_4V;
2963 defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
2965 defm VMINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P, 0>,
2966 basic_sse12_fp_binop_p_y<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>, VEX_4V;
2967 let Constraints = "$src1 = $dst" in {
2968 defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>,
2969 basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
2970 defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>,
2971 basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
2976 /// In addition, we also have a special variant of the scalar form here to
2977 /// represent the associated intrinsic operation. This form is unlike the
2978 /// plain scalar form, in that it takes an entire vector (instead of a
2979 /// scalar) and leaves the top elements undefined.
2981 /// And, we have a special variant form for a full-vector intrinsic form.
2983 def SSE_SQRTP : OpndItins<
2984 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2987 def SSE_SQRTS : OpndItins<
2988 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2991 def SSE_RCPP : OpndItins<
2992 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2995 def SSE_RCPS : OpndItins<
2996 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2999 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3000 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3001 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3002 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3003 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3004 [(set FR32:$dst, (OpNode FR32:$src))]>;
3005 // For scalar unary operations, fold a load into the operation
3006 // only in OptForSize mode. It eliminates an instruction, but it also
3007 // eliminates a whole-register clobber (the load), so it introduces a
3008 // partial register update condition.
3009 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3010 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3011 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3012 Requires<[UseSSE1, OptForSize]>;
3013 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3014 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3015 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
3016 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3017 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3018 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
3021 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
3022 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3023 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
3024 !strconcat(OpcodeStr,
3025 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3026 let mayLoad = 1 in {
3027 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
3028 !strconcat(OpcodeStr,
3029 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3030 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3031 (ins VR128:$src1, ssmem:$src2),
3032 !strconcat(OpcodeStr,
3033 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3037 /// sse1_fp_unop_p - SSE1 unops in packed form.
3038 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3040 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3041 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3042 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3043 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3044 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3045 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3048 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3049 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3051 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3052 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3053 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3055 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3056 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3057 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3061 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3062 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3063 Intrinsic V4F32Int, OpndItins itins> {
3064 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3065 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3066 [(set VR128:$dst, (V4F32Int VR128:$src))],
3068 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3069 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3070 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3074 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3075 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3076 Intrinsic V4F32Int, OpndItins itins> {
3077 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3078 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3079 [(set VR256:$dst, (V4F32Int VR256:$src))],
3081 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3082 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3083 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3087 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3088 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3089 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3090 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3091 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3092 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3093 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3094 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3095 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3096 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3097 Requires<[UseSSE2, OptForSize]>;
3098 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3099 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3100 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3101 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3102 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3103 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3106 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3107 let hasSideEffects = 0 in
3108 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3109 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3110 !strconcat(OpcodeStr,
3111 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3112 let mayLoad = 1 in {
3113 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3114 !strconcat(OpcodeStr,
3115 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3116 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3117 (ins VR128:$src1, sdmem:$src2),
3118 !strconcat(OpcodeStr,
3119 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3123 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3124 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3125 SDNode OpNode, OpndItins itins> {
3126 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3127 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3128 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3129 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3130 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3131 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3134 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3135 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3137 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3138 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3139 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3141 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3142 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3143 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3147 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3148 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3149 Intrinsic V2F64Int, OpndItins itins> {
3150 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3151 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3152 [(set VR128:$dst, (V2F64Int VR128:$src))],
3154 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3155 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3156 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3160 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3161 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3162 Intrinsic V2F64Int, OpndItins itins> {
3163 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3164 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3165 [(set VR256:$dst, (V2F64Int VR256:$src))],
3167 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3168 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3169 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3173 let Predicates = [HasAVX] in {
3175 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3176 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3178 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3179 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3180 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3181 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3182 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3184 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3186 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3188 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3192 // Reciprocal approximations. Note that these typically require refinement
3193 // in order to obtain suitable precision.
3194 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3195 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3196 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3197 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3199 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3202 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3203 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3204 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3205 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3207 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3211 def : Pat<(f32 (fsqrt FR32:$src)),
3212 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3213 def : Pat<(f32 (fsqrt (load addr:$src))),
3214 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3215 Requires<[HasAVX, OptForSize]>;
3216 def : Pat<(f64 (fsqrt FR64:$src)),
3217 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3218 def : Pat<(f64 (fsqrt (load addr:$src))),
3219 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3220 Requires<[HasAVX, OptForSize]>;
3222 def : Pat<(f32 (X86frsqrt FR32:$src)),
3223 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3224 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3225 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3226 Requires<[HasAVX, OptForSize]>;
3228 def : Pat<(f32 (X86frcp FR32:$src)),
3229 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3230 def : Pat<(f32 (X86frcp (load addr:$src))),
3231 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3232 Requires<[HasAVX, OptForSize]>;
3234 let Predicates = [HasAVX] in {
3235 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3236 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3237 (COPY_TO_REGCLASS VR128:$src, FR32)),
3239 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3240 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3242 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3243 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3244 (COPY_TO_REGCLASS VR128:$src, FR64)),
3246 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3247 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3249 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3250 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3251 (COPY_TO_REGCLASS VR128:$src, FR32)),
3253 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3254 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3256 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3257 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3258 (COPY_TO_REGCLASS VR128:$src, FR32)),
3260 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3261 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3265 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3267 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3268 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3269 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3271 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3272 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3274 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3275 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3276 Intrinsic F32Int, OpndItins itins> {
3277 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3278 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3279 [(set FR32:$dst, (OpNode FR32:$src))]>;
3280 // For scalar unary operations, fold a load into the operation
3281 // only in OptForSize mode. It eliminates an instruction, but it also
3282 // eliminates a whole-register clobber (the load), so it introduces a
3283 // partial register update condition.
3284 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3285 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3286 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3287 Requires<[UseSSE1, OptForSize]>;
3288 let Constraints = "$src1 = $dst" in {
3289 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3290 (ins VR128:$src1, VR128:$src2),
3291 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3293 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3294 (ins VR128:$src1, ssmem:$src2),
3295 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3300 // Reciprocal approximations. Note that these typically require refinement
3301 // in order to obtain suitable precision.
3302 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3304 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3305 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3307 let Predicates = [UseSSE1] in {
3308 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3309 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3312 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3314 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3315 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3316 let Predicates = [UseSSE1] in {
3317 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3318 (RCPSSr_Int VR128:$src, VR128:$src)>;
3321 // There is no f64 version of the reciprocal approximation instructions.
3323 //===----------------------------------------------------------------------===//
3324 // SSE 1 & 2 - Non-temporal stores
3325 //===----------------------------------------------------------------------===//
3327 let AddedComplexity = 400 in { // Prefer non-temporal versions
3328 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3329 (ins f128mem:$dst, VR128:$src),
3330 "movntps\t{$src, $dst|$dst, $src}",
3331 [(alignednontemporalstore (v4f32 VR128:$src),
3333 IIC_SSE_MOVNT>, VEX;
3334 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3335 (ins f128mem:$dst, VR128:$src),
3336 "movntpd\t{$src, $dst|$dst, $src}",
3337 [(alignednontemporalstore (v2f64 VR128:$src),
3339 IIC_SSE_MOVNT>, VEX;
3341 let ExeDomain = SSEPackedInt in
3342 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3343 (ins f128mem:$dst, VR128:$src),
3344 "movntdq\t{$src, $dst|$dst, $src}",
3345 [(alignednontemporalstore (v2i64 VR128:$src),
3347 IIC_SSE_MOVNT>, VEX;
3349 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3350 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3352 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3353 (ins f256mem:$dst, VR256:$src),
3354 "movntps\t{$src, $dst|$dst, $src}",
3355 [(alignednontemporalstore (v8f32 VR256:$src),
3357 IIC_SSE_MOVNT>, VEX, VEX_L;
3358 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3359 (ins f256mem:$dst, VR256:$src),
3360 "movntpd\t{$src, $dst|$dst, $src}",
3361 [(alignednontemporalstore (v4f64 VR256:$src),
3363 IIC_SSE_MOVNT>, VEX, VEX_L;
3364 let ExeDomain = SSEPackedInt in
3365 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3366 (ins f256mem:$dst, VR256:$src),
3367 "movntdq\t{$src, $dst|$dst, $src}",
3368 [(alignednontemporalstore (v4i64 VR256:$src),
3370 IIC_SSE_MOVNT>, VEX, VEX_L;
3373 let AddedComplexity = 400 in { // Prefer non-temporal versions
3374 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3375 "movntps\t{$src, $dst|$dst, $src}",
3376 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3378 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3379 "movntpd\t{$src, $dst|$dst, $src}",
3380 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3383 let ExeDomain = SSEPackedInt in
3384 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3385 "movntdq\t{$src, $dst|$dst, $src}",
3386 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3389 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3390 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3392 // There is no AVX form for instructions below this point
3393 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3394 "movnti{l}\t{$src, $dst|$dst, $src}",
3395 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3397 TB, Requires<[HasSSE2]>;
3398 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3399 "movnti{q}\t{$src, $dst|$dst, $src}",
3400 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3402 TB, Requires<[HasSSE2]>;
3405 //===----------------------------------------------------------------------===//
3406 // SSE 1 & 2 - Prefetch and memory fence
3407 //===----------------------------------------------------------------------===//
3409 // Prefetch intrinsic.
3410 let Predicates = [HasSSE1] in {
3411 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3412 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3413 IIC_SSE_PREFETCH>, TB;
3414 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3415 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3416 IIC_SSE_PREFETCH>, TB;
3417 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3418 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3419 IIC_SSE_PREFETCH>, TB;
3420 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3421 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3422 IIC_SSE_PREFETCH>, TB;
3426 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3427 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3428 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3430 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3431 // was introduced with SSE2, it's backward compatible.
3432 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3434 // Load, store, and memory fence
3435 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3436 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3437 TB, Requires<[HasSSE1]>;
3438 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3439 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3440 TB, Requires<[HasSSE2]>;
3441 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3442 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3443 TB, Requires<[HasSSE2]>;
3445 def : Pat<(X86SFence), (SFENCE)>;
3446 def : Pat<(X86LFence), (LFENCE)>;
3447 def : Pat<(X86MFence), (MFENCE)>;
3449 //===----------------------------------------------------------------------===//
3450 // SSE 1 & 2 - Load/Store XCSR register
3451 //===----------------------------------------------------------------------===//
3453 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3454 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3455 IIC_SSE_LDMXCSR>, VEX;
3456 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3457 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3458 IIC_SSE_STMXCSR>, VEX;
3460 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3461 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3463 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3464 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3467 //===---------------------------------------------------------------------===//
3468 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3469 //===---------------------------------------------------------------------===//
3471 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3473 let neverHasSideEffects = 1 in {
3474 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3475 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3477 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3478 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3480 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3481 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3483 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3484 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3489 let isCodeGenOnly = 1 in {
3490 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3491 "movdqa\t{$src, $dst|$dst, $src}", [],
3494 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3495 "movdqa\t{$src, $dst|$dst, $src}", [],
3496 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3497 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3498 "movdqu\t{$src, $dst|$dst, $src}", [],
3501 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3502 "movdqu\t{$src, $dst|$dst, $src}", [],
3503 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3506 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3507 neverHasSideEffects = 1 in {
3508 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3509 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3511 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3512 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3514 let Predicates = [HasAVX] in {
3515 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3516 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3518 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3519 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3524 let mayStore = 1, neverHasSideEffects = 1 in {
3525 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3526 (ins i128mem:$dst, VR128:$src),
3527 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3529 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3530 (ins i256mem:$dst, VR256:$src),
3531 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3533 let Predicates = [HasAVX] in {
3534 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3535 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3537 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3538 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3543 let neverHasSideEffects = 1 in
3544 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3545 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3547 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3548 "movdqu\t{$src, $dst|$dst, $src}",
3549 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3552 let isCodeGenOnly = 1 in {
3553 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3554 "movdqa\t{$src, $dst|$dst, $src}", [],
3557 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3558 "movdqu\t{$src, $dst|$dst, $src}",
3559 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3562 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3563 neverHasSideEffects = 1 in {
3564 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3565 "movdqa\t{$src, $dst|$dst, $src}",
3566 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3568 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3569 "movdqu\t{$src, $dst|$dst, $src}",
3570 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3572 XS, Requires<[UseSSE2]>;
3575 let mayStore = 1 in {
3576 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3577 "movdqa\t{$src, $dst|$dst, $src}",
3578 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3580 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3581 "movdqu\t{$src, $dst|$dst, $src}",
3582 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3584 XS, Requires<[UseSSE2]>;
3587 } // ExeDomain = SSEPackedInt
3589 let Predicates = [HasAVX] in {
3590 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3591 (VMOVDQUmr addr:$dst, VR128:$src)>;
3592 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3593 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3595 let Predicates = [UseSSE2] in
3596 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3597 (MOVDQUmr addr:$dst, VR128:$src)>;
3599 //===---------------------------------------------------------------------===//
3600 // SSE2 - Packed Integer Arithmetic Instructions
3601 //===---------------------------------------------------------------------===//
3603 def SSE_PMADD : OpndItins<
3604 IIC_SSE_PMADD, IIC_SSE_PMADD
3607 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3609 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3610 RegisterClass RC, PatFrag memop_frag,
3611 X86MemOperand x86memop,
3613 bit IsCommutable = 0,
3615 let isCommutable = IsCommutable in
3616 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3617 (ins RC:$src1, RC:$src2),
3619 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3620 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3621 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3622 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3623 (ins RC:$src1, x86memop:$src2),
3625 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3626 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3627 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3631 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3632 Intrinsic IntId256, OpndItins itins,
3633 bit IsCommutable = 0> {
3634 let Predicates = [HasAVX] in
3635 defm VP#NAME# : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3636 VR128, memopv2i64, i128mem, itins,
3637 IsCommutable, 0>, VEX_4V;
3639 let Constraints = "$src1 = $dst" in
3640 defm P#NAME# : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3641 i128mem, itins, IsCommutable, 1>;
3643 let Predicates = [HasAVX2] in
3644 defm VP#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3645 VR256, memopv4i64, i256mem, itins,
3646 IsCommutable, 0>, VEX_4V, VEX_L;
3649 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3650 string OpcodeStr, SDNode OpNode,
3651 SDNode OpNode2, RegisterClass RC,
3652 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3653 ShiftOpndItins itins,
3655 // src2 is always 128-bit
3656 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3657 (ins RC:$src1, VR128:$src2),
3659 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3660 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3661 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3663 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3664 (ins RC:$src1, i128mem:$src2),
3666 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3667 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3668 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3669 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3670 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3671 (ins RC:$src1, i32i8imm:$src2),
3673 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3674 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3675 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3678 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3679 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3680 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3681 PatFrag memop_frag, X86MemOperand x86memop,
3683 bit IsCommutable = 0, bit Is2Addr = 1> {
3684 let isCommutable = IsCommutable in
3685 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3686 (ins RC:$src1, RC:$src2),
3688 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3689 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3690 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3691 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3692 (ins RC:$src1, x86memop:$src2),
3694 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3695 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3696 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3697 (bitconvert (memop_frag addr:$src2)))))]>;
3699 } // ExeDomain = SSEPackedInt
3701 defm ADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3702 SSE_INTALU_ITINS_P, 1>;
3703 defm ADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3704 SSE_INTALU_ITINS_P, 1>;
3705 defm ADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3706 SSE_INTALU_ITINS_P, 1>;
3707 defm ADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3708 SSE_INTALUQ_ITINS_P, 1>;
3709 defm MULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3710 SSE_INTMUL_ITINS_P, 1>;
3711 defm SUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3712 SSE_INTALU_ITINS_P, 0>;
3713 defm SUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3714 SSE_INTALU_ITINS_P, 0>;
3715 defm SUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3716 SSE_INTALU_ITINS_P, 0>;
3717 defm SUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3718 SSE_INTALUQ_ITINS_P, 0>;
3719 defm SUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3720 SSE_INTALU_ITINS_P, 0>;
3721 defm SUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3722 SSE_INTALU_ITINS_P, 0>;
3723 defm MINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3724 SSE_INTALU_ITINS_P, 1>;
3725 defm MINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3726 SSE_INTALU_ITINS_P, 1>;
3727 defm MAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3728 SSE_INTALU_ITINS_P, 1>;
3729 defm MAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3730 SSE_INTALU_ITINS_P, 1>;
3733 defm SUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3734 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3735 defm SUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3736 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3737 defm ADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3738 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3739 defm ADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3740 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3741 defm ADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3742 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3743 defm ADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3744 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3745 defm MULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3746 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3747 defm MULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3748 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3749 defm MADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3750 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3751 defm AVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3752 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3753 defm AVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3754 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3755 defm SADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3756 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3758 let Predicates = [HasAVX] in
3759 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3760 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3762 let Predicates = [HasAVX2] in
3763 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3764 VR256, memopv4i64, i256mem,
3765 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3766 let Constraints = "$src1 = $dst" in
3767 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3768 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3770 //===---------------------------------------------------------------------===//
3771 // SSE2 - Packed Integer Logical Instructions
3772 //===---------------------------------------------------------------------===//
3774 let Predicates = [HasAVX] in {
3775 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3776 VR128, v8i16, v8i16, bc_v8i16,
3777 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3778 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3779 VR128, v4i32, v4i32, bc_v4i32,
3780 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3781 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3782 VR128, v2i64, v2i64, bc_v2i64,
3783 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3785 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3786 VR128, v8i16, v8i16, bc_v8i16,
3787 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3788 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3789 VR128, v4i32, v4i32, bc_v4i32,
3790 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3791 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3792 VR128, v2i64, v2i64, bc_v2i64,
3793 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3795 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3796 VR128, v8i16, v8i16, bc_v8i16,
3797 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3798 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3799 VR128, v4i32, v4i32, bc_v4i32,
3800 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3802 let ExeDomain = SSEPackedInt in {
3803 // 128-bit logical shifts.
3804 def VPSLLDQri : PDIi8<0x73, MRM7r,
3805 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3806 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3808 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3810 def VPSRLDQri : PDIi8<0x73, MRM3r,
3811 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3812 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3814 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3816 // PSRADQri doesn't exist in SSE[1-3].
3818 } // Predicates = [HasAVX]
3820 let Predicates = [HasAVX2] in {
3821 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3822 VR256, v16i16, v8i16, bc_v8i16,
3823 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3824 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3825 VR256, v8i32, v4i32, bc_v4i32,
3826 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3827 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3828 VR256, v4i64, v2i64, bc_v2i64,
3829 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3831 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3832 VR256, v16i16, v8i16, bc_v8i16,
3833 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3834 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3835 VR256, v8i32, v4i32, bc_v4i32,
3836 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3837 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3838 VR256, v4i64, v2i64, bc_v2i64,
3839 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3841 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3842 VR256, v16i16, v8i16, bc_v8i16,
3843 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3844 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3845 VR256, v8i32, v4i32, bc_v4i32,
3846 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3848 let ExeDomain = SSEPackedInt in {
3849 // 256-bit logical shifts.
3850 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3851 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3852 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3854 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3856 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3857 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3858 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3860 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3862 // PSRADQYri doesn't exist in SSE[1-3].
3864 } // Predicates = [HasAVX2]
3866 let Constraints = "$src1 = $dst" in {
3867 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3868 VR128, v8i16, v8i16, bc_v8i16,
3869 SSE_INTSHIFT_ITINS_P>;
3870 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3871 VR128, v4i32, v4i32, bc_v4i32,
3872 SSE_INTSHIFT_ITINS_P>;
3873 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3874 VR128, v2i64, v2i64, bc_v2i64,
3875 SSE_INTSHIFT_ITINS_P>;
3877 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3878 VR128, v8i16, v8i16, bc_v8i16,
3879 SSE_INTSHIFT_ITINS_P>;
3880 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3881 VR128, v4i32, v4i32, bc_v4i32,
3882 SSE_INTSHIFT_ITINS_P>;
3883 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3884 VR128, v2i64, v2i64, bc_v2i64,
3885 SSE_INTSHIFT_ITINS_P>;
3887 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3888 VR128, v8i16, v8i16, bc_v8i16,
3889 SSE_INTSHIFT_ITINS_P>;
3890 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3891 VR128, v4i32, v4i32, bc_v4i32,
3892 SSE_INTSHIFT_ITINS_P>;
3894 let ExeDomain = SSEPackedInt in {
3895 // 128-bit logical shifts.
3896 def PSLLDQri : PDIi8<0x73, MRM7r,
3897 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3898 "pslldq\t{$src2, $dst|$dst, $src2}",
3900 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3901 def PSRLDQri : PDIi8<0x73, MRM3r,
3902 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3903 "psrldq\t{$src2, $dst|$dst, $src2}",
3905 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3906 // PSRADQri doesn't exist in SSE[1-3].
3908 } // Constraints = "$src1 = $dst"
3910 let Predicates = [HasAVX] in {
3911 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3912 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3913 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3914 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3915 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3916 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3918 // Shift up / down and insert zero's.
3919 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3920 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3921 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3922 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3925 let Predicates = [HasAVX2] in {
3926 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3927 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3928 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3929 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3932 let Predicates = [UseSSE2] in {
3933 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3934 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3935 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3936 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3937 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3938 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3940 // Shift up / down and insert zero's.
3941 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3942 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3943 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3944 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3947 //===---------------------------------------------------------------------===//
3948 // SSE2 - Packed Integer Comparison Instructions
3949 //===---------------------------------------------------------------------===//
3951 defm CMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
3952 SSE_INTALU_ITINS_P, 1>;
3953 defm CMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
3954 SSE_INTALU_ITINS_P, 1>;
3955 defm CMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
3956 SSE_INTALU_ITINS_P, 1>;
3957 defm CMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
3958 SSE_INTALU_ITINS_P, 0>;
3959 defm CMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
3960 SSE_INTALU_ITINS_P, 0>;
3961 defm CMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
3962 SSE_INTALU_ITINS_P, 0>;
3964 //===---------------------------------------------------------------------===//
3965 // SSE2 - Packed Integer Pack Instructions
3966 //===---------------------------------------------------------------------===//
3968 let Predicates = [HasAVX] in {
3969 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3970 VR128, memopv2i64, i128mem,
3971 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3972 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3973 VR128, memopv2i64, i128mem,
3974 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3975 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3976 VR128, memopv2i64, i128mem,
3977 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V;
3980 let Predicates = [HasAVX2] in {
3981 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3982 VR256, memopv4i64, i256mem,
3983 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
3984 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
3985 VR256, memopv4i64, i256mem,
3986 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
3987 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
3988 VR256, memopv4i64, i256mem,
3989 SSE_INTALU_ITINS_P, 0, 0>, VEX_4V, VEX_L;
3992 let Constraints = "$src1 = $dst" in {
3993 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3994 VR128, memopv2i64, i128mem,
3995 SSE_INTALU_ITINS_P>;
3996 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3997 VR128, memopv2i64, i128mem,
3998 SSE_INTALU_ITINS_P>;
3999 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4000 VR128, memopv2i64, i128mem,
4001 SSE_INTALU_ITINS_P>;
4002 } // Constraints = "$src1 = $dst"
4004 //===---------------------------------------------------------------------===//
4005 // SSE2 - Packed Integer Shuffle Instructions
4006 //===---------------------------------------------------------------------===//
4008 let ExeDomain = SSEPackedInt in {
4009 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
4010 def ri : Ii8<0x70, MRMSrcReg,
4011 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4012 !strconcat(OpcodeStr,
4013 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4014 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
4016 def mi : Ii8<0x70, MRMSrcMem,
4017 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4018 !strconcat(OpcodeStr,
4019 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4021 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
4026 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
4027 def Yri : Ii8<0x70, MRMSrcReg,
4028 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4029 !strconcat(OpcodeStr,
4030 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4031 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
4032 def Ymi : Ii8<0x70, MRMSrcMem,
4033 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4034 !strconcat(OpcodeStr,
4035 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4037 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
4038 (i8 imm:$src2))))]>;
4040 } // ExeDomain = SSEPackedInt
4042 let Predicates = [HasAVX] in {
4043 let AddedComplexity = 5 in
4044 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
4046 // SSE2 with ImmT == Imm8 and XS prefix.
4047 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
4049 // SSE2 with ImmT == Imm8 and XD prefix.
4050 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
4052 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4053 (VPSHUFDmi addr:$src1, imm:$imm)>;
4054 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4055 (VPSHUFDri VR128:$src1, imm:$imm)>;
4058 let Predicates = [HasAVX2] in {
4059 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>,
4060 TB, OpSize, VEX,VEX_L;
4061 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>,
4063 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>,
4067 let Predicates = [UseSSE2] in {
4068 let AddedComplexity = 5 in
4069 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
4071 // SSE2 with ImmT == Imm8 and XS prefix.
4072 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
4074 // SSE2 with ImmT == Imm8 and XD prefix.
4075 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
4077 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4078 (PSHUFDmi addr:$src1, imm:$imm)>;
4079 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4080 (PSHUFDri VR128:$src1, imm:$imm)>;
4083 //===---------------------------------------------------------------------===//
4084 // SSE2 - Packed Integer Unpack Instructions
4085 //===---------------------------------------------------------------------===//
4087 let ExeDomain = SSEPackedInt in {
4088 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4089 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4090 def rr : PDI<opc, MRMSrcReg,
4091 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4093 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4094 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4095 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4097 def rm : PDI<opc, MRMSrcMem,
4098 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4100 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4101 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4102 [(set VR128:$dst, (OpNode VR128:$src1,
4103 (bc_frag (memopv2i64
4108 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4109 SDNode OpNode, PatFrag bc_frag> {
4110 def Yrr : PDI<opc, MRMSrcReg,
4111 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4112 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4113 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4114 def Yrm : PDI<opc, MRMSrcMem,
4115 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4116 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4117 [(set VR256:$dst, (OpNode VR256:$src1,
4118 (bc_frag (memopv4i64 addr:$src2))))]>;
4121 let Predicates = [HasAVX] in {
4122 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4123 bc_v16i8, 0>, VEX_4V;
4124 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4125 bc_v8i16, 0>, VEX_4V;
4126 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4127 bc_v4i32, 0>, VEX_4V;
4128 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4129 bc_v2i64, 0>, VEX_4V;
4131 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4132 bc_v16i8, 0>, VEX_4V;
4133 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4134 bc_v8i16, 0>, VEX_4V;
4135 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4136 bc_v4i32, 0>, VEX_4V;
4137 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4138 bc_v2i64, 0>, VEX_4V;
4141 let Predicates = [HasAVX2] in {
4142 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4143 bc_v32i8>, VEX_4V, VEX_L;
4144 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4145 bc_v16i16>, VEX_4V, VEX_L;
4146 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4147 bc_v8i32>, VEX_4V, VEX_L;
4148 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4149 bc_v4i64>, VEX_4V, VEX_L;
4151 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4152 bc_v32i8>, VEX_4V, VEX_L;
4153 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4154 bc_v16i16>, VEX_4V, VEX_L;
4155 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4156 bc_v8i32>, VEX_4V, VEX_L;
4157 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4158 bc_v4i64>, VEX_4V, VEX_L;
4161 let Constraints = "$src1 = $dst" in {
4162 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4164 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4166 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4168 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4171 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4173 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4175 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4177 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4180 } // ExeDomain = SSEPackedInt
4182 //===---------------------------------------------------------------------===//
4183 // SSE2 - Packed Integer Extract and Insert
4184 //===---------------------------------------------------------------------===//
4186 let ExeDomain = SSEPackedInt in {
4187 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4188 def rri : Ii8<0xC4, MRMSrcReg,
4189 (outs VR128:$dst), (ins VR128:$src1,
4190 GR32:$src2, i32i8imm:$src3),
4192 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4193 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4195 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4196 def rmi : Ii8<0xC4, MRMSrcMem,
4197 (outs VR128:$dst), (ins VR128:$src1,
4198 i16mem:$src2, i32i8imm:$src3),
4200 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4201 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4203 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4204 imm:$src3))], IIC_SSE_PINSRW>;
4208 let Predicates = [HasAVX] in
4209 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4210 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4211 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4212 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4213 imm:$src2))]>, TB, OpSize, VEX;
4214 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4215 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4216 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4217 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4218 imm:$src2))], IIC_SSE_PEXTRW>;
4221 let Predicates = [HasAVX] in {
4222 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4223 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4224 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4225 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4226 []>, TB, OpSize, VEX_4V;
4229 let Constraints = "$src1 = $dst" in
4230 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4232 } // ExeDomain = SSEPackedInt
4234 //===---------------------------------------------------------------------===//
4235 // SSE2 - Packed Mask Creation
4236 //===---------------------------------------------------------------------===//
4238 let ExeDomain = SSEPackedInt in {
4240 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4241 "pmovmskb\t{$src, $dst|$dst, $src}",
4242 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4243 IIC_SSE_MOVMSK>, VEX;
4244 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4245 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4247 let Predicates = [HasAVX2] in {
4248 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4249 "pmovmskb\t{$src, $dst|$dst, $src}",
4250 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4251 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4252 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4255 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4256 "pmovmskb\t{$src, $dst|$dst, $src}",
4257 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4260 } // ExeDomain = SSEPackedInt
4262 //===---------------------------------------------------------------------===//
4263 // SSE2 - Conditional Store
4264 //===---------------------------------------------------------------------===//
4266 let ExeDomain = SSEPackedInt in {
4269 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4270 (ins VR128:$src, VR128:$mask),
4271 "maskmovdqu\t{$mask, $src|$src, $mask}",
4272 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4273 IIC_SSE_MASKMOV>, VEX;
4275 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4276 (ins VR128:$src, VR128:$mask),
4277 "maskmovdqu\t{$mask, $src|$src, $mask}",
4278 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4279 IIC_SSE_MASKMOV>, VEX;
4282 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4283 "maskmovdqu\t{$mask, $src|$src, $mask}",
4284 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4287 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4288 "maskmovdqu\t{$mask, $src|$src, $mask}",
4289 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4292 } // ExeDomain = SSEPackedInt
4294 //===---------------------------------------------------------------------===//
4295 // SSE2 - Move Doubleword
4296 //===---------------------------------------------------------------------===//
4298 //===---------------------------------------------------------------------===//
4299 // Move Int Doubleword to Packed Double Int
4301 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4302 "movd\t{$src, $dst|$dst, $src}",
4304 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4306 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4307 "movd\t{$src, $dst|$dst, $src}",
4309 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4312 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4313 "mov{d|q}\t{$src, $dst|$dst, $src}",
4315 (v2i64 (scalar_to_vector GR64:$src)))],
4316 IIC_SSE_MOVDQ>, VEX;
4317 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4318 "mov{d|q}\t{$src, $dst|$dst, $src}",
4319 [(set FR64:$dst, (bitconvert GR64:$src))],
4320 IIC_SSE_MOVDQ>, VEX;
4322 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4323 "movd\t{$src, $dst|$dst, $src}",
4325 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4326 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4327 "movd\t{$src, $dst|$dst, $src}",
4329 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4331 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4332 "mov{d|q}\t{$src, $dst|$dst, $src}",
4334 (v2i64 (scalar_to_vector GR64:$src)))],
4336 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4337 "mov{d|q}\t{$src, $dst|$dst, $src}",
4338 [(set FR64:$dst, (bitconvert GR64:$src))],
4341 //===---------------------------------------------------------------------===//
4342 // Move Int Doubleword to Single Scalar
4344 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4345 "movd\t{$src, $dst|$dst, $src}",
4346 [(set FR32:$dst, (bitconvert GR32:$src))],
4347 IIC_SSE_MOVDQ>, VEX;
4349 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4350 "movd\t{$src, $dst|$dst, $src}",
4351 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4354 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4355 "movd\t{$src, $dst|$dst, $src}",
4356 [(set FR32:$dst, (bitconvert GR32:$src))],
4359 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4360 "movd\t{$src, $dst|$dst, $src}",
4361 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4364 //===---------------------------------------------------------------------===//
4365 // Move Packed Doubleword Int to Packed Double Int
4367 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4368 "movd\t{$src, $dst|$dst, $src}",
4369 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4370 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4371 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4372 (ins i32mem:$dst, VR128:$src),
4373 "movd\t{$src, $dst|$dst, $src}",
4374 [(store (i32 (vector_extract (v4i32 VR128:$src),
4375 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4377 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4378 "movd\t{$src, $dst|$dst, $src}",
4379 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4380 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4381 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4382 "movd\t{$src, $dst|$dst, $src}",
4383 [(store (i32 (vector_extract (v4i32 VR128:$src),
4384 (iPTR 0))), addr:$dst)],
4387 //===---------------------------------------------------------------------===//
4388 // Move Packed Doubleword Int first element to Doubleword Int
4390 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4391 "vmov{d|q}\t{$src, $dst|$dst, $src}",
4392 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4395 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4397 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4398 "mov{d|q}\t{$src, $dst|$dst, $src}",
4399 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4403 //===---------------------------------------------------------------------===//
4404 // Bitcast FR64 <-> GR64
4406 let Predicates = [HasAVX] in
4407 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4408 "vmovq\t{$src, $dst|$dst, $src}",
4409 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4411 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4412 "mov{d|q}\t{$src, $dst|$dst, $src}",
4413 [(set GR64:$dst, (bitconvert FR64:$src))],
4414 IIC_SSE_MOVDQ>, VEX;
4415 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4416 "movq\t{$src, $dst|$dst, $src}",
4417 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4418 IIC_SSE_MOVDQ>, VEX;
4420 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4421 "movq\t{$src, $dst|$dst, $src}",
4422 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4424 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4425 "mov{d|q}\t{$src, $dst|$dst, $src}",
4426 [(set GR64:$dst, (bitconvert FR64:$src))],
4428 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4429 "movq\t{$src, $dst|$dst, $src}",
4430 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4433 //===---------------------------------------------------------------------===//
4434 // Move Scalar Single to Double Int
4436 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4437 "movd\t{$src, $dst|$dst, $src}",
4438 [(set GR32:$dst, (bitconvert FR32:$src))],
4439 IIC_SSE_MOVD_ToGP>, VEX;
4440 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4441 "movd\t{$src, $dst|$dst, $src}",
4442 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4443 IIC_SSE_MOVDQ>, VEX;
4444 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4445 "movd\t{$src, $dst|$dst, $src}",
4446 [(set GR32:$dst, (bitconvert FR32:$src))],
4448 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4449 "movd\t{$src, $dst|$dst, $src}",
4450 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4453 //===---------------------------------------------------------------------===//
4454 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4456 let AddedComplexity = 15 in {
4457 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4459 [(set VR128:$dst, (v4i32 (X86vzmovl
4460 (v4i32 (scalar_to_vector GR32:$src)))))],
4461 IIC_SSE_MOVDQ>, VEX;
4462 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4463 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4464 [(set VR128:$dst, (v2i64 (X86vzmovl
4465 (v2i64 (scalar_to_vector GR64:$src)))))],
4469 let AddedComplexity = 15 in {
4470 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4471 "movd\t{$src, $dst|$dst, $src}",
4472 [(set VR128:$dst, (v4i32 (X86vzmovl
4473 (v4i32 (scalar_to_vector GR32:$src)))))],
4475 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4476 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4477 [(set VR128:$dst, (v2i64 (X86vzmovl
4478 (v2i64 (scalar_to_vector GR64:$src)))))],
4482 let AddedComplexity = 20 in {
4483 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4484 "movd\t{$src, $dst|$dst, $src}",
4486 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4487 (loadi32 addr:$src))))))],
4488 IIC_SSE_MOVDQ>, VEX;
4489 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4490 "movd\t{$src, $dst|$dst, $src}",
4492 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4493 (loadi32 addr:$src))))))],
4497 let Predicates = [HasAVX] in {
4498 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4499 let AddedComplexity = 20 in {
4500 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4501 (VMOVZDI2PDIrm addr:$src)>;
4502 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4503 (VMOVZDI2PDIrm addr:$src)>;
4505 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4506 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4507 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4508 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4509 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4510 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4511 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4514 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4515 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4516 (MOVZDI2PDIrm addr:$src)>;
4517 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4518 (MOVZDI2PDIrm addr:$src)>;
4521 // These are the correct encodings of the instructions so that we know how to
4522 // read correct assembly, even though we continue to emit the wrong ones for
4523 // compatibility with Darwin's buggy assembler.
4524 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4525 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4526 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4527 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4528 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4529 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4530 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4531 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4532 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4533 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4534 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4535 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4537 //===---------------------------------------------------------------------===//
4538 // SSE2 - Move Quadword
4539 //===---------------------------------------------------------------------===//
4541 //===---------------------------------------------------------------------===//
4542 // Move Quadword Int to Packed Quadword Int
4544 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4545 "vmovq\t{$src, $dst|$dst, $src}",
4547 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4548 VEX, Requires<[HasAVX]>;
4549 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4550 "movq\t{$src, $dst|$dst, $src}",
4552 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4554 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4556 //===---------------------------------------------------------------------===//
4557 // Move Packed Quadword Int to Quadword Int
4559 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4560 "movq\t{$src, $dst|$dst, $src}",
4561 [(store (i64 (vector_extract (v2i64 VR128:$src),
4562 (iPTR 0))), addr:$dst)],
4563 IIC_SSE_MOVDQ>, VEX;
4564 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4565 "movq\t{$src, $dst|$dst, $src}",
4566 [(store (i64 (vector_extract (v2i64 VR128:$src),
4567 (iPTR 0))), addr:$dst)],
4570 //===---------------------------------------------------------------------===//
4571 // Store / copy lower 64-bits of a XMM register.
4573 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4574 "movq\t{$src, $dst|$dst, $src}",
4575 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4576 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4577 "movq\t{$src, $dst|$dst, $src}",
4578 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4581 let AddedComplexity = 20 in
4582 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4583 "vmovq\t{$src, $dst|$dst, $src}",
4585 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4586 (loadi64 addr:$src))))))],
4588 XS, VEX, Requires<[HasAVX]>;
4590 let AddedComplexity = 20 in
4591 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4592 "movq\t{$src, $dst|$dst, $src}",
4594 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4595 (loadi64 addr:$src))))))],
4597 XS, Requires<[UseSSE2]>;
4599 let Predicates = [HasAVX], AddedComplexity = 20 in {
4600 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4601 (VMOVZQI2PQIrm addr:$src)>;
4602 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4603 (VMOVZQI2PQIrm addr:$src)>;
4604 def : Pat<(v2i64 (X86vzload addr:$src)),
4605 (VMOVZQI2PQIrm addr:$src)>;
4608 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4609 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4610 (MOVZQI2PQIrm addr:$src)>;
4611 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4612 (MOVZQI2PQIrm addr:$src)>;
4613 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4616 let Predicates = [HasAVX] in {
4617 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4618 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4619 def : Pat<(v4i64 (X86vzload addr:$src)),
4620 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4623 //===---------------------------------------------------------------------===//
4624 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4625 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4627 let AddedComplexity = 15 in
4628 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4629 "vmovq\t{$src, $dst|$dst, $src}",
4630 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4632 XS, VEX, Requires<[HasAVX]>;
4633 let AddedComplexity = 15 in
4634 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4635 "movq\t{$src, $dst|$dst, $src}",
4636 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4638 XS, Requires<[UseSSE2]>;
4640 let AddedComplexity = 20 in
4641 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4642 "vmovq\t{$src, $dst|$dst, $src}",
4643 [(set VR128:$dst, (v2i64 (X86vzmovl
4644 (loadv2i64 addr:$src))))],
4646 XS, VEX, Requires<[HasAVX]>;
4647 let AddedComplexity = 20 in {
4648 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4649 "movq\t{$src, $dst|$dst, $src}",
4650 [(set VR128:$dst, (v2i64 (X86vzmovl
4651 (loadv2i64 addr:$src))))],
4653 XS, Requires<[UseSSE2]>;
4656 let AddedComplexity = 20 in {
4657 let Predicates = [HasAVX] in {
4658 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4659 (VMOVZPQILo2PQIrm addr:$src)>;
4660 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4661 (VMOVZPQILo2PQIrr VR128:$src)>;
4663 let Predicates = [UseSSE2] in {
4664 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4665 (MOVZPQILo2PQIrm addr:$src)>;
4666 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4667 (MOVZPQILo2PQIrr VR128:$src)>;
4671 // Instructions to match in the assembler
4672 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4673 "movq\t{$src, $dst|$dst, $src}", [],
4674 IIC_SSE_MOVDQ>, VEX, VEX_W;
4675 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4676 "movq\t{$src, $dst|$dst, $src}", [],
4677 IIC_SSE_MOVDQ>, VEX, VEX_W;
4678 // Recognize "movd" with GR64 destination, but encode as a "movq"
4679 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4680 "movd\t{$src, $dst|$dst, $src}", [],
4681 IIC_SSE_MOVDQ>, VEX, VEX_W;
4683 // Instructions for the disassembler
4684 // xr = XMM register
4687 let Predicates = [HasAVX] in
4688 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4689 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4690 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4691 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4693 //===---------------------------------------------------------------------===//
4694 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4695 //===---------------------------------------------------------------------===//
4696 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4697 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4698 X86MemOperand x86memop> {
4699 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4700 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4701 [(set RC:$dst, (vt (OpNode RC:$src)))],
4703 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4704 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4705 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4709 let Predicates = [HasAVX] in {
4710 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4711 v4f32, VR128, memopv4f32, f128mem>, VEX;
4712 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4713 v4f32, VR128, memopv4f32, f128mem>, VEX;
4714 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4715 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4716 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4717 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4719 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4720 memopv4f32, f128mem>;
4721 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4722 memopv4f32, f128mem>;
4724 let Predicates = [HasAVX] in {
4725 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4726 (VMOVSHDUPrr VR128:$src)>;
4727 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4728 (VMOVSHDUPrm addr:$src)>;
4729 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4730 (VMOVSLDUPrr VR128:$src)>;
4731 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4732 (VMOVSLDUPrm addr:$src)>;
4733 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4734 (VMOVSHDUPYrr VR256:$src)>;
4735 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4736 (VMOVSHDUPYrm addr:$src)>;
4737 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4738 (VMOVSLDUPYrr VR256:$src)>;
4739 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4740 (VMOVSLDUPYrm addr:$src)>;
4743 let Predicates = [UseSSE3] in {
4744 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4745 (MOVSHDUPrr VR128:$src)>;
4746 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4747 (MOVSHDUPrm addr:$src)>;
4748 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4749 (MOVSLDUPrr VR128:$src)>;
4750 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4751 (MOVSLDUPrm addr:$src)>;
4754 //===---------------------------------------------------------------------===//
4755 // SSE3 - Replicate Double FP - MOVDDUP
4756 //===---------------------------------------------------------------------===//
4758 multiclass sse3_replicate_dfp<string OpcodeStr> {
4759 let neverHasSideEffects = 1 in
4760 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4761 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4762 [], IIC_SSE_MOV_LH>;
4763 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4764 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4767 (scalar_to_vector (loadf64 addr:$src)))))],
4771 // FIXME: Merge with above classe when there're patterns for the ymm version
4772 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4773 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4774 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4775 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4776 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4777 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4780 (scalar_to_vector (loadf64 addr:$src)))))]>;
4783 let Predicates = [HasAVX] in {
4784 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4785 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4788 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4790 let Predicates = [HasAVX] in {
4791 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4792 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4793 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4794 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4795 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4796 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4797 def : Pat<(X86Movddup (bc_v2f64
4798 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4799 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4802 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4803 (VMOVDDUPYrm addr:$src)>;
4804 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4805 (VMOVDDUPYrm addr:$src)>;
4806 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4807 (VMOVDDUPYrm addr:$src)>;
4808 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4809 (VMOVDDUPYrr VR256:$src)>;
4812 let Predicates = [UseSSE3] in {
4813 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4814 (MOVDDUPrm addr:$src)>;
4815 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4816 (MOVDDUPrm addr:$src)>;
4817 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4818 (MOVDDUPrm addr:$src)>;
4819 def : Pat<(X86Movddup (bc_v2f64
4820 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4821 (MOVDDUPrm addr:$src)>;
4824 //===---------------------------------------------------------------------===//
4825 // SSE3 - Move Unaligned Integer
4826 //===---------------------------------------------------------------------===//
4828 let Predicates = [HasAVX] in {
4829 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4830 "vlddqu\t{$src, $dst|$dst, $src}",
4831 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4832 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4833 "vlddqu\t{$src, $dst|$dst, $src}",
4834 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4837 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4838 "lddqu\t{$src, $dst|$dst, $src}",
4839 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4842 //===---------------------------------------------------------------------===//
4843 // SSE3 - Arithmetic
4844 //===---------------------------------------------------------------------===//
4846 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4847 X86MemOperand x86memop, OpndItins itins,
4849 def rr : I<0xD0, MRMSrcReg,
4850 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4852 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4853 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4854 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4855 def rm : I<0xD0, MRMSrcMem,
4856 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4858 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4859 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4860 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4863 let Predicates = [HasAVX] in {
4864 let ExeDomain = SSEPackedSingle in {
4865 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4866 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4867 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4868 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4870 let ExeDomain = SSEPackedDouble in {
4871 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4872 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4873 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4874 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4877 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4878 let ExeDomain = SSEPackedSingle in
4879 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4880 f128mem, SSE_ALU_F32P>, TB, XD;
4881 let ExeDomain = SSEPackedDouble in
4882 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4883 f128mem, SSE_ALU_F64P>, TB, OpSize;
4886 //===---------------------------------------------------------------------===//
4887 // SSE3 Instructions
4888 //===---------------------------------------------------------------------===//
4891 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4892 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4893 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4895 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4896 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4897 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4899 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4901 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4902 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4903 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4904 IIC_SSE_HADDSUB_RM>;
4906 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4907 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4908 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4910 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4911 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4912 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4914 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4916 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4917 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4918 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4919 IIC_SSE_HADDSUB_RM>;
4922 let Predicates = [HasAVX] in {
4923 let ExeDomain = SSEPackedSingle in {
4924 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4925 X86fhadd, 0>, VEX_4V;
4926 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4927 X86fhsub, 0>, VEX_4V;
4928 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4929 X86fhadd, 0>, VEX_4V, VEX_L;
4930 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4931 X86fhsub, 0>, VEX_4V, VEX_L;
4933 let ExeDomain = SSEPackedDouble in {
4934 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4935 X86fhadd, 0>, VEX_4V;
4936 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4937 X86fhsub, 0>, VEX_4V;
4938 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4939 X86fhadd, 0>, VEX_4V, VEX_L;
4940 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4941 X86fhsub, 0>, VEX_4V, VEX_L;
4945 let Constraints = "$src1 = $dst" in {
4946 let ExeDomain = SSEPackedSingle in {
4947 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4948 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4950 let ExeDomain = SSEPackedDouble in {
4951 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4952 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4956 //===---------------------------------------------------------------------===//
4957 // SSSE3 - Packed Absolute Instructions
4958 //===---------------------------------------------------------------------===//
4961 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4962 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4963 Intrinsic IntId128> {
4964 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4966 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4967 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
4970 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4972 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4975 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
4979 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4980 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4981 Intrinsic IntId256> {
4982 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4984 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4985 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4988 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4990 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4993 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4996 let Predicates = [HasAVX] in {
4997 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4998 int_x86_ssse3_pabs_b_128>, VEX;
4999 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5000 int_x86_ssse3_pabs_w_128>, VEX;
5001 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5002 int_x86_ssse3_pabs_d_128>, VEX;
5005 let Predicates = [HasAVX2] in {
5006 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5007 int_x86_avx2_pabs_b>, VEX, VEX_L;
5008 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5009 int_x86_avx2_pabs_w>, VEX, VEX_L;
5010 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5011 int_x86_avx2_pabs_d>, VEX, VEX_L;
5014 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5015 int_x86_ssse3_pabs_b_128>;
5016 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5017 int_x86_ssse3_pabs_w_128>;
5018 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5019 int_x86_ssse3_pabs_d_128>;
5021 //===---------------------------------------------------------------------===//
5022 // SSSE3 - Packed Binary Operator Instructions
5023 //===---------------------------------------------------------------------===//
5025 def SSE_PHADDSUBD : OpndItins<
5026 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5028 def SSE_PHADDSUBSW : OpndItins<
5029 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5031 def SSE_PHADDSUBW : OpndItins<
5032 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5034 def SSE_PSHUFB : OpndItins<
5035 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5037 def SSE_PSIGN : OpndItins<
5038 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5040 def SSE_PMULHRSW : OpndItins<
5041 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5044 /// SS3I_binop_rm - Simple SSSE3 bin op
5045 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5046 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5047 X86MemOperand x86memop, OpndItins itins,
5049 let isCommutable = 1 in
5050 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5051 (ins RC:$src1, RC:$src2),
5053 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5054 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5055 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5057 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5058 (ins RC:$src1, x86memop:$src2),
5060 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5061 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5063 (OpVT (OpNode RC:$src1,
5064 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5067 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5068 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5069 Intrinsic IntId128, OpndItins itins,
5071 let isCommutable = 1 in
5072 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5073 (ins VR128:$src1, VR128:$src2),
5075 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5076 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5077 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5079 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5080 (ins VR128:$src1, i128mem:$src2),
5082 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5083 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5085 (IntId128 VR128:$src1,
5086 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5089 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5090 Intrinsic IntId256> {
5091 let isCommutable = 1 in
5092 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5093 (ins VR256:$src1, VR256:$src2),
5094 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5095 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5097 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5098 (ins VR256:$src1, i256mem:$src2),
5099 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5101 (IntId256 VR256:$src1,
5102 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5105 let ImmT = NoImm, Predicates = [HasAVX] in {
5106 let isCommutable = 0 in {
5107 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5108 memopv2i64, i128mem,
5109 SSE_PHADDSUBW, 0>, VEX_4V;
5110 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5111 memopv2i64, i128mem,
5112 SSE_PHADDSUBD, 0>, VEX_4V;
5113 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5114 memopv2i64, i128mem,
5115 SSE_PHADDSUBW, 0>, VEX_4V;
5116 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5117 memopv2i64, i128mem,
5118 SSE_PHADDSUBD, 0>, VEX_4V;
5119 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5120 memopv2i64, i128mem,
5121 SSE_PSIGN, 0>, VEX_4V;
5122 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5123 memopv2i64, i128mem,
5124 SSE_PSIGN, 0>, VEX_4V;
5125 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5126 memopv2i64, i128mem,
5127 SSE_PSIGN, 0>, VEX_4V;
5128 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5129 memopv2i64, i128mem,
5130 SSE_PSHUFB, 0>, VEX_4V;
5131 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5132 int_x86_ssse3_phadd_sw_128,
5133 SSE_PHADDSUBSW, 0>, VEX_4V;
5134 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5135 int_x86_ssse3_phsub_sw_128,
5136 SSE_PHADDSUBSW, 0>, VEX_4V;
5137 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5138 int_x86_ssse3_pmadd_ub_sw_128,
5139 SSE_PMADD, 0>, VEX_4V;
5141 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5142 int_x86_ssse3_pmul_hr_sw_128,
5143 SSE_PMULHRSW, 0>, VEX_4V;
5146 let ImmT = NoImm, Predicates = [HasAVX2] in {
5147 let isCommutable = 0 in {
5148 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5149 memopv4i64, i256mem,
5150 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5151 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5152 memopv4i64, i256mem,
5153 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5154 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5155 memopv4i64, i256mem,
5156 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5157 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5158 memopv4i64, i256mem,
5159 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5160 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5161 memopv4i64, i256mem,
5162 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5163 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5164 memopv4i64, i256mem,
5165 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5166 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5167 memopv4i64, i256mem,
5168 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5169 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5170 memopv4i64, i256mem,
5171 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5172 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5173 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5174 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5175 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5176 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5177 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5179 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5180 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5183 // None of these have i8 immediate fields.
5184 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5185 let isCommutable = 0 in {
5186 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5187 memopv2i64, i128mem, SSE_PHADDSUBW>;
5188 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5189 memopv2i64, i128mem, SSE_PHADDSUBD>;
5190 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5191 memopv2i64, i128mem, SSE_PHADDSUBW>;
5192 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5193 memopv2i64, i128mem, SSE_PHADDSUBD>;
5194 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5195 memopv2i64, i128mem, SSE_PSIGN>;
5196 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5197 memopv2i64, i128mem, SSE_PSIGN>;
5198 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5199 memopv2i64, i128mem, SSE_PSIGN>;
5200 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5201 memopv2i64, i128mem, SSE_PSHUFB>;
5202 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5203 int_x86_ssse3_phadd_sw_128,
5205 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5206 int_x86_ssse3_phsub_sw_128,
5208 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5209 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5211 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5212 int_x86_ssse3_pmul_hr_sw_128,
5216 //===---------------------------------------------------------------------===//
5217 // SSSE3 - Packed Align Instruction Patterns
5218 //===---------------------------------------------------------------------===//
5220 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5221 let neverHasSideEffects = 1 in {
5222 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5223 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5225 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5227 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5228 [], IIC_SSE_PALIGNR>, OpSize;
5230 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5231 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5233 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5235 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5236 [], IIC_SSE_PALIGNR>, OpSize;
5240 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5241 let neverHasSideEffects = 1 in {
5242 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5243 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5245 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5248 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5249 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5251 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5256 let Predicates = [HasAVX] in
5257 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5258 let Predicates = [HasAVX2] in
5259 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
5260 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5261 defm PALIGN : ssse3_palign<"palignr">;
5263 let Predicates = [HasAVX2] in {
5264 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5265 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5266 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5267 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5268 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5269 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5270 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5271 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5274 let Predicates = [HasAVX] in {
5275 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5276 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5277 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5278 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5279 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5280 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5281 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5282 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5285 let Predicates = [UseSSSE3] in {
5286 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5287 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5288 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5289 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5290 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5291 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5292 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5293 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5296 //===---------------------------------------------------------------------===//
5297 // SSSE3 - Thread synchronization
5298 //===---------------------------------------------------------------------===//
5300 let usesCustomInserter = 1 in {
5301 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5302 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5303 Requires<[HasSSE3]>;
5306 let Uses = [EAX, ECX, EDX] in
5307 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5308 TB, Requires<[HasSSE3]>;
5309 let Uses = [ECX, EAX] in
5310 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5311 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5312 TB, Requires<[HasSSE3]>;
5314 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5315 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5317 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5318 Requires<[In32BitMode]>;
5319 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5320 Requires<[In64BitMode]>;
5322 //===----------------------------------------------------------------------===//
5323 // SSE4.1 - Packed Move with Sign/Zero Extend
5324 //===----------------------------------------------------------------------===//
5326 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5327 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5328 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5329 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5331 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5332 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5334 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5338 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5340 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5341 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5342 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5344 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5345 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5346 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5349 let Predicates = [HasAVX] in {
5350 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5352 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5354 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5356 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5358 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5360 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5364 let Predicates = [HasAVX2] in {
5365 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5366 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5367 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5368 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5369 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5370 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5371 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5372 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5373 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5374 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5375 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5376 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5379 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5380 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5381 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5382 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5383 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5384 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5386 let Predicates = [HasAVX] in {
5387 // Common patterns involving scalar load.
5388 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5389 (VPMOVSXBWrm addr:$src)>;
5390 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5391 (VPMOVSXBWrm addr:$src)>;
5392 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5393 (VPMOVSXBWrm addr:$src)>;
5395 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5396 (VPMOVSXWDrm addr:$src)>;
5397 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5398 (VPMOVSXWDrm addr:$src)>;
5399 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5400 (VPMOVSXWDrm addr:$src)>;
5402 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5403 (VPMOVSXDQrm addr:$src)>;
5404 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5405 (VPMOVSXDQrm addr:$src)>;
5406 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5407 (VPMOVSXDQrm addr:$src)>;
5409 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5410 (VPMOVZXBWrm addr:$src)>;
5411 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5412 (VPMOVZXBWrm addr:$src)>;
5413 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5414 (VPMOVZXBWrm addr:$src)>;
5416 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5417 (VPMOVZXWDrm addr:$src)>;
5418 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5419 (VPMOVZXWDrm addr:$src)>;
5420 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5421 (VPMOVZXWDrm addr:$src)>;
5423 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5424 (VPMOVZXDQrm addr:$src)>;
5425 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5426 (VPMOVZXDQrm addr:$src)>;
5427 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5428 (VPMOVZXDQrm addr:$src)>;
5431 let Predicates = [UseSSE41] in {
5432 // Common patterns involving scalar load.
5433 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5434 (PMOVSXBWrm addr:$src)>;
5435 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5436 (PMOVSXBWrm addr:$src)>;
5437 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5438 (PMOVSXBWrm addr:$src)>;
5440 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5441 (PMOVSXWDrm addr:$src)>;
5442 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5443 (PMOVSXWDrm addr:$src)>;
5444 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5445 (PMOVSXWDrm addr:$src)>;
5447 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5448 (PMOVSXDQrm addr:$src)>;
5449 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5450 (PMOVSXDQrm addr:$src)>;
5451 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5452 (PMOVSXDQrm addr:$src)>;
5454 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5455 (PMOVZXBWrm addr:$src)>;
5456 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5457 (PMOVZXBWrm addr:$src)>;
5458 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5459 (PMOVZXBWrm addr:$src)>;
5461 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5462 (PMOVZXWDrm addr:$src)>;
5463 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5464 (PMOVZXWDrm addr:$src)>;
5465 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5466 (PMOVZXWDrm addr:$src)>;
5468 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5469 (PMOVZXDQrm addr:$src)>;
5470 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5471 (PMOVZXDQrm addr:$src)>;
5472 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5473 (PMOVZXDQrm addr:$src)>;
5476 let Predicates = [HasAVX2] in {
5477 let AddedComplexity = 15 in {
5478 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5479 (VPMOVZXDQYrr VR128:$src)>;
5480 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5481 (VPMOVZXWDYrr VR128:$src)>;
5484 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5485 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5488 let Predicates = [HasAVX] in {
5489 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5490 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5493 let Predicates = [UseSSE41] in {
5494 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5495 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5499 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5500 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5501 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5502 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5504 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5505 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5507 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5511 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5513 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5514 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5515 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5517 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5518 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5520 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5524 let Predicates = [HasAVX] in {
5525 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5527 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5529 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5531 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5535 let Predicates = [HasAVX2] in {
5536 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5537 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5538 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5539 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5540 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5541 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5542 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5543 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5546 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5547 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5548 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5549 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5551 let Predicates = [HasAVX] in {
5552 // Common patterns involving scalar load
5553 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5554 (VPMOVSXBDrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5556 (VPMOVSXWQrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5559 (VPMOVZXBDrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5561 (VPMOVZXWQrm addr:$src)>;
5564 let Predicates = [UseSSE41] in {
5565 // Common patterns involving scalar load
5566 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5567 (PMOVSXBDrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5569 (PMOVSXWQrm addr:$src)>;
5571 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5572 (PMOVZXBDrm addr:$src)>;
5573 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5574 (PMOVZXWQrm addr:$src)>;
5577 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5578 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5579 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5580 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5582 // Expecting a i16 load any extended to i32 value.
5583 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5584 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5585 [(set VR128:$dst, (IntId (bitconvert
5586 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5590 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5592 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5593 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5594 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5596 // Expecting a i16 load any extended to i32 value.
5597 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5598 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5599 [(set VR256:$dst, (IntId (bitconvert
5600 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5604 let Predicates = [HasAVX] in {
5605 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5607 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5610 let Predicates = [HasAVX2] in {
5611 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5612 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5613 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5614 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5616 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5617 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5619 let Predicates = [HasAVX2] in {
5620 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5621 (VPMOVSXWDYrm addr:$src)>;
5622 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5623 (VPMOVSXDQYrm addr:$src)>;
5625 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5626 (scalar_to_vector (loadi64 addr:$src))))))),
5627 (VPMOVSXBDYrm addr:$src)>;
5628 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5629 (scalar_to_vector (loadf64 addr:$src))))))),
5630 (VPMOVSXBDYrm addr:$src)>;
5632 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5633 (scalar_to_vector (loadi64 addr:$src))))))),
5634 (VPMOVSXWQYrm addr:$src)>;
5635 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5636 (scalar_to_vector (loadf64 addr:$src))))))),
5637 (VPMOVSXWQYrm addr:$src)>;
5639 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5640 (scalar_to_vector (loadi32 addr:$src))))))),
5641 (VPMOVSXBQYrm addr:$src)>;
5644 let Predicates = [HasAVX] in {
5645 // Common patterns involving scalar load
5646 def : Pat<(int_x86_sse41_pmovsxbq
5647 (bitconvert (v4i32 (X86vzmovl
5648 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5649 (VPMOVSXBQrm addr:$src)>;
5651 def : Pat<(int_x86_sse41_pmovzxbq
5652 (bitconvert (v4i32 (X86vzmovl
5653 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5654 (VPMOVZXBQrm addr:$src)>;
5657 let Predicates = [UseSSE41] in {
5658 // Common patterns involving scalar load
5659 def : Pat<(int_x86_sse41_pmovsxbq
5660 (bitconvert (v4i32 (X86vzmovl
5661 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5662 (PMOVSXBQrm addr:$src)>;
5664 def : Pat<(int_x86_sse41_pmovzxbq
5665 (bitconvert (v4i32 (X86vzmovl
5666 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5667 (PMOVZXBQrm addr:$src)>;
5669 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5670 (scalar_to_vector (loadi64 addr:$src))))))),
5671 (PMOVSXWDrm addr:$src)>;
5672 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5673 (scalar_to_vector (loadf64 addr:$src))))))),
5674 (PMOVSXWDrm addr:$src)>;
5675 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5676 (scalar_to_vector (loadi32 addr:$src))))))),
5677 (PMOVSXBDrm addr:$src)>;
5678 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5679 (scalar_to_vector (loadi32 addr:$src))))))),
5680 (PMOVSXWQrm addr:$src)>;
5681 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5682 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5683 (PMOVSXBQrm addr:$src)>;
5684 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5685 (scalar_to_vector (loadi64 addr:$src))))))),
5686 (PMOVSXDQrm addr:$src)>;
5687 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5688 (scalar_to_vector (loadf64 addr:$src))))))),
5689 (PMOVSXDQrm addr:$src)>;
5690 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5691 (scalar_to_vector (loadi64 addr:$src))))))),
5692 (PMOVSXBWrm addr:$src)>;
5693 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5694 (scalar_to_vector (loadf64 addr:$src))))))),
5695 (PMOVSXBWrm addr:$src)>;
5698 let Predicates = [HasAVX2] in {
5699 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5700 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5701 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5703 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5704 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5706 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5708 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5709 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5710 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5711 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5712 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5713 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5715 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5716 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5717 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5718 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5720 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5721 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5724 let Predicates = [HasAVX] in {
5725 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5726 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5727 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5729 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5730 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5732 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5734 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5735 (VPMOVZXBWrm addr:$src)>;
5736 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5737 (VPMOVZXBWrm addr:$src)>;
5738 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5739 (VPMOVZXBDrm addr:$src)>;
5740 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5741 (VPMOVZXBQrm addr:$src)>;
5743 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5744 (VPMOVZXWDrm addr:$src)>;
5745 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5746 (VPMOVZXWDrm addr:$src)>;
5747 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5748 (VPMOVZXWQrm addr:$src)>;
5750 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5751 (VPMOVZXDQrm addr:$src)>;
5752 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5753 (VPMOVZXDQrm addr:$src)>;
5754 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5755 (VPMOVZXDQrm addr:$src)>;
5757 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5758 (scalar_to_vector (loadi64 addr:$src))))))),
5759 (VPMOVSXWDrm addr:$src)>;
5760 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5761 (scalar_to_vector (loadi64 addr:$src))))))),
5762 (VPMOVSXDQrm addr:$src)>;
5763 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5764 (scalar_to_vector (loadf64 addr:$src))))))),
5765 (VPMOVSXWDrm addr:$src)>;
5766 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5767 (scalar_to_vector (loadf64 addr:$src))))))),
5768 (VPMOVSXDQrm addr:$src)>;
5769 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5770 (scalar_to_vector (loadi64 addr:$src))))))),
5771 (VPMOVSXBWrm addr:$src)>;
5772 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5773 (scalar_to_vector (loadf64 addr:$src))))))),
5774 (VPMOVSXBWrm addr:$src)>;
5776 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5777 (scalar_to_vector (loadi32 addr:$src))))))),
5778 (VPMOVSXBDrm addr:$src)>;
5779 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5780 (scalar_to_vector (loadi32 addr:$src))))))),
5781 (VPMOVSXWQrm addr:$src)>;
5782 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5783 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5784 (VPMOVSXBQrm addr:$src)>;
5787 let Predicates = [UseSSE41] in {
5788 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5789 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5790 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5792 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5793 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5795 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5797 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5798 (PMOVZXBWrm addr:$src)>;
5799 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5800 (PMOVZXBWrm addr:$src)>;
5801 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5802 (PMOVZXBDrm addr:$src)>;
5803 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5804 (PMOVZXBQrm addr:$src)>;
5806 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5807 (PMOVZXWDrm addr:$src)>;
5808 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5809 (PMOVZXWDrm addr:$src)>;
5810 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5811 (PMOVZXWQrm addr:$src)>;
5813 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5814 (PMOVZXDQrm addr:$src)>;
5815 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5816 (PMOVZXDQrm addr:$src)>;
5817 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5818 (PMOVZXDQrm addr:$src)>;
5821 //===----------------------------------------------------------------------===//
5822 // SSE4.1 - Extract Instructions
5823 //===----------------------------------------------------------------------===//
5825 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5826 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5827 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5828 (ins VR128:$src1, i32i8imm:$src2),
5829 !strconcat(OpcodeStr,
5830 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5831 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5833 let neverHasSideEffects = 1, mayStore = 1 in
5834 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5835 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5836 !strconcat(OpcodeStr,
5837 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5840 // There's an AssertZext in the way of writing the store pattern
5841 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5844 let Predicates = [HasAVX] in {
5845 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5846 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5847 (ins VR128:$src1, i32i8imm:$src2),
5848 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5851 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5854 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5855 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5856 let neverHasSideEffects = 1, mayStore = 1 in
5857 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5858 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5859 !strconcat(OpcodeStr,
5860 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5863 // There's an AssertZext in the way of writing the store pattern
5864 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5867 let Predicates = [HasAVX] in
5868 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5870 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5873 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5874 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5875 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5876 (ins VR128:$src1, i32i8imm:$src2),
5877 !strconcat(OpcodeStr,
5878 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5880 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5881 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5882 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5883 !strconcat(OpcodeStr,
5884 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5885 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5886 addr:$dst)]>, OpSize;
5889 let Predicates = [HasAVX] in
5890 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5892 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5894 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5895 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5896 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5897 (ins VR128:$src1, i32i8imm:$src2),
5898 !strconcat(OpcodeStr,
5899 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5901 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5902 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5903 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5904 !strconcat(OpcodeStr,
5905 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5906 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5907 addr:$dst)]>, OpSize, REX_W;
5910 let Predicates = [HasAVX] in
5911 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5913 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5915 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5917 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5918 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5919 (ins VR128:$src1, i32i8imm:$src2),
5920 !strconcat(OpcodeStr,
5921 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5923 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5925 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5926 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5927 !strconcat(OpcodeStr,
5928 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5929 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5930 addr:$dst)]>, OpSize;
5933 let ExeDomain = SSEPackedSingle in {
5934 let Predicates = [HasAVX] in {
5935 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5936 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5937 (ins VR128:$src1, i32i8imm:$src2),
5938 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5941 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5944 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5945 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5948 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5950 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5953 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5954 Requires<[UseSSE41]>;
5956 //===----------------------------------------------------------------------===//
5957 // SSE4.1 - Insert Instructions
5958 //===----------------------------------------------------------------------===//
5960 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5961 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5962 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5964 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5966 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5968 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5969 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5970 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5972 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5974 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5976 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5977 imm:$src3))]>, OpSize;
5980 let Predicates = [HasAVX] in
5981 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5982 let Constraints = "$src1 = $dst" in
5983 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5985 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5986 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5987 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5989 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5991 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5993 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5995 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5996 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5998 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6000 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6002 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6003 imm:$src3)))]>, OpSize;
6006 let Predicates = [HasAVX] in
6007 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6008 let Constraints = "$src1 = $dst" in
6009 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6011 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6012 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6013 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6015 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6017 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6019 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6021 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6022 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6024 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6026 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6028 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6029 imm:$src3)))]>, OpSize;
6032 let Predicates = [HasAVX] in
6033 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6034 let Constraints = "$src1 = $dst" in
6035 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6037 // insertps has a few different modes, there's the first two here below which
6038 // are optimized inserts that won't zero arbitrary elements in the destination
6039 // vector. The next one matches the intrinsic and could zero arbitrary elements
6040 // in the target vector.
6041 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6042 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6043 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6045 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6047 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6049 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6051 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6052 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6054 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6056 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6058 (X86insrtps VR128:$src1,
6059 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6060 imm:$src3))]>, OpSize;
6063 let ExeDomain = SSEPackedSingle in {
6064 let Predicates = [HasAVX] in
6065 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6066 let Constraints = "$src1 = $dst" in
6067 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6070 //===----------------------------------------------------------------------===//
6071 // SSE4.1 - Round Instructions
6072 //===----------------------------------------------------------------------===//
6074 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6075 X86MemOperand x86memop, RegisterClass RC,
6076 PatFrag mem_frag32, PatFrag mem_frag64,
6077 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6078 let ExeDomain = SSEPackedSingle in {
6079 // Intrinsic operation, reg.
6080 // Vector intrinsic operation, reg
6081 def PSr : SS4AIi8<opcps, MRMSrcReg,
6082 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6083 !strconcat(OpcodeStr,
6084 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6085 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6088 // Vector intrinsic operation, mem
6089 def PSm : SS4AIi8<opcps, MRMSrcMem,
6090 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6091 !strconcat(OpcodeStr,
6092 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6094 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6096 } // ExeDomain = SSEPackedSingle
6098 let ExeDomain = SSEPackedDouble in {
6099 // Vector intrinsic operation, reg
6100 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6101 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6102 !strconcat(OpcodeStr,
6103 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6104 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6107 // Vector intrinsic operation, mem
6108 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6109 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6110 !strconcat(OpcodeStr,
6111 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6113 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6115 } // ExeDomain = SSEPackedDouble
6118 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6121 Intrinsic F64Int, bit Is2Addr = 1> {
6122 let ExeDomain = GenericDomain in {
6124 def SSr : SS4AIi8<opcss, MRMSrcReg,
6125 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6127 !strconcat(OpcodeStr,
6128 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6129 !strconcat(OpcodeStr,
6130 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6133 // Intrinsic operation, reg.
6134 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6135 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6137 !strconcat(OpcodeStr,
6138 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6139 !strconcat(OpcodeStr,
6140 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6141 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6144 // Intrinsic operation, mem.
6145 def SSm : SS4AIi8<opcss, MRMSrcMem,
6146 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6148 !strconcat(OpcodeStr,
6149 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6150 !strconcat(OpcodeStr,
6151 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6153 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6157 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6158 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6160 !strconcat(OpcodeStr,
6161 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6162 !strconcat(OpcodeStr,
6163 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6166 // Intrinsic operation, reg.
6167 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6168 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6170 !strconcat(OpcodeStr,
6171 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6172 !strconcat(OpcodeStr,
6173 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6174 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6177 // Intrinsic operation, mem.
6178 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6179 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6181 !strconcat(OpcodeStr,
6182 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6183 !strconcat(OpcodeStr,
6184 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6186 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6188 } // ExeDomain = GenericDomain
6191 // FP round - roundss, roundps, roundsd, roundpd
6192 let Predicates = [HasAVX] in {
6194 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6195 memopv4f32, memopv2f64,
6196 int_x86_sse41_round_ps,
6197 int_x86_sse41_round_pd>, VEX;
6198 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6199 memopv8f32, memopv4f64,
6200 int_x86_avx_round_ps_256,
6201 int_x86_avx_round_pd_256>, VEX, VEX_L;
6202 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6203 int_x86_sse41_round_ss,
6204 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6206 def : Pat<(ffloor FR32:$src),
6207 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6208 def : Pat<(f64 (ffloor FR64:$src)),
6209 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6210 def : Pat<(f32 (fnearbyint FR32:$src)),
6211 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6212 def : Pat<(f64 (fnearbyint FR64:$src)),
6213 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6214 def : Pat<(f32 (fceil FR32:$src)),
6215 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6216 def : Pat<(f64 (fceil FR64:$src)),
6217 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6218 def : Pat<(f32 (frint FR32:$src)),
6219 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6220 def : Pat<(f64 (frint FR64:$src)),
6221 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6222 def : Pat<(f32 (ftrunc FR32:$src)),
6223 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6224 def : Pat<(f64 (ftrunc FR64:$src)),
6225 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6227 def : Pat<(v4f32 (ffloor VR128:$src)),
6228 (VROUNDPSr VR128:$src, (i32 0x1))>;
6229 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6230 (VROUNDPSr VR128:$src, (i32 0xC))>;
6231 def : Pat<(v4f32 (fceil VR128:$src)),
6232 (VROUNDPSr VR128:$src, (i32 0x2))>;
6233 def : Pat<(v4f32 (frint VR128:$src)),
6234 (VROUNDPSr VR128:$src, (i32 0x4))>;
6235 def : Pat<(v4f32 (ftrunc VR128:$src)),
6236 (VROUNDPSr VR128:$src, (i32 0x3))>;
6238 def : Pat<(v2f64 (ffloor VR128:$src)),
6239 (VROUNDPDr VR128:$src, (i32 0x1))>;
6240 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6241 (VROUNDPDr VR128:$src, (i32 0xC))>;
6242 def : Pat<(v2f64 (fceil VR128:$src)),
6243 (VROUNDPDr VR128:$src, (i32 0x2))>;
6244 def : Pat<(v2f64 (frint VR128:$src)),
6245 (VROUNDPDr VR128:$src, (i32 0x4))>;
6246 def : Pat<(v2f64 (ftrunc VR128:$src)),
6247 (VROUNDPDr VR128:$src, (i32 0x3))>;
6249 def : Pat<(v8f32 (ffloor VR256:$src)),
6250 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6251 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6252 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6253 def : Pat<(v8f32 (fceil VR256:$src)),
6254 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6255 def : Pat<(v8f32 (frint VR256:$src)),
6256 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6257 def : Pat<(v8f32 (ftrunc VR256:$src)),
6258 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6260 def : Pat<(v4f64 (ffloor VR256:$src)),
6261 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6262 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6263 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6264 def : Pat<(v4f64 (fceil VR256:$src)),
6265 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6266 def : Pat<(v4f64 (frint VR256:$src)),
6267 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6268 def : Pat<(v4f64 (ftrunc VR256:$src)),
6269 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6272 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6273 memopv4f32, memopv2f64,
6274 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6275 let Constraints = "$src1 = $dst" in
6276 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6277 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6279 let Predicates = [UseSSE41] in {
6280 def : Pat<(ffloor FR32:$src),
6281 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6282 def : Pat<(f64 (ffloor FR64:$src)),
6283 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6284 def : Pat<(f32 (fnearbyint FR32:$src)),
6285 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6286 def : Pat<(f64 (fnearbyint FR64:$src)),
6287 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6288 def : Pat<(f32 (fceil FR32:$src)),
6289 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6290 def : Pat<(f64 (fceil FR64:$src)),
6291 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6292 def : Pat<(f32 (frint FR32:$src)),
6293 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6294 def : Pat<(f64 (frint FR64:$src)),
6295 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6296 def : Pat<(f32 (ftrunc FR32:$src)),
6297 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6298 def : Pat<(f64 (ftrunc FR64:$src)),
6299 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6301 def : Pat<(v4f32 (ffloor VR128:$src)),
6302 (ROUNDPSr VR128:$src, (i32 0x1))>;
6303 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6304 (ROUNDPSr VR128:$src, (i32 0xC))>;
6305 def : Pat<(v4f32 (fceil VR128:$src)),
6306 (ROUNDPSr VR128:$src, (i32 0x2))>;
6307 def : Pat<(v4f32 (frint VR128:$src)),
6308 (ROUNDPSr VR128:$src, (i32 0x4))>;
6309 def : Pat<(v4f32 (ftrunc VR128:$src)),
6310 (ROUNDPSr VR128:$src, (i32 0x3))>;
6312 def : Pat<(v2f64 (ffloor VR128:$src)),
6313 (ROUNDPDr VR128:$src, (i32 0x1))>;
6314 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6315 (ROUNDPDr VR128:$src, (i32 0xC))>;
6316 def : Pat<(v2f64 (fceil VR128:$src)),
6317 (ROUNDPDr VR128:$src, (i32 0x2))>;
6318 def : Pat<(v2f64 (frint VR128:$src)),
6319 (ROUNDPDr VR128:$src, (i32 0x4))>;
6320 def : Pat<(v2f64 (ftrunc VR128:$src)),
6321 (ROUNDPDr VR128:$src, (i32 0x3))>;
6324 //===----------------------------------------------------------------------===//
6325 // SSE4.1 - Packed Bit Test
6326 //===----------------------------------------------------------------------===//
6328 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6329 // the intel intrinsic that corresponds to this.
6330 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6331 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6332 "vptest\t{$src2, $src1|$src1, $src2}",
6333 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6335 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6336 "vptest\t{$src2, $src1|$src1, $src2}",
6337 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6340 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6341 "vptest\t{$src2, $src1|$src1, $src2}",
6342 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6344 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6345 "vptest\t{$src2, $src1|$src1, $src2}",
6346 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6350 let Defs = [EFLAGS] in {
6351 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6352 "ptest\t{$src2, $src1|$src1, $src2}",
6353 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6355 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6356 "ptest\t{$src2, $src1|$src1, $src2}",
6357 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6361 // The bit test instructions below are AVX only
6362 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6363 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6364 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6365 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6366 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6367 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6368 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6369 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6373 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6374 let ExeDomain = SSEPackedSingle in {
6375 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6376 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6379 let ExeDomain = SSEPackedDouble in {
6380 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6381 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6386 //===----------------------------------------------------------------------===//
6387 // SSE4.1 - Misc Instructions
6388 //===----------------------------------------------------------------------===//
6390 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6391 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6392 "popcnt{w}\t{$src, $dst|$dst, $src}",
6393 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6395 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6396 "popcnt{w}\t{$src, $dst|$dst, $src}",
6397 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6398 (implicit EFLAGS)]>, OpSize, XS;
6400 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6401 "popcnt{l}\t{$src, $dst|$dst, $src}",
6402 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6404 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6405 "popcnt{l}\t{$src, $dst|$dst, $src}",
6406 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6407 (implicit EFLAGS)]>, XS;
6409 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6410 "popcnt{q}\t{$src, $dst|$dst, $src}",
6411 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6413 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6414 "popcnt{q}\t{$src, $dst|$dst, $src}",
6415 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6416 (implicit EFLAGS)]>, XS;
6421 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6422 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6423 Intrinsic IntId128> {
6424 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6426 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6427 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6428 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6430 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6433 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6436 let Predicates = [HasAVX] in
6437 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6438 int_x86_sse41_phminposuw>, VEX;
6439 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6440 int_x86_sse41_phminposuw>;
6442 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6443 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6444 Intrinsic IntId128, bit Is2Addr = 1> {
6445 let isCommutable = 1 in
6446 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6447 (ins VR128:$src1, VR128:$src2),
6449 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6450 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6451 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6452 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6453 (ins VR128:$src1, i128mem:$src2),
6455 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6456 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6458 (IntId128 VR128:$src1,
6459 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6462 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6463 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6464 Intrinsic IntId256> {
6465 let isCommutable = 1 in
6466 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6467 (ins VR256:$src1, VR256:$src2),
6468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6469 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6470 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6471 (ins VR256:$src1, i256mem:$src2),
6472 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6474 (IntId256 VR256:$src1,
6475 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6479 /// SS48I_binop_rm - Simple SSE41 binary operator.
6480 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6481 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6482 X86MemOperand x86memop, bit Is2Addr = 1> {
6483 let isCommutable = 1 in
6484 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6485 (ins RC:$src1, RC:$src2),
6487 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6488 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6489 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6490 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6491 (ins RC:$src1, x86memop:$src2),
6493 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6494 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6496 (OpVT (OpNode RC:$src1,
6497 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6500 let Predicates = [HasAVX] in {
6501 let isCommutable = 0 in
6502 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6504 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6505 memopv2i64, i128mem, 0>, VEX_4V;
6506 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6507 memopv2i64, i128mem, 0>, VEX_4V;
6508 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6509 memopv2i64, i128mem, 0>, VEX_4V;
6510 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6511 memopv2i64, i128mem, 0>, VEX_4V;
6512 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6513 memopv2i64, i128mem, 0>, VEX_4V;
6514 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6515 memopv2i64, i128mem, 0>, VEX_4V;
6516 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6517 memopv2i64, i128mem, 0>, VEX_4V;
6518 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6519 memopv2i64, i128mem, 0>, VEX_4V;
6520 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6524 let Predicates = [HasAVX2] in {
6525 let isCommutable = 0 in
6526 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6527 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6528 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6529 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6530 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6531 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6532 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6533 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6534 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6535 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6536 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6537 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6538 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6539 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6540 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6541 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6542 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6543 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6544 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6545 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6548 let Constraints = "$src1 = $dst" in {
6549 let isCommutable = 0 in
6550 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6551 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6552 memopv2i64, i128mem>;
6553 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6554 memopv2i64, i128mem>;
6555 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6556 memopv2i64, i128mem>;
6557 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6558 memopv2i64, i128mem>;
6559 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6560 memopv2i64, i128mem>;
6561 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6562 memopv2i64, i128mem>;
6563 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6564 memopv2i64, i128mem>;
6565 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6566 memopv2i64, i128mem>;
6567 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6570 let Predicates = [HasAVX] in {
6571 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6572 memopv2i64, i128mem, 0>, VEX_4V;
6573 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6574 memopv2i64, i128mem, 0>, VEX_4V;
6576 let Predicates = [HasAVX2] in {
6577 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6578 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6579 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6580 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6583 let Constraints = "$src1 = $dst" in {
6584 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6585 memopv2i64, i128mem>;
6586 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6587 memopv2i64, i128mem>;
6590 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6591 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6592 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6593 X86MemOperand x86memop, bit Is2Addr = 1> {
6594 let isCommutable = 1 in
6595 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6596 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6598 !strconcat(OpcodeStr,
6599 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6600 !strconcat(OpcodeStr,
6601 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6602 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6604 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6605 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6607 !strconcat(OpcodeStr,
6608 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6609 !strconcat(OpcodeStr,
6610 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6613 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6617 let Predicates = [HasAVX] in {
6618 let isCommutable = 0 in {
6619 let ExeDomain = SSEPackedSingle in {
6620 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6621 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6622 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6623 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6624 f256mem, 0>, VEX_4V, VEX_L;
6626 let ExeDomain = SSEPackedDouble in {
6627 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6628 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6629 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6630 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6631 f256mem, 0>, VEX_4V, VEX_L;
6633 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6634 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6635 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6636 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6638 let ExeDomain = SSEPackedSingle in
6639 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6640 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6641 let ExeDomain = SSEPackedDouble in
6642 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6643 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6644 let ExeDomain = SSEPackedSingle in
6645 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6646 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6649 let Predicates = [HasAVX2] in {
6650 let isCommutable = 0 in {
6651 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6652 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6653 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6654 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6658 let Constraints = "$src1 = $dst" in {
6659 let isCommutable = 0 in {
6660 let ExeDomain = SSEPackedSingle in
6661 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6662 VR128, memopv4f32, f128mem>;
6663 let ExeDomain = SSEPackedDouble in
6664 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6665 VR128, memopv2f64, f128mem>;
6666 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6667 VR128, memopv2i64, i128mem>;
6668 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6669 VR128, memopv2i64, i128mem>;
6671 let ExeDomain = SSEPackedSingle in
6672 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6673 VR128, memopv4f32, f128mem>;
6674 let ExeDomain = SSEPackedDouble in
6675 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6676 VR128, memopv2f64, f128mem>;
6679 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6680 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6681 RegisterClass RC, X86MemOperand x86memop,
6682 PatFrag mem_frag, Intrinsic IntId> {
6683 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6684 (ins RC:$src1, RC:$src2, RC:$src3),
6685 !strconcat(OpcodeStr,
6686 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6687 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6688 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6690 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6691 (ins RC:$src1, x86memop:$src2, RC:$src3),
6692 !strconcat(OpcodeStr,
6693 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6695 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6697 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6700 let Predicates = [HasAVX] in {
6701 let ExeDomain = SSEPackedDouble in {
6702 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6703 memopv2f64, int_x86_sse41_blendvpd>;
6704 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6705 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6706 } // ExeDomain = SSEPackedDouble
6707 let ExeDomain = SSEPackedSingle in {
6708 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6709 memopv4f32, int_x86_sse41_blendvps>;
6710 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6711 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6712 } // ExeDomain = SSEPackedSingle
6713 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6714 memopv2i64, int_x86_sse41_pblendvb>;
6717 let Predicates = [HasAVX2] in {
6718 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6719 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6722 let Predicates = [HasAVX] in {
6723 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6724 (v16i8 VR128:$src2))),
6725 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6726 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6727 (v4i32 VR128:$src2))),
6728 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6729 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6730 (v4f32 VR128:$src2))),
6731 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6732 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6733 (v2i64 VR128:$src2))),
6734 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6735 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6736 (v2f64 VR128:$src2))),
6737 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6738 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6739 (v8i32 VR256:$src2))),
6740 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6741 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6742 (v8f32 VR256:$src2))),
6743 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6744 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6745 (v4i64 VR256:$src2))),
6746 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6747 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6748 (v4f64 VR256:$src2))),
6749 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6751 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6753 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6754 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6756 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6758 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6760 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6761 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6763 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6764 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6766 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6769 let Predicates = [HasAVX2] in {
6770 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6771 (v32i8 VR256:$src2))),
6772 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6773 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6775 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6778 /// SS41I_ternary_int - SSE 4.1 ternary operator
6779 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6780 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6781 X86MemOperand x86memop, Intrinsic IntId> {
6782 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6783 (ins VR128:$src1, VR128:$src2),
6784 !strconcat(OpcodeStr,
6785 "\t{$src2, $dst|$dst, $src2}"),
6786 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6789 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6790 (ins VR128:$src1, x86memop:$src2),
6791 !strconcat(OpcodeStr,
6792 "\t{$src2, $dst|$dst, $src2}"),
6795 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6799 let ExeDomain = SSEPackedDouble in
6800 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6801 int_x86_sse41_blendvpd>;
6802 let ExeDomain = SSEPackedSingle in
6803 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6804 int_x86_sse41_blendvps>;
6805 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6806 int_x86_sse41_pblendvb>;
6808 // Aliases with the implicit xmm0 argument
6809 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6810 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6811 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6812 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6813 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6814 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6815 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6816 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6817 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6818 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6819 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6820 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6822 let Predicates = [UseSSE41] in {
6823 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6824 (v16i8 VR128:$src2))),
6825 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6826 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6827 (v4i32 VR128:$src2))),
6828 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6829 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6830 (v4f32 VR128:$src2))),
6831 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6832 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6833 (v2i64 VR128:$src2))),
6834 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6835 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6836 (v2f64 VR128:$src2))),
6837 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6839 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6841 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6842 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6844 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6845 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6847 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6851 let Predicates = [HasAVX] in
6852 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6853 "vmovntdqa\t{$src, $dst|$dst, $src}",
6854 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6856 let Predicates = [HasAVX2] in
6857 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6858 "vmovntdqa\t{$src, $dst|$dst, $src}",
6859 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6861 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6862 "movntdqa\t{$src, $dst|$dst, $src}",
6863 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6866 //===----------------------------------------------------------------------===//
6867 // SSE4.2 - Compare Instructions
6868 //===----------------------------------------------------------------------===//
6870 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6871 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6872 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6873 X86MemOperand x86memop, bit Is2Addr = 1> {
6874 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6875 (ins RC:$src1, RC:$src2),
6877 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6878 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6879 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6881 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6882 (ins RC:$src1, x86memop:$src2),
6884 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6885 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6887 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6890 let Predicates = [HasAVX] in
6891 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6892 memopv2i64, i128mem, 0>, VEX_4V;
6894 let Predicates = [HasAVX2] in
6895 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6896 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6898 let Constraints = "$src1 = $dst" in
6899 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6900 memopv2i64, i128mem>;
6902 //===----------------------------------------------------------------------===//
6903 // SSE4.2 - String/text Processing Instructions
6904 //===----------------------------------------------------------------------===//
6906 // Packed Compare Implicit Length Strings, Return Mask
6907 multiclass pseudo_pcmpistrm<string asm> {
6908 def REG : PseudoI<(outs VR128:$dst),
6909 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6910 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6912 def MEM : PseudoI<(outs VR128:$dst),
6913 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6914 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
6915 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6918 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6919 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6920 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
6923 multiclass pcmpistrm_SS42AI<string asm> {
6924 def rr : SS42AI<0x62, MRMSrcReg, (outs),
6925 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6926 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6929 def rm :SS42AI<0x62, MRMSrcMem, (outs),
6930 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6931 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6935 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6936 let Predicates = [HasAVX] in
6937 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
6938 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
6941 // Packed Compare Explicit Length Strings, Return Mask
6942 multiclass pseudo_pcmpestrm<string asm> {
6943 def REG : PseudoI<(outs VR128:$dst),
6944 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6945 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6946 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6947 def MEM : PseudoI<(outs VR128:$dst),
6948 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6949 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
6950 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
6953 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6954 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6955 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
6958 multiclass SS42AI_pcmpestrm<string asm> {
6959 def rr : SS42AI<0x60, MRMSrcReg, (outs),
6960 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6961 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6964 def rm : SS42AI<0x60, MRMSrcMem, (outs),
6965 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6966 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6970 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6971 let Predicates = [HasAVX] in
6972 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
6973 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
6976 // Packed Compare Implicit Length Strings, Return Index
6977 multiclass pseudo_pcmpistri<string asm> {
6978 def REG : PseudoI<(outs GR32:$dst),
6979 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6980 [(set GR32:$dst, EFLAGS,
6981 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
6982 def MEM : PseudoI<(outs GR32:$dst),
6983 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6984 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
6985 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6988 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6989 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
6990 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
6993 multiclass SS42AI_pcmpistri<string asm> {
6994 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6995 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6996 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6999 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7000 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7001 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7005 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7006 let Predicates = [HasAVX] in
7007 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7008 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7011 // Packed Compare Explicit Length Strings, Return Index
7012 multiclass pseudo_pcmpestri<string asm> {
7013 def REG : PseudoI<(outs GR32:$dst),
7014 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7015 [(set GR32:$dst, EFLAGS,
7016 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7017 def MEM : PseudoI<(outs GR32:$dst),
7018 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7019 [(set GR32:$dst, EFLAGS,
7020 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7024 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7025 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7026 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7029 multiclass SS42AI_pcmpestri<string asm> {
7030 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7031 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7032 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7035 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7036 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7037 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7041 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7042 let Predicates = [HasAVX] in
7043 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7044 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7047 //===----------------------------------------------------------------------===//
7048 // SSE4.2 - CRC Instructions
7049 //===----------------------------------------------------------------------===//
7051 // No CRC instructions have AVX equivalents
7053 // crc intrinsic instruction
7054 // This set of instructions are only rm, the only difference is the size
7056 let Constraints = "$src1 = $dst" in {
7057 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7058 (ins GR32:$src1, i8mem:$src2),
7059 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7061 (int_x86_sse42_crc32_32_8 GR32:$src1,
7062 (load addr:$src2)))]>;
7063 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7064 (ins GR32:$src1, GR8:$src2),
7065 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7067 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7068 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7069 (ins GR32:$src1, i16mem:$src2),
7070 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7072 (int_x86_sse42_crc32_32_16 GR32:$src1,
7073 (load addr:$src2)))]>,
7075 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7076 (ins GR32:$src1, GR16:$src2),
7077 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7079 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7081 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7082 (ins GR32:$src1, i32mem:$src2),
7083 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7085 (int_x86_sse42_crc32_32_32 GR32:$src1,
7086 (load addr:$src2)))]>;
7087 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7088 (ins GR32:$src1, GR32:$src2),
7089 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7091 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7092 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7093 (ins GR64:$src1, i8mem:$src2),
7094 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7096 (int_x86_sse42_crc32_64_8 GR64:$src1,
7097 (load addr:$src2)))]>,
7099 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7100 (ins GR64:$src1, GR8:$src2),
7101 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7103 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7105 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7106 (ins GR64:$src1, i64mem:$src2),
7107 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7109 (int_x86_sse42_crc32_64_64 GR64:$src1,
7110 (load addr:$src2)))]>,
7112 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7113 (ins GR64:$src1, GR64:$src2),
7114 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7116 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7120 //===----------------------------------------------------------------------===//
7121 // AES-NI Instructions
7122 //===----------------------------------------------------------------------===//
7124 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7125 Intrinsic IntId128, bit Is2Addr = 1> {
7126 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7127 (ins VR128:$src1, VR128:$src2),
7129 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7130 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7131 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7133 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7134 (ins VR128:$src1, i128mem:$src2),
7136 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7137 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7139 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7142 // Perform One Round of an AES Encryption/Decryption Flow
7143 let Predicates = [HasAVX, HasAES] in {
7144 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7145 int_x86_aesni_aesenc, 0>, VEX_4V;
7146 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7147 int_x86_aesni_aesenclast, 0>, VEX_4V;
7148 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7149 int_x86_aesni_aesdec, 0>, VEX_4V;
7150 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7151 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7154 let Constraints = "$src1 = $dst" in {
7155 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7156 int_x86_aesni_aesenc>;
7157 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7158 int_x86_aesni_aesenclast>;
7159 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7160 int_x86_aesni_aesdec>;
7161 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7162 int_x86_aesni_aesdeclast>;
7165 // Perform the AES InvMixColumn Transformation
7166 let Predicates = [HasAVX, HasAES] in {
7167 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7169 "vaesimc\t{$src1, $dst|$dst, $src1}",
7171 (int_x86_aesni_aesimc VR128:$src1))]>,
7173 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7174 (ins i128mem:$src1),
7175 "vaesimc\t{$src1, $dst|$dst, $src1}",
7176 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7179 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7181 "aesimc\t{$src1, $dst|$dst, $src1}",
7183 (int_x86_aesni_aesimc VR128:$src1))]>,
7185 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7186 (ins i128mem:$src1),
7187 "aesimc\t{$src1, $dst|$dst, $src1}",
7188 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7191 // AES Round Key Generation Assist
7192 let Predicates = [HasAVX, HasAES] in {
7193 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7194 (ins VR128:$src1, i8imm:$src2),
7195 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7197 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7199 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7200 (ins i128mem:$src1, i8imm:$src2),
7201 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7203 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7206 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7207 (ins VR128:$src1, i8imm:$src2),
7208 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7210 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7212 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7213 (ins i128mem:$src1, i8imm:$src2),
7214 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7216 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7219 //===----------------------------------------------------------------------===//
7220 // PCLMUL Instructions
7221 //===----------------------------------------------------------------------===//
7223 // AVX carry-less Multiplication instructions
7224 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7225 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7226 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7228 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7230 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7231 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7232 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7233 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7234 (memopv2i64 addr:$src2), imm:$src3))]>;
7236 // Carry-less Multiplication instructions
7237 let Constraints = "$src1 = $dst" in {
7238 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7239 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7240 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7242 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7244 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7245 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7246 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7247 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7248 (memopv2i64 addr:$src2), imm:$src3))]>;
7249 } // Constraints = "$src1 = $dst"
7252 multiclass pclmul_alias<string asm, int immop> {
7253 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7254 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7256 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7257 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7259 def : InstAlias<!strconcat("vpclmul", asm,
7260 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7261 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7263 def : InstAlias<!strconcat("vpclmul", asm,
7264 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7265 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7267 defm : pclmul_alias<"hqhq", 0x11>;
7268 defm : pclmul_alias<"hqlq", 0x01>;
7269 defm : pclmul_alias<"lqhq", 0x10>;
7270 defm : pclmul_alias<"lqlq", 0x00>;
7272 //===----------------------------------------------------------------------===//
7273 // SSE4A Instructions
7274 //===----------------------------------------------------------------------===//
7276 let Predicates = [HasSSE4A] in {
7278 let Constraints = "$src = $dst" in {
7279 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7280 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7281 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7282 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7283 imm:$idx))]>, TB, OpSize;
7284 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7285 (ins VR128:$src, VR128:$mask),
7286 "extrq\t{$mask, $src|$src, $mask}",
7287 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7288 VR128:$mask))]>, TB, OpSize;
7290 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7291 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7292 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7293 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7294 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7295 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7296 (ins VR128:$src, VR128:$mask),
7297 "insertq\t{$mask, $src|$src, $mask}",
7298 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7299 VR128:$mask))]>, XD;
7302 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7303 "movntss\t{$src, $dst|$dst, $src}",
7304 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7306 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7307 "movntsd\t{$src, $dst|$dst, $src}",
7308 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7311 //===----------------------------------------------------------------------===//
7313 //===----------------------------------------------------------------------===//
7315 //===----------------------------------------------------------------------===//
7316 // VBROADCAST - Load from memory and broadcast to all elements of the
7317 // destination operand
7319 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7320 X86MemOperand x86memop, Intrinsic Int> :
7321 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7322 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7323 [(set RC:$dst, (Int addr:$src))]>, VEX;
7325 // AVX2 adds register forms
7326 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7328 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7329 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7330 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7332 let ExeDomain = SSEPackedSingle in {
7333 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7334 int_x86_avx_vbroadcast_ss>;
7335 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7336 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7338 let ExeDomain = SSEPackedDouble in
7339 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7340 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7341 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7342 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7344 let ExeDomain = SSEPackedSingle in {
7345 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7346 int_x86_avx2_vbroadcast_ss_ps>;
7347 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7348 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7350 let ExeDomain = SSEPackedDouble in
7351 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7352 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7354 let Predicates = [HasAVX2] in
7355 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7356 int_x86_avx2_vbroadcasti128>, VEX_L;
7358 let Predicates = [HasAVX] in
7359 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7360 (VBROADCASTF128 addr:$src)>;
7363 //===----------------------------------------------------------------------===//
7364 // VINSERTF128 - Insert packed floating-point values
7366 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7367 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7368 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7369 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7372 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7373 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7374 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7378 let Predicates = [HasAVX] in {
7379 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7381 (VINSERTF128rr VR256:$src1, VR128:$src2,
7382 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7383 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7385 (VINSERTF128rr VR256:$src1, VR128:$src2,
7386 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7388 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7390 (VINSERTF128rm VR256:$src1, addr:$src2,
7391 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7392 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7394 (VINSERTF128rm VR256:$src1, addr:$src2,
7395 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7398 let Predicates = [HasAVX1Only] in {
7399 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7401 (VINSERTF128rr VR256:$src1, VR128:$src2,
7402 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7403 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7405 (VINSERTF128rr VR256:$src1, VR128:$src2,
7406 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7407 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7409 (VINSERTF128rr VR256:$src1, VR128:$src2,
7410 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7411 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7413 (VINSERTF128rr VR256:$src1, VR128:$src2,
7414 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7416 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7418 (VINSERTF128rm VR256:$src1, addr:$src2,
7419 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7420 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7421 (bc_v4i32 (memopv2i64 addr:$src2)),
7423 (VINSERTF128rm VR256:$src1, addr:$src2,
7424 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7425 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7426 (bc_v16i8 (memopv2i64 addr:$src2)),
7428 (VINSERTF128rm VR256:$src1, addr:$src2,
7429 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7430 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7431 (bc_v8i16 (memopv2i64 addr:$src2)),
7433 (VINSERTF128rm VR256:$src1, addr:$src2,
7434 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7437 //===----------------------------------------------------------------------===//
7438 // VEXTRACTF128 - Extract packed floating-point values
7440 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7441 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7442 (ins VR256:$src1, i8imm:$src2),
7443 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7446 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7447 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7448 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7453 let Predicates = [HasAVX] in {
7454 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7455 (v4f32 (VEXTRACTF128rr
7456 (v8f32 VR256:$src1),
7457 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7458 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7459 (v2f64 (VEXTRACTF128rr
7460 (v4f64 VR256:$src1),
7461 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7463 def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
7464 (iPTR imm))), addr:$dst),
7465 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7466 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7467 def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
7468 (iPTR imm))), addr:$dst),
7469 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7470 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7473 let Predicates = [HasAVX1Only] in {
7474 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7475 (v2i64 (VEXTRACTF128rr
7476 (v4i64 VR256:$src1),
7477 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7478 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7479 (v4i32 (VEXTRACTF128rr
7480 (v8i32 VR256:$src1),
7481 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7482 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7483 (v8i16 (VEXTRACTF128rr
7484 (v16i16 VR256:$src1),
7485 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7486 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7487 (v16i8 (VEXTRACTF128rr
7488 (v32i8 VR256:$src1),
7489 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7491 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7492 (iPTR imm))), addr:$dst),
7493 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7494 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7495 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7496 (iPTR imm))), addr:$dst),
7497 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7498 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7499 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7500 (iPTR imm))), addr:$dst),
7501 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7502 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7503 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7504 (iPTR imm))), addr:$dst),
7505 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7506 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7509 //===----------------------------------------------------------------------===//
7510 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7512 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7513 Intrinsic IntLd, Intrinsic IntLd256,
7514 Intrinsic IntSt, Intrinsic IntSt256> {
7515 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7516 (ins VR128:$src1, f128mem:$src2),
7517 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7518 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7520 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7521 (ins VR256:$src1, f256mem:$src2),
7522 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7523 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7525 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7526 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7527 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7528 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7529 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7530 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7531 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7532 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7535 let ExeDomain = SSEPackedSingle in
7536 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7537 int_x86_avx_maskload_ps,
7538 int_x86_avx_maskload_ps_256,
7539 int_x86_avx_maskstore_ps,
7540 int_x86_avx_maskstore_ps_256>;
7541 let ExeDomain = SSEPackedDouble in
7542 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7543 int_x86_avx_maskload_pd,
7544 int_x86_avx_maskload_pd_256,
7545 int_x86_avx_maskstore_pd,
7546 int_x86_avx_maskstore_pd_256>;
7548 //===----------------------------------------------------------------------===//
7549 // VPERMIL - Permute Single and Double Floating-Point Values
7551 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7552 RegisterClass RC, X86MemOperand x86memop_f,
7553 X86MemOperand x86memop_i, PatFrag i_frag,
7554 Intrinsic IntVar, ValueType vt> {
7555 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7556 (ins RC:$src1, RC:$src2),
7557 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7558 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7559 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7560 (ins RC:$src1, x86memop_i:$src2),
7561 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7562 [(set RC:$dst, (IntVar RC:$src1,
7563 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7565 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7566 (ins RC:$src1, i8imm:$src2),
7567 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7568 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7569 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7570 (ins x86memop_f:$src1, i8imm:$src2),
7571 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7573 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7576 let ExeDomain = SSEPackedSingle in {
7577 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7578 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7579 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7580 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7582 let ExeDomain = SSEPackedDouble in {
7583 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7584 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7585 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7586 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7589 let Predicates = [HasAVX] in {
7590 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7591 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7592 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7593 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7594 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7596 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7597 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7598 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7600 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7601 (VPERMILPDri VR128:$src1, imm:$imm)>;
7602 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7603 (VPERMILPDmi addr:$src1, imm:$imm)>;
7606 //===----------------------------------------------------------------------===//
7607 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7609 let ExeDomain = SSEPackedSingle in {
7610 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7611 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7612 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7613 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7614 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7615 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7616 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7617 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7618 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7619 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7622 let Predicates = [HasAVX] in {
7623 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7624 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7625 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7626 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7627 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7630 let Predicates = [HasAVX1Only] in {
7631 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7632 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7633 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7634 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7635 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7636 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7637 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7638 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7640 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7641 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7642 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7643 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7644 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7645 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7646 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7647 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7648 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7649 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7650 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7651 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7654 //===----------------------------------------------------------------------===//
7655 // VZERO - Zero YMM registers
7657 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7658 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7659 // Zero All YMM registers
7660 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7661 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7663 // Zero Upper bits of YMM registers
7664 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7665 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7668 //===----------------------------------------------------------------------===//
7669 // Half precision conversion instructions
7670 //===----------------------------------------------------------------------===//
7671 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7672 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7673 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7674 [(set RC:$dst, (Int VR128:$src))]>,
7676 let neverHasSideEffects = 1, mayLoad = 1 in
7677 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7678 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7681 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7682 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7683 (ins RC:$src1, i32i8imm:$src2),
7684 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7685 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7687 let neverHasSideEffects = 1, mayStore = 1 in
7688 def mr : Ii8<0x1D, MRMDestMem, (outs),
7689 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7690 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7694 let Predicates = [HasAVX, HasF16C] in {
7695 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7696 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7697 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7698 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7701 //===----------------------------------------------------------------------===//
7702 // AVX2 Instructions
7703 //===----------------------------------------------------------------------===//
7705 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7706 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7707 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7708 X86MemOperand x86memop> {
7709 let isCommutable = 1 in
7710 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7711 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7712 !strconcat(OpcodeStr,
7713 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7714 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7716 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7717 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7718 !strconcat(OpcodeStr,
7719 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7722 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7726 let isCommutable = 0 in {
7727 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7728 VR128, memopv2i64, i128mem>;
7729 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7730 VR256, memopv4i64, i256mem>, VEX_L;
7733 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7735 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7736 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7738 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7740 //===----------------------------------------------------------------------===//
7741 // VPBROADCAST - Load from memory and broadcast to all elements of the
7742 // destination operand
7744 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7745 X86MemOperand x86memop, PatFrag ld_frag,
7746 Intrinsic Int128, Intrinsic Int256> {
7747 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7748 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7749 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7750 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7751 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7753 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7754 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7755 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7756 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7757 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7758 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7760 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7764 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7765 int_x86_avx2_pbroadcastb_128,
7766 int_x86_avx2_pbroadcastb_256>;
7767 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7768 int_x86_avx2_pbroadcastw_128,
7769 int_x86_avx2_pbroadcastw_256>;
7770 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7771 int_x86_avx2_pbroadcastd_128,
7772 int_x86_avx2_pbroadcastd_256>;
7773 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7774 int_x86_avx2_pbroadcastq_128,
7775 int_x86_avx2_pbroadcastq_256>;
7777 let Predicates = [HasAVX2] in {
7778 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7779 (VPBROADCASTBrm addr:$src)>;
7780 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7781 (VPBROADCASTBYrm addr:$src)>;
7782 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7783 (VPBROADCASTWrm addr:$src)>;
7784 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7785 (VPBROADCASTWYrm addr:$src)>;
7786 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7787 (VPBROADCASTDrm addr:$src)>;
7788 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7789 (VPBROADCASTDYrm addr:$src)>;
7790 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7791 (VPBROADCASTQrm addr:$src)>;
7792 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7793 (VPBROADCASTQYrm addr:$src)>;
7795 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7796 (VPBROADCASTBrr VR128:$src)>;
7797 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7798 (VPBROADCASTBYrr VR128:$src)>;
7799 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7800 (VPBROADCASTWrr VR128:$src)>;
7801 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7802 (VPBROADCASTWYrr VR128:$src)>;
7803 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7804 (VPBROADCASTDrr VR128:$src)>;
7805 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7806 (VPBROADCASTDYrr VR128:$src)>;
7807 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7808 (VPBROADCASTQrr VR128:$src)>;
7809 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7810 (VPBROADCASTQYrr VR128:$src)>;
7811 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7812 (VBROADCASTSSrr VR128:$src)>;
7813 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7814 (VBROADCASTSSYrr VR128:$src)>;
7815 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7816 (VPBROADCASTQrr VR128:$src)>;
7817 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7818 (VBROADCASTSDYrr VR128:$src)>;
7820 // Provide fallback in case the load node that is used in the patterns above
7821 // is used by additional users, which prevents the pattern selection.
7822 let AddedComplexity = 20 in {
7823 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7824 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7825 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7826 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7827 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7828 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7830 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7831 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7832 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7833 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7834 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7835 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7839 // AVX1 broadcast patterns
7840 let Predicates = [HasAVX1Only] in {
7841 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7842 (VBROADCASTSSYrm addr:$src)>;
7843 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7844 (VBROADCASTSDYrm addr:$src)>;
7845 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7846 (VBROADCASTSSrm addr:$src)>;
7849 let Predicates = [HasAVX] in {
7850 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7851 (VBROADCASTSSYrm addr:$src)>;
7852 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7853 (VBROADCASTSDYrm addr:$src)>;
7854 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7855 (VBROADCASTSSrm addr:$src)>;
7857 // Provide fallback in case the load node that is used in the patterns above
7858 // is used by additional users, which prevents the pattern selection.
7859 let AddedComplexity = 20 in {
7860 // 128bit broadcasts:
7861 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7862 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7863 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7864 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7865 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7866 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7867 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7868 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7869 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7870 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7872 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7873 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7874 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7875 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7876 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7877 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7878 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7879 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7880 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7881 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7885 //===----------------------------------------------------------------------===//
7886 // VPERM - Permute instructions
7889 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7891 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7892 (ins VR256:$src1, VR256:$src2),
7893 !strconcat(OpcodeStr,
7894 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7896 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
7898 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7899 (ins VR256:$src1, i256mem:$src2),
7900 !strconcat(OpcodeStr,
7901 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7903 (OpVT (X86VPermv VR256:$src1,
7904 (bitconvert (mem_frag addr:$src2)))))]>,
7908 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7909 let ExeDomain = SSEPackedSingle in
7910 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7912 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7914 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7915 (ins VR256:$src1, i8imm:$src2),
7916 !strconcat(OpcodeStr,
7917 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7919 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
7921 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7922 (ins i256mem:$src1, i8imm:$src2),
7923 !strconcat(OpcodeStr,
7924 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7926 (OpVT (X86VPermi (mem_frag addr:$src1),
7927 (i8 imm:$src2))))]>, VEX, VEX_L;
7930 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7931 let ExeDomain = SSEPackedDouble in
7932 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7934 //===----------------------------------------------------------------------===//
7935 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7937 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7938 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7939 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7940 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7941 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7942 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7943 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7944 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7945 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7946 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7948 let Predicates = [HasAVX2] in {
7949 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7950 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7951 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7952 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7953 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7954 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7956 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7958 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7959 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7960 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7961 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7962 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7964 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7968 //===----------------------------------------------------------------------===//
7969 // VINSERTI128 - Insert packed integer values
7971 let neverHasSideEffects = 1 in {
7972 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7973 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7974 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7977 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7978 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7979 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7983 let Predicates = [HasAVX2] in {
7984 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7986 (VINSERTI128rr VR256:$src1, VR128:$src2,
7987 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7988 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7990 (VINSERTI128rr VR256:$src1, VR128:$src2,
7991 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7992 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7994 (VINSERTI128rr VR256:$src1, VR128:$src2,
7995 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7996 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7998 (VINSERTI128rr VR256:$src1, VR128:$src2,
7999 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8001 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
8003 (VINSERTI128rm VR256:$src1, addr:$src2,
8004 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8005 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
8006 (bc_v4i32 (memopv2i64 addr:$src2)),
8008 (VINSERTI128rm VR256:$src1, addr:$src2,
8009 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8010 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
8011 (bc_v16i8 (memopv2i64 addr:$src2)),
8013 (VINSERTI128rm VR256:$src1, addr:$src2,
8014 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8015 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
8016 (bc_v8i16 (memopv2i64 addr:$src2)),
8018 (VINSERTI128rm VR256:$src1, addr:$src2,
8019 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8022 //===----------------------------------------------------------------------===//
8023 // VEXTRACTI128 - Extract packed integer values
8025 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8026 (ins VR256:$src1, i8imm:$src2),
8027 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8029 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8031 let neverHasSideEffects = 1, mayStore = 1 in
8032 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8033 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8034 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8037 let Predicates = [HasAVX2] in {
8038 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8039 (v2i64 (VEXTRACTI128rr
8040 (v4i64 VR256:$src1),
8041 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8042 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8043 (v4i32 (VEXTRACTI128rr
8044 (v8i32 VR256:$src1),
8045 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8046 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8047 (v8i16 (VEXTRACTI128rr
8048 (v16i16 VR256:$src1),
8049 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8050 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8051 (v16i8 (VEXTRACTI128rr
8052 (v32i8 VR256:$src1),
8053 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8055 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
8056 (iPTR imm))), addr:$dst),
8057 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8058 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8059 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
8060 (iPTR imm))), addr:$dst),
8061 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8062 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8063 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
8064 (iPTR imm))), addr:$dst),
8065 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8066 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8067 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
8068 (iPTR imm))), addr:$dst),
8069 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8070 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8073 //===----------------------------------------------------------------------===//
8074 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8076 multiclass avx2_pmovmask<string OpcodeStr,
8077 Intrinsic IntLd128, Intrinsic IntLd256,
8078 Intrinsic IntSt128, Intrinsic IntSt256> {
8079 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8080 (ins VR128:$src1, i128mem:$src2),
8081 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8082 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8083 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8084 (ins VR256:$src1, i256mem:$src2),
8085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8086 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8088 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8089 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8091 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8092 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8093 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8094 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8095 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8098 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8099 int_x86_avx2_maskload_d,
8100 int_x86_avx2_maskload_d_256,
8101 int_x86_avx2_maskstore_d,
8102 int_x86_avx2_maskstore_d_256>;
8103 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8104 int_x86_avx2_maskload_q,
8105 int_x86_avx2_maskload_q_256,
8106 int_x86_avx2_maskstore_q,
8107 int_x86_avx2_maskstore_q_256>, VEX_W;
8110 //===----------------------------------------------------------------------===//
8111 // Variable Bit Shifts
8113 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8114 ValueType vt128, ValueType vt256> {
8115 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8116 (ins VR128:$src1, VR128:$src2),
8117 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8119 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8121 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8122 (ins VR128:$src1, i128mem:$src2),
8123 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8125 (vt128 (OpNode VR128:$src1,
8126 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8128 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8129 (ins VR256:$src1, VR256:$src2),
8130 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8132 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8134 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8135 (ins VR256:$src1, i256mem:$src2),
8136 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8138 (vt256 (OpNode VR256:$src1,
8139 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8143 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8144 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8145 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8146 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8147 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8149 //===----------------------------------------------------------------------===//
8150 // VGATHER - GATHER Operations
8151 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8152 X86MemOperand memop128, X86MemOperand memop256> {
8153 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8154 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8155 !strconcat(OpcodeStr,
8156 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8158 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8159 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8160 !strconcat(OpcodeStr,
8161 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8162 []>, VEX_4VOp3, VEX_L;
8165 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8166 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8167 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8168 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8169 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8170 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8171 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8172 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8173 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;