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 let isCommutable = 1, hasSideEffects = 0 in
208 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
210 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
211 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
212 pat_rr, IIC_DEFAULT, d>;
213 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
215 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
216 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
217 pat_rm, IIC_DEFAULT, d>;
220 //===----------------------------------------------------------------------===//
221 // Non-instruction patterns
222 //===----------------------------------------------------------------------===//
224 // A vector extract of the first f32/f64 position is a subregister copy
225 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
226 (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32)>;
227 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
228 (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64)>;
230 // A 128-bit subvector extract from the first 256-bit vector position
231 // is a subregister copy that needs no instruction.
232 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (iPTR 0))),
233 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
234 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (iPTR 0))),
235 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
237 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (iPTR 0))),
238 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
239 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (iPTR 0))),
240 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
242 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (iPTR 0))),
243 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
244 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (iPTR 0))),
245 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
247 // A 128-bit subvector insert to the first 256-bit vector position
248 // is a subregister copy that needs no instruction.
249 let AddedComplexity = 25 in { // to give priority over vinsertf128rm
250 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)),
251 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
252 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)),
253 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
254 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)),
255 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
256 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)),
257 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
258 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (iPTR 0)),
259 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
260 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (iPTR 0)),
261 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
264 // Implicitly promote a 32-bit scalar to a vector.
265 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
266 (COPY_TO_REGCLASS FR32:$src, VR128)>;
267 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
268 (COPY_TO_REGCLASS FR32:$src, VR128)>;
269 // Implicitly promote a 64-bit scalar to a vector.
270 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
271 (COPY_TO_REGCLASS FR64:$src, VR128)>;
272 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
273 (COPY_TO_REGCLASS FR64:$src, VR128)>;
275 // Bitcasts between 128-bit vector types. Return the original type since
276 // no instruction is needed for the conversion
277 let Predicates = [HasSSE2] in {
278 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
279 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
280 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
281 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
282 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
283 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
284 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
285 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
286 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
287 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
288 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
289 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
290 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
291 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
292 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
293 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
294 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
295 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
296 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
297 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
298 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
299 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
300 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
301 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
302 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
303 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
304 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
305 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
306 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
307 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
310 // Bitcasts between 256-bit vector types. Return the original type since
311 // no instruction is needed for the conversion
312 let Predicates = [HasAVX] in {
313 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
314 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
315 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
316 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
317 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
318 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
319 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
320 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
321 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
322 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
323 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
324 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
325 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
326 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
327 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
328 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
329 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
330 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
331 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
332 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
333 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
334 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
335 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
336 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
337 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
338 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
339 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
340 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
341 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
342 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
345 // Alias instructions that map fld0 to xorps for sse or vxorps for avx.
346 // This is expanded by ExpandPostRAPseudos.
347 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
349 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
350 [(set FR32:$dst, fp32imm0)]>, Requires<[HasSSE1]>;
351 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
352 [(set FR64:$dst, fpimm0)]>, Requires<[HasSSE2]>;
355 //===----------------------------------------------------------------------===//
356 // AVX & SSE - Zero/One Vectors
357 //===----------------------------------------------------------------------===//
359 // Alias instruction that maps zero vector to pxor / xorp* for sse.
360 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
361 // swizzled by ExecutionDepsFix to pxor.
362 // We set canFoldAsLoad because this can be converted to a constant-pool
363 // load of an all-zeros value if folding it would be beneficial.
364 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
366 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "",
367 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
370 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
371 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
372 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
373 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
374 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
377 // The same as done above but for AVX. The 256-bit AVX1 ISA doesn't support PI,
378 // and doesn't need it because on sandy bridge the register is set to zero
379 // at the rename stage without using any execution unit, so SET0PSY
380 // and SET0PDY can be used for vector int instructions without penalty
381 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
382 isPseudo = 1, Predicates = [HasAVX] in {
383 def AVX_SET0 : I<0, Pseudo, (outs VR256:$dst), (ins), "",
384 [(set VR256:$dst, (v8f32 immAllZerosV))]>;
387 let Predicates = [HasAVX] in
388 def : Pat<(v4f64 immAllZerosV), (AVX_SET0)>;
390 let Predicates = [HasAVX2] in {
391 def : Pat<(v4i64 immAllZerosV), (AVX_SET0)>;
392 def : Pat<(v8i32 immAllZerosV), (AVX_SET0)>;
393 def : Pat<(v16i16 immAllZerosV), (AVX_SET0)>;
394 def : Pat<(v32i8 immAllZerosV), (AVX_SET0)>;
397 // AVX1 has no support for 256-bit integer instructions, but since the 128-bit
398 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
399 let Predicates = [HasAVX1Only] in {
400 def : Pat<(v32i8 immAllZerosV), (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
401 def : Pat<(bc_v32i8 (v8f32 immAllZerosV)),
402 (SUBREG_TO_REG (i8 0), (V_SET0), sub_xmm)>;
404 def : Pat<(v16i16 immAllZerosV), (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
405 def : Pat<(bc_v16i16 (v8f32 immAllZerosV)),
406 (SUBREG_TO_REG (i16 0), (V_SET0), sub_xmm)>;
408 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
409 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
410 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
412 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
413 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
414 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
417 // We set canFoldAsLoad because this can be converted to a constant-pool
418 // load of an all-ones value if folding it would be beneficial.
419 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
421 def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
422 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
423 let Predicates = [HasAVX2] in
424 def AVX2_SETALLONES : I<0, Pseudo, (outs VR256:$dst), (ins), "",
425 [(set VR256:$dst, (v8i32 immAllOnesV))]>;
429 //===----------------------------------------------------------------------===//
430 // SSE 1 & 2 - Move FP Scalar Instructions
432 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
433 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
434 // is used instead. Register-to-register movss/movsd is not modeled as an
435 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
436 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
437 //===----------------------------------------------------------------------===//
439 class sse12_move_rr<RegisterClass RC, SDNode OpNode, ValueType vt, string asm> :
440 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
441 [(set VR128:$dst, (vt (OpNode VR128:$src1,
442 (scalar_to_vector RC:$src2))))],
445 // Loading from memory automatically zeroing upper bits.
446 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
447 PatFrag mem_pat, string OpcodeStr> :
448 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
449 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
450 [(set RC:$dst, (mem_pat addr:$src))],
454 def VMOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
455 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
457 def VMOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
458 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
461 // For the disassembler
462 let isCodeGenOnly = 1, hasSideEffects = 0 in {
463 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
464 (ins VR128:$src1, FR32:$src2),
465 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
468 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
469 (ins VR128:$src1, FR64:$src2),
470 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
475 let canFoldAsLoad = 1, isReMaterializable = 1 in {
476 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
478 let AddedComplexity = 20 in
479 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
483 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
484 "movss\t{$src, $dst|$dst, $src}",
485 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
487 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
488 "movsd\t{$src, $dst|$dst, $src}",
489 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>,
493 let Constraints = "$src1 = $dst" in {
494 def MOVSSrr : sse12_move_rr<FR32, X86Movss, v4f32,
495 "movss\t{$src2, $dst|$dst, $src2}">, XS;
496 def MOVSDrr : sse12_move_rr<FR64, X86Movsd, v2f64,
497 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
499 // For the disassembler
500 let isCodeGenOnly = 1, hasSideEffects = 0 in {
501 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
502 (ins VR128:$src1, FR32:$src2),
503 "movss\t{$src2, $dst|$dst, $src2}", [],
504 IIC_SSE_MOV_S_RR>, XS;
505 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
506 (ins VR128:$src1, FR64:$src2),
507 "movsd\t{$src2, $dst|$dst, $src2}", [],
508 IIC_SSE_MOV_S_RR>, XD;
512 let canFoldAsLoad = 1, isReMaterializable = 1 in {
513 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
515 let AddedComplexity = 20 in
516 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
519 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
520 "movss\t{$src, $dst|$dst, $src}",
521 [(store FR32:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
522 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
523 "movsd\t{$src, $dst|$dst, $src}",
524 [(store FR64:$src, addr:$dst)], IIC_SSE_MOV_S_MR>;
527 let Predicates = [HasAVX] in {
528 let AddedComplexity = 15 in {
529 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
530 // MOVS{S,D} to the lower bits.
531 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
532 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
533 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
534 (VMOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
535 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
536 (VMOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
537 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
538 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
540 // Move low f32 and clear high bits.
541 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
542 (SUBREG_TO_REG (i32 0),
543 (VMOVSSrr (v4f32 (V_SET0)),
544 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm)), sub_xmm)>;
545 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
546 (SUBREG_TO_REG (i32 0),
547 (VMOVSSrr (v4i32 (V_SET0)),
548 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm)), sub_xmm)>;
551 let AddedComplexity = 20 in {
552 // MOVSSrm zeros the high parts of the register; represent this
553 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
554 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
555 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
556 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
557 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
558 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
559 (COPY_TO_REGCLASS (VMOVSSrm addr:$src), VR128)>;
561 // MOVSDrm zeros the high parts of the register; represent this
562 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
563 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
564 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
565 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
566 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
567 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
568 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
569 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
570 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
571 def : Pat<(v2f64 (X86vzload addr:$src)),
572 (COPY_TO_REGCLASS (VMOVSDrm addr:$src), VR128)>;
574 // Represent the same patterns above but in the form they appear for
576 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
577 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (iPTR 0)))),
578 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
579 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
580 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (iPTR 0)))),
581 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_xmm)>;
582 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
583 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (iPTR 0)))),
584 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_xmm)>;
586 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
587 (v4f32 (scalar_to_vector FR32:$src)), (iPTR 0)))),
588 (SUBREG_TO_REG (i32 0),
589 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
591 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
592 (v2f64 (scalar_to_vector FR64:$src)), (iPTR 0)))),
593 (SUBREG_TO_REG (i64 0),
594 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
596 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
597 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (iPTR 0)))),
598 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_xmm)>;
600 // Move low f64 and clear high bits.
601 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
602 (SUBREG_TO_REG (i32 0),
603 (VMOVSDrr (v2f64 (V_SET0)),
604 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm)), sub_xmm)>;
606 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
607 (SUBREG_TO_REG (i32 0),
608 (VMOVSDrr (v2i64 (V_SET0)),
609 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm)), sub_xmm)>;
611 // Extract and store.
612 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
614 (VMOVSSmr addr:$dst, (COPY_TO_REGCLASS (v4f32 VR128:$src), FR32))>;
615 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
617 (VMOVSDmr addr:$dst, (COPY_TO_REGCLASS (v2f64 VR128:$src), FR64))>;
619 // Shuffle with VMOVSS
620 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
621 (VMOVSSrr (v4i32 VR128:$src1),
622 (COPY_TO_REGCLASS (v4i32 VR128:$src2), FR32))>;
623 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
624 (VMOVSSrr (v4f32 VR128:$src1),
625 (COPY_TO_REGCLASS (v4f32 VR128:$src2), FR32))>;
628 def : Pat<(v8i32 (X86Movss VR256:$src1, VR256:$src2)),
629 (SUBREG_TO_REG (i32 0),
630 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_xmm),
631 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_xmm)),
633 def : Pat<(v8f32 (X86Movss VR256:$src1, VR256:$src2)),
634 (SUBREG_TO_REG (i32 0),
635 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_xmm),
636 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_xmm)),
639 // Shuffle with VMOVSD
640 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
641 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
642 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
643 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
644 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
645 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
646 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
647 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
650 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
651 (SUBREG_TO_REG (i32 0),
652 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_xmm),
653 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_xmm)),
655 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
656 (SUBREG_TO_REG (i32 0),
657 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_xmm),
658 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_xmm)),
662 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
663 // is during lowering, where it's not possible to recognize the fold cause
664 // it has two uses through a bitcast. One use disappears at isel time and the
665 // fold opportunity reappears.
666 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
667 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
668 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
669 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
670 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
672 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
673 (VMOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
676 let Predicates = [UseSSE1] in {
677 let AddedComplexity = 15 in {
678 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
679 // MOVSS to the lower bits.
680 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
681 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
682 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
683 (MOVSSrr (v4f32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
684 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
685 (MOVSSrr (v4i32 (V_SET0)), (COPY_TO_REGCLASS VR128:$src, FR32))>;
688 let AddedComplexity = 20 in {
689 // MOVSSrm already zeros the high parts of the register.
690 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
691 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
692 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
693 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
694 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
695 (COPY_TO_REGCLASS (MOVSSrm addr:$src), VR128)>;
698 // Extract and store.
699 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
701 (MOVSSmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR32))>;
703 // Shuffle with MOVSS
704 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
705 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
706 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
707 (MOVSSrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR32))>;
710 let Predicates = [UseSSE2] in {
711 let AddedComplexity = 15 in {
712 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
713 // MOVSD to the lower bits.
714 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
715 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
718 let AddedComplexity = 20 in {
719 // MOVSDrm already zeros the high parts of the register.
720 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
721 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
722 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
723 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
724 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
725 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
726 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
727 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
728 def : Pat<(v2f64 (X86vzload addr:$src)),
729 (COPY_TO_REGCLASS (MOVSDrm addr:$src), VR128)>;
732 // Extract and store.
733 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
735 (MOVSDmr addr:$dst, (COPY_TO_REGCLASS VR128:$src, FR64))>;
737 // Shuffle with MOVSD
738 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
739 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
740 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
741 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
742 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
743 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
744 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
745 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
747 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
748 // is during lowering, where it's not possible to recognize the fold cause
749 // it has two uses through a bitcast. One use disappears at isel time and the
750 // fold opportunity reappears.
751 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
752 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
753 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
754 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
755 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
756 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
757 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
758 (MOVSDrr VR128:$src1, (COPY_TO_REGCLASS VR128:$src2, FR64))>;
761 //===----------------------------------------------------------------------===//
762 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
763 //===----------------------------------------------------------------------===//
765 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
766 X86MemOperand x86memop, PatFrag ld_frag,
767 string asm, Domain d,
769 bit IsReMaterializable = 1> {
770 let neverHasSideEffects = 1 in
771 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
772 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], itins.rr, d>;
773 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
774 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
775 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
776 [(set RC:$dst, (ld_frag addr:$src))], itins.rm, d>;
779 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
780 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
782 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
783 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
785 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
786 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
788 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
789 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
792 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
793 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
795 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
796 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
797 TB, OpSize, VEX, VEX_L;
798 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
799 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
801 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
802 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
803 TB, OpSize, VEX, VEX_L;
804 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
805 "movaps", SSEPackedSingle, SSE_MOVA_ITINS>,
807 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
808 "movapd", SSEPackedDouble, SSE_MOVA_ITINS>,
810 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
811 "movups", SSEPackedSingle, SSE_MOVU_ITINS>,
813 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
814 "movupd", SSEPackedDouble, SSE_MOVU_ITINS, 0>,
817 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
818 "movaps\t{$src, $dst|$dst, $src}",
819 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
820 IIC_SSE_MOVA_P_MR>, VEX;
821 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
822 "movapd\t{$src, $dst|$dst, $src}",
823 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
824 IIC_SSE_MOVA_P_MR>, VEX;
825 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
826 "movups\t{$src, $dst|$dst, $src}",
827 [(store (v4f32 VR128:$src), addr:$dst)],
828 IIC_SSE_MOVU_P_MR>, VEX;
829 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
830 "movupd\t{$src, $dst|$dst, $src}",
831 [(store (v2f64 VR128:$src), addr:$dst)],
832 IIC_SSE_MOVU_P_MR>, VEX;
833 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
834 "movaps\t{$src, $dst|$dst, $src}",
835 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)],
836 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
837 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
838 "movapd\t{$src, $dst|$dst, $src}",
839 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)],
840 IIC_SSE_MOVA_P_MR>, VEX, VEX_L;
841 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
842 "movups\t{$src, $dst|$dst, $src}",
843 [(store (v8f32 VR256:$src), addr:$dst)],
844 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
845 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
846 "movupd\t{$src, $dst|$dst, $src}",
847 [(store (v4f64 VR256:$src), addr:$dst)],
848 IIC_SSE_MOVU_P_MR>, VEX, VEX_L;
851 let isCodeGenOnly = 1, hasSideEffects = 0 in {
852 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
854 "movaps\t{$src, $dst|$dst, $src}", [],
855 IIC_SSE_MOVA_P_RR>, VEX;
856 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
858 "movapd\t{$src, $dst|$dst, $src}", [],
859 IIC_SSE_MOVA_P_RR>, VEX;
860 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
862 "movups\t{$src, $dst|$dst, $src}", [],
863 IIC_SSE_MOVU_P_RR>, VEX;
864 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
866 "movupd\t{$src, $dst|$dst, $src}", [],
867 IIC_SSE_MOVU_P_RR>, VEX;
868 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
870 "movaps\t{$src, $dst|$dst, $src}", [],
871 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
872 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
874 "movapd\t{$src, $dst|$dst, $src}", [],
875 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
876 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
878 "movups\t{$src, $dst|$dst, $src}", [],
879 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
880 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
882 "movupd\t{$src, $dst|$dst, $src}", [],
883 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
886 let Predicates = [HasAVX] in {
887 def : Pat<(v8i32 (X86vzmovl
888 (insert_subvector undef, (v4i32 VR128:$src), (iPTR 0)))),
889 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
890 def : Pat<(v4i64 (X86vzmovl
891 (insert_subvector undef, (v2i64 VR128:$src), (iPTR 0)))),
892 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
893 def : Pat<(v8f32 (X86vzmovl
894 (insert_subvector undef, (v4f32 VR128:$src), (iPTR 0)))),
895 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
896 def : Pat<(v4f64 (X86vzmovl
897 (insert_subvector undef, (v2f64 VR128:$src), (iPTR 0)))),
898 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
902 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
903 (VMOVUPSYmr addr:$dst, VR256:$src)>;
904 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
905 (VMOVUPDYmr addr:$dst, VR256:$src)>;
907 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
908 "movaps\t{$src, $dst|$dst, $src}",
909 [(alignedstore (v4f32 VR128:$src), addr:$dst)],
911 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
912 "movapd\t{$src, $dst|$dst, $src}",
913 [(alignedstore (v2f64 VR128:$src), addr:$dst)],
915 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
916 "movups\t{$src, $dst|$dst, $src}",
917 [(store (v4f32 VR128:$src), addr:$dst)],
919 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
920 "movupd\t{$src, $dst|$dst, $src}",
921 [(store (v2f64 VR128:$src), addr:$dst)],
925 let isCodeGenOnly = 1, hasSideEffects = 0 in {
926 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
927 "movaps\t{$src, $dst|$dst, $src}", [],
929 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
930 "movapd\t{$src, $dst|$dst, $src}", [],
932 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
933 "movups\t{$src, $dst|$dst, $src}", [],
935 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
936 "movupd\t{$src, $dst|$dst, $src}", [],
940 let Predicates = [HasAVX] in {
941 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
942 (VMOVUPSmr addr:$dst, VR128:$src)>;
943 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
944 (VMOVUPDmr addr:$dst, VR128:$src)>;
947 let Predicates = [UseSSE1] in
948 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
949 (MOVUPSmr addr:$dst, VR128:$src)>;
950 let Predicates = [UseSSE2] in
951 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
952 (MOVUPDmr addr:$dst, VR128:$src)>;
954 // Use vmovaps/vmovups for AVX integer load/store.
955 let Predicates = [HasAVX] in {
956 // 128-bit load/store
957 def : Pat<(alignedloadv2i64 addr:$src),
958 (VMOVAPSrm addr:$src)>;
959 def : Pat<(loadv2i64 addr:$src),
960 (VMOVUPSrm addr:$src)>;
962 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
963 (VMOVAPSmr addr:$dst, VR128:$src)>;
964 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
965 (VMOVAPSmr addr:$dst, VR128:$src)>;
966 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
967 (VMOVAPSmr addr:$dst, VR128:$src)>;
968 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
969 (VMOVAPSmr addr:$dst, VR128:$src)>;
970 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
971 (VMOVUPSmr addr:$dst, VR128:$src)>;
972 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
973 (VMOVUPSmr addr:$dst, VR128:$src)>;
974 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
975 (VMOVUPSmr addr:$dst, VR128:$src)>;
976 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
977 (VMOVUPSmr addr:$dst, VR128:$src)>;
979 // 256-bit load/store
980 def : Pat<(alignedloadv4i64 addr:$src),
981 (VMOVAPSYrm addr:$src)>;
982 def : Pat<(loadv4i64 addr:$src),
983 (VMOVUPSYrm addr:$src)>;
984 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
985 (VMOVAPSYmr addr:$dst, VR256:$src)>;
986 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
987 (VMOVAPSYmr addr:$dst, VR256:$src)>;
988 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
989 (VMOVAPSYmr addr:$dst, VR256:$src)>;
990 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
991 (VMOVAPSYmr addr:$dst, VR256:$src)>;
992 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
993 (VMOVUPSYmr addr:$dst, VR256:$src)>;
994 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
995 (VMOVUPSYmr addr:$dst, VR256:$src)>;
996 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
997 (VMOVUPSYmr addr:$dst, VR256:$src)>;
998 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
999 (VMOVUPSYmr addr:$dst, VR256:$src)>;
1001 // Special patterns for storing subvector extracts of lower 128-bits
1002 // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
1003 def : Pat<(alignedstore (v2f64 (extract_subvector
1004 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1005 (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1006 def : Pat<(alignedstore (v4f32 (extract_subvector
1007 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1008 (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1009 def : Pat<(alignedstore (v2i64 (extract_subvector
1010 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1011 (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1012 def : Pat<(alignedstore (v4i32 (extract_subvector
1013 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1014 (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1015 def : Pat<(alignedstore (v8i16 (extract_subvector
1016 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1017 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1018 def : Pat<(alignedstore (v16i8 (extract_subvector
1019 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1020 (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1022 def : Pat<(store (v2f64 (extract_subvector
1023 (v4f64 VR256:$src), (iPTR 0))), addr:$dst),
1024 (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1025 def : Pat<(store (v4f32 (extract_subvector
1026 (v8f32 VR256:$src), (iPTR 0))), addr:$dst),
1027 (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1028 def : Pat<(store (v2i64 (extract_subvector
1029 (v4i64 VR256:$src), (iPTR 0))), addr:$dst),
1030 (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1031 def : Pat<(store (v4i32 (extract_subvector
1032 (v8i32 VR256:$src), (iPTR 0))), addr:$dst),
1033 (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1034 def : Pat<(store (v8i16 (extract_subvector
1035 (v16i16 VR256:$src), (iPTR 0))), addr:$dst),
1036 (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1037 def : Pat<(store (v16i8 (extract_subvector
1038 (v32i8 VR256:$src), (iPTR 0))), addr:$dst),
1039 (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
1042 // Use movaps / movups for SSE integer load / store (one byte shorter).
1043 // The instructions selected below are then converted to MOVDQA/MOVDQU
1044 // during the SSE domain pass.
1045 let Predicates = [UseSSE1] in {
1046 def : Pat<(alignedloadv2i64 addr:$src),
1047 (MOVAPSrm addr:$src)>;
1048 def : Pat<(loadv2i64 addr:$src),
1049 (MOVUPSrm addr:$src)>;
1051 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
1052 (MOVAPSmr addr:$dst, VR128:$src)>;
1053 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
1054 (MOVAPSmr addr:$dst, VR128:$src)>;
1055 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
1056 (MOVAPSmr addr:$dst, VR128:$src)>;
1057 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
1058 (MOVAPSmr addr:$dst, VR128:$src)>;
1059 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1060 (MOVUPSmr addr:$dst, VR128:$src)>;
1061 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1062 (MOVUPSmr addr:$dst, VR128:$src)>;
1063 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1064 (MOVUPSmr addr:$dst, VR128:$src)>;
1065 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1066 (MOVUPSmr addr:$dst, VR128:$src)>;
1069 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1070 // bits are disregarded. FIXME: Set encoding to pseudo!
1071 let neverHasSideEffects = 1 in {
1072 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1073 "movaps\t{$src, $dst|$dst, $src}", [],
1074 IIC_SSE_MOVA_P_RR>, VEX;
1075 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1076 "movapd\t{$src, $dst|$dst, $src}", [],
1077 IIC_SSE_MOVA_P_RR>, VEX;
1078 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1079 "movaps\t{$src, $dst|$dst, $src}", [],
1081 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1082 "movapd\t{$src, $dst|$dst, $src}", [],
1086 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1087 // bits are disregarded. FIXME: Set encoding to pseudo!
1088 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1089 let isCodeGenOnly = 1 in {
1090 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1091 "movaps\t{$src, $dst|$dst, $src}",
1092 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1093 IIC_SSE_MOVA_P_RM>, VEX;
1094 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1095 "movapd\t{$src, $dst|$dst, $src}",
1096 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1097 IIC_SSE_MOVA_P_RM>, VEX;
1099 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1100 "movaps\t{$src, $dst|$dst, $src}",
1101 [(set FR32:$dst, (alignedloadfsf32 addr:$src))],
1103 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1104 "movapd\t{$src, $dst|$dst, $src}",
1105 [(set FR64:$dst, (alignedloadfsf64 addr:$src))],
1109 //===----------------------------------------------------------------------===//
1110 // SSE 1 & 2 - Move Low packed FP Instructions
1111 //===----------------------------------------------------------------------===//
1113 multiclass sse12_mov_hilo_packed_base<bits<8>opc, SDNode psnode, SDNode pdnode,
1114 string base_opc, string asm_opr,
1115 InstrItinClass itin> {
1116 def PSrm : PI<opc, MRMSrcMem,
1117 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1118 !strconcat(base_opc, "s", asm_opr),
1120 (psnode VR128:$src1,
1121 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1122 itin, SSEPackedSingle>, TB;
1124 def PDrm : PI<opc, MRMSrcMem,
1125 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1126 !strconcat(base_opc, "d", asm_opr),
1127 [(set VR128:$dst, (v2f64 (pdnode VR128:$src1,
1128 (scalar_to_vector (loadf64 addr:$src2)))))],
1129 itin, SSEPackedDouble>, TB, OpSize;
1133 multiclass sse12_mov_hilo_packed<bits<8>opc, SDNode psnode, SDNode pdnode,
1134 string base_opc, InstrItinClass itin> {
1135 defm V#NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1136 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1139 let Constraints = "$src1 = $dst" in
1140 defm NAME : sse12_mov_hilo_packed_base<opc, psnode, pdnode, base_opc,
1141 "\t{$src2, $dst|$dst, $src2}",
1145 let AddedComplexity = 20 in {
1146 defm MOVL : sse12_mov_hilo_packed<0x12, X86Movlps, X86Movlpd, "movlp",
1150 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1151 "movlps\t{$src, $dst|$dst, $src}",
1152 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1153 (iPTR 0))), addr:$dst)],
1154 IIC_SSE_MOV_LH>, VEX;
1155 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1156 "movlpd\t{$src, $dst|$dst, $src}",
1157 [(store (f64 (vector_extract (v2f64 VR128:$src),
1158 (iPTR 0))), addr:$dst)],
1159 IIC_SSE_MOV_LH>, VEX;
1160 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1161 "movlps\t{$src, $dst|$dst, $src}",
1162 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1163 (iPTR 0))), addr:$dst)],
1165 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1166 "movlpd\t{$src, $dst|$dst, $src}",
1167 [(store (f64 (vector_extract (v2f64 VR128:$src),
1168 (iPTR 0))), addr:$dst)],
1171 let Predicates = [HasAVX] in {
1172 // Shuffle with VMOVLPS
1173 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1174 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1175 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1176 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1178 // Shuffle with VMOVLPD
1179 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1180 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1181 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1182 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1185 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1187 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1188 def : Pat<(store (v4i32 (X86Movlps
1189 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1190 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1191 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1193 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1194 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1196 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1199 let Predicates = [UseSSE1] in {
1200 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1201 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1202 (iPTR 0))), addr:$src1),
1203 (MOVLPSmr addr:$src1, VR128:$src2)>;
1205 // Shuffle with MOVLPS
1206 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1207 (MOVLPSrm VR128:$src1, addr:$src2)>;
1208 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1209 (MOVLPSrm VR128:$src1, addr:$src2)>;
1210 def : Pat<(X86Movlps VR128:$src1,
1211 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1212 (MOVLPSrm VR128:$src1, addr:$src2)>;
1215 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1217 (MOVLPSmr addr:$src1, VR128:$src2)>;
1218 def : Pat<(store (v4i32 (X86Movlps
1219 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1221 (MOVLPSmr addr:$src1, VR128:$src2)>;
1224 let Predicates = [UseSSE2] in {
1225 // Shuffle with MOVLPD
1226 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1227 (MOVLPDrm VR128:$src1, addr:$src2)>;
1228 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1229 (MOVLPDrm VR128:$src1, addr:$src2)>;
1232 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1234 (MOVLPDmr addr:$src1, VR128:$src2)>;
1235 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1237 (MOVLPDmr addr:$src1, VR128:$src2)>;
1240 //===----------------------------------------------------------------------===//
1241 // SSE 1 & 2 - Move Hi packed FP Instructions
1242 //===----------------------------------------------------------------------===//
1244 let AddedComplexity = 20 in {
1245 defm MOVH : sse12_mov_hilo_packed<0x16, X86Movlhps, X86Movlhpd, "movhp",
1249 // v2f64 extract element 1 is always custom lowered to unpack high to low
1250 // and extract element 0 so the non-store version isn't too horrible.
1251 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1252 "movhps\t{$src, $dst|$dst, $src}",
1253 [(store (f64 (vector_extract
1254 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1255 (bc_v2f64 (v4f32 VR128:$src))),
1256 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1257 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1258 "movhpd\t{$src, $dst|$dst, $src}",
1259 [(store (f64 (vector_extract
1260 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1261 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1262 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1263 "movhps\t{$src, $dst|$dst, $src}",
1264 [(store (f64 (vector_extract
1265 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1266 (bc_v2f64 (v4f32 VR128:$src))),
1267 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1268 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1269 "movhpd\t{$src, $dst|$dst, $src}",
1270 [(store (f64 (vector_extract
1271 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1272 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1274 let Predicates = [HasAVX] in {
1276 def : Pat<(X86Movlhps VR128:$src1,
1277 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1278 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1279 def : Pat<(X86Movlhps VR128:$src1,
1280 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1281 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1283 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1284 // is during lowering, where it's not possible to recognize the load fold
1285 // cause it has two uses through a bitcast. One use disappears at isel time
1286 // and the fold opportunity reappears.
1287 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1288 (scalar_to_vector (loadf64 addr:$src2)))),
1289 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1292 let Predicates = [UseSSE1] in {
1294 def : Pat<(X86Movlhps VR128:$src1,
1295 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1296 (MOVHPSrm VR128:$src1, addr:$src2)>;
1297 def : Pat<(X86Movlhps VR128:$src1,
1298 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1299 (MOVHPSrm VR128:$src1, addr:$src2)>;
1302 let Predicates = [UseSSE2] in {
1303 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1304 // is during lowering, where it's not possible to recognize the load fold
1305 // cause it has two uses through a bitcast. One use disappears at isel time
1306 // and the fold opportunity reappears.
1307 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1308 (scalar_to_vector (loadf64 addr:$src2)))),
1309 (MOVHPDrm VR128:$src1, addr:$src2)>;
1312 //===----------------------------------------------------------------------===//
1313 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1314 //===----------------------------------------------------------------------===//
1316 let AddedComplexity = 20 in {
1317 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1318 (ins VR128:$src1, VR128:$src2),
1319 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1321 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1324 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1325 (ins VR128:$src1, VR128:$src2),
1326 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1328 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1332 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1333 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1334 (ins VR128:$src1, VR128:$src2),
1335 "movlhps\t{$src2, $dst|$dst, $src2}",
1337 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1339 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1340 (ins VR128:$src1, VR128:$src2),
1341 "movhlps\t{$src2, $dst|$dst, $src2}",
1343 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1347 let Predicates = [HasAVX] in {
1349 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1350 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1351 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1352 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1355 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1356 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1359 let Predicates = [UseSSE1] in {
1361 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1362 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1363 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1364 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1367 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1368 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1371 //===----------------------------------------------------------------------===//
1372 // SSE 1 & 2 - Conversion Instructions
1373 //===----------------------------------------------------------------------===//
1375 def SSE_CVT_PD : OpndItins<
1376 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1379 def SSE_CVT_PS : OpndItins<
1380 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1383 def SSE_CVT_Scalar : OpndItins<
1384 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1387 def SSE_CVT_SS2SI_32 : OpndItins<
1388 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1391 def SSE_CVT_SS2SI_64 : OpndItins<
1392 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1395 def SSE_CVT_SD2SI : OpndItins<
1396 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1399 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1400 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1401 string asm, OpndItins itins> {
1402 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1403 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1405 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1406 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1410 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1411 X86MemOperand x86memop, string asm, Domain d,
1413 let neverHasSideEffects = 1 in {
1414 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1417 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1422 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1423 X86MemOperand x86memop, string asm> {
1424 let neverHasSideEffects = 1 in {
1425 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1426 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1428 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1429 (ins DstRC:$src1, x86memop:$src),
1430 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1431 } // neverHasSideEffects = 1
1434 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1435 "cvttss2si\t{$src, $dst|$dst, $src}",
1438 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1439 "cvttss2si\t{$src, $dst|$dst, $src}",
1441 XS, VEX, VEX_W, VEX_LIG;
1442 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1443 "cvttsd2si\t{$src, $dst|$dst, $src}",
1446 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1447 "cvttsd2si\t{$src, $dst|$dst, $src}",
1449 XD, VEX, VEX_W, VEX_LIG;
1451 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1452 (VCVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1453 def : InstAlias<"vcvttss2si{l}\t{$src, $dst|$dst, $src}",
1454 (VCVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1455 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1456 (VCVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1457 def : InstAlias<"vcvttsd2si{l}\t{$src, $dst|$dst, $src}",
1458 (VCVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1459 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1460 (VCVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1461 def : InstAlias<"vcvttss2si{q}\t{$src, $dst|$dst, $src}",
1462 (VCVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1463 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1464 (VCVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1465 def : InstAlias<"vcvttsd2si{q}\t{$src, $dst|$dst, $src}",
1466 (VCVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1468 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1469 // register, but the same isn't true when only using memory operands,
1470 // provide other assembly "l" and "q" forms to address this explicitly
1471 // where appropriate to do so.
1472 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss{l}">,
1473 XS, VEX_4V, VEX_LIG;
1474 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1475 XS, VEX_4V, VEX_W, VEX_LIG;
1476 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">,
1477 XD, VEX_4V, VEX_LIG;
1478 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1479 XD, VEX_4V, VEX_W, VEX_LIG;
1481 def : InstAlias<"vcvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1482 (VCVTSI2SSrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1483 def : InstAlias<"vcvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1484 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1486 let Predicates = [HasAVX] in {
1487 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1488 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1489 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1490 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1491 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1492 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1493 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1494 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1496 def : Pat<(f32 (sint_to_fp GR32:$src)),
1497 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1498 def : Pat<(f32 (sint_to_fp GR64:$src)),
1499 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1500 def : Pat<(f64 (sint_to_fp GR32:$src)),
1501 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1502 def : Pat<(f64 (sint_to_fp GR64:$src)),
1503 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1506 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1507 "cvttss2si\t{$src, $dst|$dst, $src}",
1508 SSE_CVT_SS2SI_32>, XS;
1509 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1510 "cvttss2si\t{$src, $dst|$dst, $src}",
1511 SSE_CVT_SS2SI_64>, XS, REX_W;
1512 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1513 "cvttsd2si\t{$src, $dst|$dst, $src}",
1515 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1516 "cvttsd2si\t{$src, $dst|$dst, $src}",
1517 SSE_CVT_SD2SI>, XD, REX_W;
1518 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1519 "cvtsi2ss{l}\t{$src, $dst|$dst, $src}",
1520 SSE_CVT_Scalar>, XS;
1521 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1522 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1523 SSE_CVT_Scalar>, XS, REX_W;
1524 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1525 "cvtsi2sd{l}\t{$src, $dst|$dst, $src}",
1526 SSE_CVT_Scalar>, XD;
1527 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1528 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1529 SSE_CVT_Scalar>, XD, REX_W;
1531 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1532 (CVTTSS2SIrr GR32:$dst, FR32:$src), 0>;
1533 def : InstAlias<"cvttss2si{l}\t{$src, $dst|$dst, $src}",
1534 (CVTTSS2SIrm GR32:$dst, f32mem:$src), 0>;
1535 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1536 (CVTTSD2SIrr GR32:$dst, FR64:$src), 0>;
1537 def : InstAlias<"cvttsd2si{l}\t{$src, $dst|$dst, $src}",
1538 (CVTTSD2SIrm GR32:$dst, f64mem:$src), 0>;
1539 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1540 (CVTTSS2SI64rr GR64:$dst, FR32:$src), 0>;
1541 def : InstAlias<"cvttss2si{q}\t{$src, $dst|$dst, $src}",
1542 (CVTTSS2SI64rm GR64:$dst, f32mem:$src), 0>;
1543 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1544 (CVTTSD2SI64rr GR64:$dst, FR64:$src), 0>;
1545 def : InstAlias<"cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1546 (CVTTSD2SI64rm GR64:$dst, f64mem:$src), 0>;
1548 def : InstAlias<"cvtsi2ss\t{$src, $dst|$dst, $src}",
1549 (CVTSI2SSrm FR64:$dst, i32mem:$src)>;
1550 def : InstAlias<"cvtsi2sd\t{$src, $dst|$dst, $src}",
1551 (CVTSI2SDrm FR64:$dst, i32mem:$src)>;
1553 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1554 // and/or XMM operand(s).
1556 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1557 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1558 string asm, OpndItins itins> {
1559 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1560 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1561 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1562 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1563 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1564 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1567 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1568 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1569 PatFrag ld_frag, string asm, OpndItins itins,
1571 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1573 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1574 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1575 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1577 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1578 (ins DstRC:$src1, x86memop:$src2),
1580 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1581 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1582 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1586 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1587 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si",
1588 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1589 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1590 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si",
1591 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1593 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1594 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD;
1595 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1596 sdmem, sse_load_f64, "cvtsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1599 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1600 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss{l}",
1601 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1602 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1603 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1604 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1606 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1607 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd{l}",
1608 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1609 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1610 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1611 SSE_CVT_Scalar, 0>, XD,
1614 let Constraints = "$src1 = $dst" in {
1615 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1616 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1617 "cvtsi2ss{l}", SSE_CVT_Scalar>, XS;
1618 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1619 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1620 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1621 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1622 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1623 "cvtsi2sd{l}", SSE_CVT_Scalar>, XD;
1624 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1625 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1626 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1631 // Aliases for intrinsics
1632 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1633 ssmem, sse_load_f32, "cvttss2si",
1634 SSE_CVT_SS2SI_32>, XS, VEX;
1635 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1636 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1637 "cvttss2si", SSE_CVT_SS2SI_64>,
1639 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1640 sdmem, sse_load_f64, "cvttsd2si",
1641 SSE_CVT_SD2SI>, XD, VEX;
1642 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1643 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1644 "cvttsd2si", SSE_CVT_SD2SI>,
1646 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1647 ssmem, sse_load_f32, "cvttss2si",
1648 SSE_CVT_SS2SI_32>, XS;
1649 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1650 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1651 "cvttss2si", SSE_CVT_SS2SI_64>, XS, REX_W;
1652 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1653 sdmem, sse_load_f64, "cvttsd2si",
1655 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1656 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1657 "cvttsd2si", SSE_CVT_SD2SI>, XD, REX_W;
1659 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1660 ssmem, sse_load_f32, "cvtss2si",
1661 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1662 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1663 ssmem, sse_load_f32, "cvtss2si",
1664 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1666 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1667 ssmem, sse_load_f32, "cvtss2si",
1668 SSE_CVT_SS2SI_32>, XS;
1669 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1670 ssmem, sse_load_f32, "cvtss2si",
1671 SSE_CVT_SS2SI_64>, XS, REX_W;
1673 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1674 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1675 SSEPackedSingle, SSE_CVT_PS>,
1676 TB, VEX, Requires<[HasAVX]>;
1677 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1678 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1679 SSEPackedSingle, SSE_CVT_PS>,
1680 TB, VEX, VEX_L, Requires<[HasAVX]>;
1682 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1683 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1684 SSEPackedSingle, SSE_CVT_PS>,
1685 TB, Requires<[UseSSE2]>;
1687 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1688 (VCVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1689 def : InstAlias<"vcvtss2si{l}\t{$src, $dst|$dst, $src}",
1690 (VCVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1691 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1692 (VCVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1693 def : InstAlias<"vcvtsd2si{l}\t{$src, $dst|$dst, $src}",
1694 (VCVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1695 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1696 (VCVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1697 def : InstAlias<"vcvtss2si{q}\t{$src, $dst|$dst, $src}",
1698 (VCVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1699 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1700 (VCVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1701 def : InstAlias<"vcvtsd2si{q}\t{$src, $dst|$dst, $src}",
1702 (VCVTSD2SI64rm GR64:$dst, sdmem:$src), 0>;
1704 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1705 (CVTSS2SIrr GR32:$dst, VR128:$src), 0>;
1706 def : InstAlias<"cvtss2si{l}\t{$src, $dst|$dst, $src}",
1707 (CVTSS2SIrm GR32:$dst, ssmem:$src), 0>;
1708 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1709 (CVTSD2SIrr GR32:$dst, VR128:$src), 0>;
1710 def : InstAlias<"cvtsd2si{l}\t{$src, $dst|$dst, $src}",
1711 (CVTSD2SIrm GR32:$dst, sdmem:$src), 0>;
1712 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1713 (CVTSS2SI64rr GR64:$dst, VR128:$src), 0>;
1714 def : InstAlias<"cvtss2si{q}\t{$src, $dst|$dst, $src}",
1715 (CVTSS2SI64rm GR64:$dst, ssmem:$src), 0>;
1716 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1717 (CVTSD2SI64rr GR64:$dst, VR128:$src), 0>;
1718 def : InstAlias<"cvtsd2si{q}\t{$src, $dst|$dst, $src}",
1719 (CVTSD2SI64rm GR64:$dst, sdmem:$src)>;
1723 // Convert scalar double to scalar single
1724 let neverHasSideEffects = 1 in {
1725 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1726 (ins FR64:$src1, FR64:$src2),
1727 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1728 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1730 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1731 (ins FR64:$src1, f64mem:$src2),
1732 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1733 [], IIC_SSE_CVT_Scalar_RM>,
1734 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1737 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1740 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1741 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1742 [(set FR32:$dst, (fround FR64:$src))],
1743 IIC_SSE_CVT_Scalar_RR>;
1744 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1745 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1746 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1747 IIC_SSE_CVT_Scalar_RM>,
1749 Requires<[UseSSE2, OptForSize]>;
1751 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1752 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1753 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1755 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1756 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
1757 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1758 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1759 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1760 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1761 VR128:$src1, sse_load_f64:$src2))],
1762 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
1764 let Constraints = "$src1 = $dst" in {
1765 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1766 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1767 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1769 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1770 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
1771 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1772 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1773 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1774 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1775 VR128:$src1, sse_load_f64:$src2))],
1776 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
1779 // Convert scalar single to scalar double
1780 // SSE2 instructions with XS prefix
1781 let neverHasSideEffects = 1 in {
1782 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1783 (ins FR32:$src1, FR32:$src2),
1784 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1785 [], IIC_SSE_CVT_Scalar_RR>,
1786 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1788 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1789 (ins FR32:$src1, f32mem:$src2),
1790 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1791 [], IIC_SSE_CVT_Scalar_RM>,
1792 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1795 def : Pat<(f64 (fextend FR32:$src)),
1796 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1797 def : Pat<(fextend (loadf32 addr:$src)),
1798 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1800 def : Pat<(extloadf32 addr:$src),
1801 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1802 Requires<[HasAVX, OptForSize]>;
1803 def : Pat<(extloadf32 addr:$src),
1804 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1805 Requires<[HasAVX, OptForSpeed]>;
1807 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1808 "cvtss2sd\t{$src, $dst|$dst, $src}",
1809 [(set FR64:$dst, (fextend FR32:$src))],
1810 IIC_SSE_CVT_Scalar_RR>, XS,
1811 Requires<[UseSSE2]>;
1812 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1813 "cvtss2sd\t{$src, $dst|$dst, $src}",
1814 [(set FR64:$dst, (extloadf32 addr:$src))],
1815 IIC_SSE_CVT_Scalar_RM>, XS,
1816 Requires<[UseSSE2, OptForSize]>;
1818 // extload f32 -> f64. This matches load+fextend because we have a hack in
1819 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1821 // Since these loads aren't folded into the fextend, we have to match it
1823 def : Pat<(fextend (loadf32 addr:$src)),
1824 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1825 def : Pat<(extloadf32 addr:$src),
1826 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1828 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1829 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1830 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1832 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1833 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
1834 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1835 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1836 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1838 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1839 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
1840 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1841 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1842 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1843 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1845 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1846 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
1847 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1848 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1849 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1851 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1852 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
1855 // Convert packed single/double fp to doubleword
1856 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1857 "cvtps2dq\t{$src, $dst|$dst, $src}",
1858 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1859 IIC_SSE_CVT_PS_RR>, VEX;
1860 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1861 "cvtps2dq\t{$src, $dst|$dst, $src}",
1863 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1864 IIC_SSE_CVT_PS_RM>, VEX;
1865 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1866 "cvtps2dq\t{$src, $dst|$dst, $src}",
1868 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1869 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1870 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1871 "cvtps2dq\t{$src, $dst|$dst, $src}",
1873 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1874 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1875 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1876 "cvtps2dq\t{$src, $dst|$dst, $src}",
1877 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1879 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1880 "cvtps2dq\t{$src, $dst|$dst, $src}",
1882 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1886 // Convert Packed Double FP to Packed DW Integers
1887 let Predicates = [HasAVX] in {
1888 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1889 // register, but the same isn't true when using memory operands instead.
1890 // Provide other assembly rr and rm forms to address this explicitly.
1891 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1892 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1893 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1897 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1898 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1899 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1900 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1902 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1905 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1906 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1908 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L;
1909 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1910 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1912 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1914 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1915 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1918 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1919 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1921 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1923 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1924 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1925 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1928 // Convert with truncation packed single/double fp to doubleword
1929 // SSE2 packed instructions with XS prefix
1930 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1931 "cvttps2dq\t{$src, $dst|$dst, $src}",
1933 (int_x86_sse2_cvttps2dq VR128:$src))],
1934 IIC_SSE_CVT_PS_RR>, VEX;
1935 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1936 "cvttps2dq\t{$src, $dst|$dst, $src}",
1937 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1938 (memopv4f32 addr:$src)))],
1939 IIC_SSE_CVT_PS_RM>, VEX;
1940 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1941 "cvttps2dq\t{$src, $dst|$dst, $src}",
1943 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1944 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1945 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1946 "cvttps2dq\t{$src, $dst|$dst, $src}",
1947 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1948 (memopv8f32 addr:$src)))],
1949 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1951 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1952 "cvttps2dq\t{$src, $dst|$dst, $src}",
1953 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1955 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1956 "cvttps2dq\t{$src, $dst|$dst, $src}",
1958 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1961 let Predicates = [HasAVX] in {
1962 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1963 (VCVTDQ2PSrr VR128:$src)>;
1964 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1965 (VCVTDQ2PSrm addr:$src)>;
1967 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1968 (VCVTDQ2PSrr VR128:$src)>;
1969 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1970 (VCVTDQ2PSrm addr:$src)>;
1972 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1973 (VCVTTPS2DQrr VR128:$src)>;
1974 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1975 (VCVTTPS2DQrm addr:$src)>;
1977 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1978 (VCVTDQ2PSYrr VR256:$src)>;
1979 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1980 (VCVTDQ2PSYrm addr:$src)>;
1982 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1983 (VCVTTPS2DQYrr VR256:$src)>;
1984 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1985 (VCVTTPS2DQYrm addr:$src)>;
1988 let Predicates = [UseSSE2] in {
1989 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1990 (CVTDQ2PSrr VR128:$src)>;
1991 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1992 (CVTDQ2PSrm addr:$src)>;
1994 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1995 (CVTDQ2PSrr VR128:$src)>;
1996 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1997 (CVTDQ2PSrm addr:$src)>;
1999 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
2000 (CVTTPS2DQrr VR128:$src)>;
2001 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
2002 (CVTTPS2DQrm addr:$src)>;
2005 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2006 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2008 (int_x86_sse2_cvttpd2dq VR128:$src))],
2009 IIC_SSE_CVT_PD_RR>, VEX;
2011 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2012 // register, but the same isn't true when using memory operands instead.
2013 // Provide other assembly rr and rm forms to address this explicitly.
2016 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
2017 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
2018 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2019 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
2020 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2021 (memopv2f64 addr:$src)))],
2022 IIC_SSE_CVT_PD_RM>, VEX;
2025 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2026 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2028 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
2029 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
2030 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2031 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
2033 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
2034 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2035 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
2036 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
2038 let Predicates = [HasAVX] in {
2039 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
2040 (VCVTTPD2DQYrr VR256:$src)>;
2041 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
2042 (VCVTTPD2DQYrm addr:$src)>;
2043 } // Predicates = [HasAVX]
2045 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2046 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2047 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
2049 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
2050 "cvttpd2dq\t{$src, $dst|$dst, $src}",
2051 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
2052 (memopv2f64 addr:$src)))],
2055 // Convert packed single to packed double
2056 let Predicates = [HasAVX] in {
2057 // SSE2 instructions without OpSize prefix
2058 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2059 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2060 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2061 IIC_SSE_CVT_PD_RR>, TB, VEX;
2062 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2063 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2064 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2065 IIC_SSE_CVT_PD_RM>, TB, VEX;
2066 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2067 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2069 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
2070 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L;
2071 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
2072 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2074 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2075 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L;
2078 let Predicates = [UseSSE2] in {
2079 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2080 "cvtps2pd\t{$src, $dst|$dst, $src}",
2081 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2082 IIC_SSE_CVT_PD_RR>, TB;
2083 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2084 "cvtps2pd\t{$src, $dst|$dst, $src}",
2085 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2086 IIC_SSE_CVT_PD_RM>, TB;
2089 // Convert Packed DW Integers to Packed Double FP
2090 let Predicates = [HasAVX] in {
2091 let neverHasSideEffects = 1, mayLoad = 1 in
2092 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2093 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2095 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2096 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2098 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2099 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2100 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2102 (int_x86_avx_cvtdq2_pd_256
2103 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L;
2104 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2105 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2107 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L;
2110 let neverHasSideEffects = 1, mayLoad = 1 in
2111 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2112 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2114 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2115 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2116 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2119 // AVX 256-bit register conversion intrinsics
2120 let Predicates = [HasAVX] in {
2121 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2122 (VCVTDQ2PDYrr VR128:$src)>;
2123 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2124 (VCVTDQ2PDYrm addr:$src)>;
2125 } // Predicates = [HasAVX]
2127 // Convert packed double to packed single
2128 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2129 // register, but the same isn't true when using memory operands instead.
2130 // Provide other assembly rr and rm forms to address this explicitly.
2131 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2132 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2133 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2134 IIC_SSE_CVT_PD_RR>, VEX;
2137 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2138 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2139 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2140 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2142 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2143 IIC_SSE_CVT_PD_RM>, VEX;
2146 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2147 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2149 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2150 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
2151 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2152 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2154 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2155 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2156 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2157 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2159 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2160 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2161 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2163 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2164 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2166 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2170 // AVX 256-bit register conversion intrinsics
2171 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2172 // whenever possible to avoid declaring two versions of each one.
2173 let Predicates = [HasAVX] in {
2174 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2175 (VCVTDQ2PSYrr VR256:$src)>;
2176 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2177 (VCVTDQ2PSYrm addr:$src)>;
2179 // Match fround and fextend for 128/256-bit conversions
2180 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2181 (VCVTPD2PSrr VR128:$src)>;
2182 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2183 (VCVTPD2PSXrm addr:$src)>;
2184 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2185 (VCVTPD2PSYrr VR256:$src)>;
2186 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2187 (VCVTPD2PSYrm addr:$src)>;
2189 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2190 (VCVTPS2PDrr VR128:$src)>;
2191 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2192 (VCVTPS2PDYrr VR128:$src)>;
2193 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2194 (VCVTPS2PDYrm addr:$src)>;
2197 let Predicates = [UseSSE2] in {
2198 // Match fround and fextend for 128 conversions
2199 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2200 (CVTPD2PSrr VR128:$src)>;
2201 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2202 (CVTPD2PSrm addr:$src)>;
2204 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2205 (CVTPS2PDrr VR128:$src)>;
2208 //===----------------------------------------------------------------------===//
2209 // SSE 1 & 2 - Compare Instructions
2210 //===----------------------------------------------------------------------===//
2212 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2213 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2214 Operand CC, SDNode OpNode, ValueType VT,
2215 PatFrag ld_frag, string asm, string asm_alt,
2217 def rr : SIi8<0xC2, MRMSrcReg,
2218 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2219 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2221 def rm : SIi8<0xC2, MRMSrcMem,
2222 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2223 [(set RC:$dst, (OpNode (VT RC:$src1),
2224 (ld_frag addr:$src2), imm:$cc))],
2227 // Accept explicit immediate argument form instead of comparison code.
2228 let neverHasSideEffects = 1 in {
2229 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2230 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2231 IIC_SSE_ALU_F32S_RR>;
2233 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2234 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2235 IIC_SSE_ALU_F32S_RM>;
2239 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2240 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2241 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2243 XS, VEX_4V, VEX_LIG;
2244 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2245 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2246 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2247 SSE_ALU_F32S>, // same latency as 32 bit compare
2248 XD, VEX_4V, VEX_LIG;
2250 let Constraints = "$src1 = $dst" in {
2251 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2252 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2253 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2255 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2256 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2257 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2258 SSE_ALU_F32S>, // same latency as 32 bit compare
2262 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2263 Intrinsic Int, string asm, OpndItins itins> {
2264 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2265 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2266 [(set VR128:$dst, (Int VR128:$src1,
2267 VR128:$src, imm:$cc))],
2269 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2270 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2271 [(set VR128:$dst, (Int VR128:$src1,
2272 (load addr:$src), imm:$cc))],
2276 // Aliases to match intrinsics which expect XMM operand(s).
2277 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2278 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2281 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2282 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2283 SSE_ALU_F32S>, // same latency as f32
2285 let Constraints = "$src1 = $dst" in {
2286 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2287 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2289 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2290 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2291 SSE_ALU_F32S>, // same latency as f32
2296 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2297 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2298 ValueType vt, X86MemOperand x86memop,
2299 PatFrag ld_frag, string OpcodeStr, Domain d> {
2300 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2301 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2302 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2303 IIC_SSE_COMIS_RR, d>;
2304 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2305 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2306 [(set EFLAGS, (OpNode (vt RC:$src1),
2307 (ld_frag addr:$src2)))],
2308 IIC_SSE_COMIS_RM, d>;
2311 let Defs = [EFLAGS] in {
2312 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2313 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2314 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2315 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2317 let Pattern = []<dag> in {
2318 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2319 "comiss", SSEPackedSingle>, TB, VEX,
2321 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2322 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2326 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2327 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2328 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2329 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2331 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2332 load, "comiss", SSEPackedSingle>, TB, VEX;
2333 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2334 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2335 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2336 "ucomiss", SSEPackedSingle>, TB;
2337 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2338 "ucomisd", SSEPackedDouble>, TB, OpSize;
2340 let Pattern = []<dag> in {
2341 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2342 "comiss", SSEPackedSingle>, TB;
2343 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2344 "comisd", SSEPackedDouble>, TB, OpSize;
2347 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2348 load, "ucomiss", SSEPackedSingle>, TB;
2349 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2350 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2352 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2353 "comiss", SSEPackedSingle>, TB;
2354 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2355 "comisd", SSEPackedDouble>, TB, OpSize;
2356 } // Defs = [EFLAGS]
2358 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2359 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2360 Operand CC, Intrinsic Int, string asm,
2361 string asm_alt, Domain d> {
2362 def rri : PIi8<0xC2, MRMSrcReg,
2363 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2364 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2365 IIC_SSE_CMPP_RR, d>;
2366 def rmi : PIi8<0xC2, MRMSrcMem,
2367 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2368 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2369 IIC_SSE_CMPP_RM, d>;
2371 // Accept explicit immediate argument form instead of comparison code.
2372 let neverHasSideEffects = 1 in {
2373 def rri_alt : PIi8<0xC2, MRMSrcReg,
2374 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2375 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2376 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2377 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2378 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2382 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2383 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2384 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2385 SSEPackedSingle>, TB, VEX_4V;
2386 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2387 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2388 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2389 SSEPackedDouble>, TB, OpSize, VEX_4V;
2390 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2391 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2392 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2393 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2394 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2395 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2396 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2397 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2398 let Constraints = "$src1 = $dst" in {
2399 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2400 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2401 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2402 SSEPackedSingle>, TB;
2403 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2404 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2405 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2406 SSEPackedDouble>, TB, OpSize;
2409 let Predicates = [HasAVX] in {
2410 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2411 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2412 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2413 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2414 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2415 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2416 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2417 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2419 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2420 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2421 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2422 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2423 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2424 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2425 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2426 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2429 let Predicates = [UseSSE1] in {
2430 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2431 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2432 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2433 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2436 let Predicates = [UseSSE2] in {
2437 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2438 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2439 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2440 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2443 //===----------------------------------------------------------------------===//
2444 // SSE 1 & 2 - Shuffle Instructions
2445 //===----------------------------------------------------------------------===//
2447 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2448 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2449 ValueType vt, string asm, PatFrag mem_frag,
2450 Domain d, bit IsConvertibleToThreeAddress = 0> {
2451 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2452 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2453 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2454 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2455 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2456 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2457 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2458 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2459 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2462 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2463 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2464 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2465 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2466 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2467 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2468 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2469 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2470 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2471 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2472 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2473 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2475 let Constraints = "$src1 = $dst" in {
2476 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2477 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2478 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2480 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2481 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2482 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2486 let Predicates = [HasAVX] in {
2487 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2488 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2489 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2490 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2491 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2493 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2494 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2495 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2496 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2497 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2500 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2501 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2502 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2503 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2504 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2506 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2507 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2508 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2509 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2510 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2513 let Predicates = [UseSSE1] in {
2514 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2515 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2516 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2517 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2518 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2521 let Predicates = [UseSSE2] in {
2522 // Generic SHUFPD patterns
2523 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2524 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2525 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2526 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2527 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2530 //===----------------------------------------------------------------------===//
2531 // SSE 1 & 2 - Unpack Instructions
2532 //===----------------------------------------------------------------------===//
2534 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2535 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2536 PatFrag mem_frag, RegisterClass RC,
2537 X86MemOperand x86memop, string asm,
2539 def rr : PI<opc, MRMSrcReg,
2540 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2542 (vt (OpNode RC:$src1, RC:$src2)))],
2544 def rm : PI<opc, MRMSrcMem,
2545 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2547 (vt (OpNode RC:$src1,
2548 (mem_frag addr:$src2))))],
2552 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2553 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2554 SSEPackedSingle>, TB, VEX_4V;
2555 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2556 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2557 SSEPackedDouble>, TB, OpSize, VEX_4V;
2558 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2559 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2560 SSEPackedSingle>, TB, VEX_4V;
2561 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2562 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2563 SSEPackedDouble>, TB, OpSize, VEX_4V;
2565 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2566 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2567 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2568 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2569 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2570 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2571 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2572 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2573 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2574 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2575 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2576 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2578 let Constraints = "$src1 = $dst" in {
2579 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2580 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2581 SSEPackedSingle>, TB;
2582 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2583 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2584 SSEPackedDouble>, TB, OpSize;
2585 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2586 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2587 SSEPackedSingle>, TB;
2588 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2589 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2590 SSEPackedDouble>, TB, OpSize;
2591 } // Constraints = "$src1 = $dst"
2593 let Predicates = [HasAVX1Only] in {
2594 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2595 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2596 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2597 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2598 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2599 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2600 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2601 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2603 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2604 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2605 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2606 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2607 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2608 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2609 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2610 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2613 let Predicates = [HasAVX] in {
2614 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2615 // problem is during lowering, where it's not possible to recognize the load
2616 // fold cause it has two uses through a bitcast. One use disappears at isel
2617 // time and the fold opportunity reappears.
2618 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2619 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2622 let Predicates = [UseSSE2] in {
2623 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2624 // problem is during lowering, where it's not possible to recognize the load
2625 // fold cause it has two uses through a bitcast. One use disappears at isel
2626 // time and the fold opportunity reappears.
2627 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2628 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2631 //===----------------------------------------------------------------------===//
2632 // SSE 1 & 2 - Extract Floating-Point Sign mask
2633 //===----------------------------------------------------------------------===//
2635 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2636 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2638 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2639 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2640 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2641 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2642 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2643 IIC_SSE_MOVMSK, d>, REX_W;
2646 let Predicates = [HasAVX] in {
2647 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2648 "movmskps", SSEPackedSingle>, TB, VEX;
2649 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2650 "movmskpd", SSEPackedDouble>, TB,
2652 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2653 "movmskps", SSEPackedSingle>, TB,
2655 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2656 "movmskpd", SSEPackedDouble>, TB,
2659 def : Pat<(i32 (X86fgetsign FR32:$src)),
2660 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2661 def : Pat<(i64 (X86fgetsign FR32:$src)),
2662 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2663 def : Pat<(i32 (X86fgetsign FR64:$src)),
2664 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2665 def : Pat<(i64 (X86fgetsign FR64:$src)),
2666 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2669 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2670 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2671 SSEPackedSingle>, TB, VEX;
2672 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2673 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2674 SSEPackedDouble>, TB,
2676 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2677 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2678 SSEPackedSingle>, TB, VEX, VEX_L;
2679 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2680 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2681 SSEPackedDouble>, TB,
2685 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2686 SSEPackedSingle>, TB;
2687 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2688 SSEPackedDouble>, TB, OpSize;
2690 def : Pat<(i32 (X86fgetsign FR32:$src)),
2691 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2692 Requires<[UseSSE1]>;
2693 def : Pat<(i64 (X86fgetsign FR32:$src)),
2694 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2695 Requires<[UseSSE1]>;
2696 def : Pat<(i32 (X86fgetsign FR64:$src)),
2697 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2698 Requires<[UseSSE2]>;
2699 def : Pat<(i64 (X86fgetsign FR64:$src)),
2700 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2701 Requires<[UseSSE2]>;
2703 //===---------------------------------------------------------------------===//
2704 // SSE2 - Packed Integer Logical Instructions
2705 //===---------------------------------------------------------------------===//
2707 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2709 /// PDI_binop_rm - Simple SSE2 binary operator.
2710 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2711 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2712 X86MemOperand x86memop, OpndItins itins,
2713 bit IsCommutable, bit Is2Addr> {
2714 let isCommutable = IsCommutable in
2715 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2716 (ins RC:$src1, RC:$src2),
2718 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2719 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2720 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2721 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2722 (ins RC:$src1, x86memop:$src2),
2724 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2725 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2726 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2727 (bitconvert (memop_frag addr:$src2)))))],
2730 } // ExeDomain = SSEPackedInt
2732 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2733 ValueType OpVT128, ValueType OpVT256,
2734 OpndItins itins, bit IsCommutable = 0> {
2735 let Predicates = [HasAVX] in
2736 defm V#NAME : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2737 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2739 let Constraints = "$src1 = $dst" in
2740 defm NAME : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2741 memopv2i64, i128mem, itins, IsCommutable, 1>;
2743 let Predicates = [HasAVX2] in
2744 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2745 OpVT256, VR256, memopv4i64, i256mem, itins,
2746 IsCommutable, 0>, VEX_4V, VEX_L;
2749 // These are ordered here for pattern ordering requirements with the fp versions
2751 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2752 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2753 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2754 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2755 SSE_BIT_ITINS_P, 0>;
2757 //===----------------------------------------------------------------------===//
2758 // SSE 1 & 2 - Logical Instructions
2759 //===----------------------------------------------------------------------===//
2761 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2763 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2764 SDNode OpNode, OpndItins itins> {
2765 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2766 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2769 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2770 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2773 let Constraints = "$src1 = $dst" in {
2774 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2775 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2778 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2779 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2784 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2785 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2787 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2789 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2792 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2793 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2796 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2798 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2800 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2801 !strconcat(OpcodeStr, "ps"), f256mem,
2802 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2803 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2804 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2806 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2807 !strconcat(OpcodeStr, "pd"), f256mem,
2808 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2809 (bc_v4i64 (v4f64 VR256:$src2))))],
2810 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2811 (memopv4i64 addr:$src2)))], 0>,
2812 TB, OpSize, VEX_4V, VEX_L;
2814 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2815 // are all promoted to v2i64, and the patterns are covered by the int
2816 // version. This is needed in SSE only, because v2i64 isn't supported on
2817 // SSE1, but only on SSE2.
2818 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2819 !strconcat(OpcodeStr, "ps"), f128mem, [],
2820 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2821 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2823 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2824 !strconcat(OpcodeStr, "pd"), f128mem,
2825 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2826 (bc_v2i64 (v2f64 VR128:$src2))))],
2827 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2828 (memopv2i64 addr:$src2)))], 0>,
2831 let Constraints = "$src1 = $dst" in {
2832 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2833 !strconcat(OpcodeStr, "ps"), f128mem,
2834 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2835 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2836 (memopv2i64 addr:$src2)))]>, TB;
2838 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2839 !strconcat(OpcodeStr, "pd"), f128mem,
2840 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2841 (bc_v2i64 (v2f64 VR128:$src2))))],
2842 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2843 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2847 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2848 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2849 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2850 let isCommutable = 0 in
2851 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2853 //===----------------------------------------------------------------------===//
2854 // SSE 1 & 2 - Arithmetic Instructions
2855 //===----------------------------------------------------------------------===//
2857 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2860 /// In addition, we also have a special variant of the scalar form here to
2861 /// represent the associated intrinsic operation. This form is unlike the
2862 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2863 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2865 /// These three forms can each be reg+reg or reg+mem.
2868 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2870 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2873 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2874 OpNode, FR32, f32mem,
2875 itins.s, Is2Addr>, XS;
2876 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2877 OpNode, FR64, f64mem,
2878 itins.d, Is2Addr>, XD;
2881 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2882 SDNode OpNode, SizeItins itins> {
2883 let Predicates = [HasAVX] in {
2884 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2885 VR128, v4f32, f128mem, memopv4f32,
2886 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2887 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2888 VR128, v2f64, f128mem, memopv2f64,
2889 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2891 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2892 OpNode, VR256, v8f32, f256mem, memopv8f32,
2893 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2894 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2895 OpNode, VR256, v4f64, f256mem, memopv4f64,
2896 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2899 let Constraints = "$src1 = $dst" in {
2900 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2901 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2903 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2904 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2905 itins.d, 1>, TB, OpSize;
2909 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2912 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2913 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2914 itins.s, Is2Addr>, XS;
2915 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2916 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2917 itins.d, Is2Addr>, XD;
2920 // Binary Arithmetic instructions
2921 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>;
2922 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>;
2923 let isCommutable = 0 in {
2924 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>;
2925 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>;
2926 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>;
2927 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>;
2930 let isCodeGenOnly = 1 in {
2931 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
2932 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
2935 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2936 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2938 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2939 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2942 let isCommutable = 0 in {
2943 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2944 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2946 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2947 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2949 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2950 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2952 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2953 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2957 let Constraints = "$src1 = $dst" in {
2958 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2959 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2960 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2961 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2963 let isCommutable = 0 in {
2964 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2965 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2966 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2967 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2968 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2969 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
2970 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2971 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
2975 let isCodeGenOnly = 1 in {
2976 defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
2978 defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
2980 let Constraints = "$src1 = $dst" in {
2981 defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
2982 defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
2987 /// In addition, we also have a special variant of the scalar form here to
2988 /// represent the associated intrinsic operation. This form is unlike the
2989 /// plain scalar form, in that it takes an entire vector (instead of a
2990 /// scalar) and leaves the top elements undefined.
2992 /// And, we have a special variant form for a full-vector intrinsic form.
2994 def SSE_SQRTP : OpndItins<
2995 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2998 def SSE_SQRTS : OpndItins<
2999 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
3002 def SSE_RCPP : OpndItins<
3003 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
3006 def SSE_RCPS : OpndItins<
3007 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
3010 /// sse1_fp_unop_s - SSE1 unops in scalar form.
3011 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
3012 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
3013 let Predicates = [HasAVX], hasSideEffects = 0 in {
3014 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3015 (ins FR32:$src1, FR32:$src2),
3016 !strconcat("v", OpcodeStr,
3017 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3018 []>, VEX_4V, VEX_LIG;
3019 let mayLoad = 1 in {
3020 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3021 (ins FR32:$src1,f32mem:$src2),
3022 !strconcat("v", OpcodeStr,
3023 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3024 []>, VEX_4V, VEX_LIG;
3025 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3026 (ins VR128:$src1, ssmem:$src2),
3027 !strconcat("v", OpcodeStr,
3028 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3029 []>, VEX_4V, VEX_LIG;
3033 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3034 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3035 [(set FR32:$dst, (OpNode FR32:$src))]>;
3036 // For scalar unary operations, fold a load into the operation
3037 // only in OptForSize mode. It eliminates an instruction, but it also
3038 // eliminates a whole-register clobber (the load), so it introduces a
3039 // partial register update condition.
3040 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3041 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3042 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3043 Requires<[UseSSE1, OptForSize]>;
3044 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3045 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3046 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
3047 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
3048 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3049 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
3052 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3053 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3055 let Predicates = [HasAVX], hasSideEffects = 0 in {
3056 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
3057 (ins FR32:$src1, FR32:$src2),
3058 !strconcat("v", OpcodeStr,
3059 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3060 []>, VEX_4V, VEX_LIG;
3061 let mayLoad = 1 in {
3062 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
3063 (ins FR32:$src1,f32mem:$src2),
3064 !strconcat("v", OpcodeStr,
3065 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3066 []>, VEX_4V, VEX_LIG;
3067 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3068 (ins VR128:$src1, ssmem:$src2),
3069 !strconcat("v", OpcodeStr,
3070 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3071 []>, VEX_4V, VEX_LIG;
3075 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3076 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3077 [(set FR32:$dst, (OpNode FR32:$src))]>;
3078 // For scalar unary operations, fold a load into the operation
3079 // only in OptForSize mode. It eliminates an instruction, but it also
3080 // eliminates a whole-register clobber (the load), so it introduces a
3081 // partial register update condition.
3082 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3083 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3084 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3085 Requires<[UseSSE1, OptForSize]>;
3086 let Constraints = "$src1 = $dst" in {
3087 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3088 (ins VR128:$src1, VR128:$src2),
3089 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3091 let mayLoad = 1, hasSideEffects = 0 in
3092 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3093 (ins VR128:$src1, ssmem:$src2),
3094 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3099 /// sse1_fp_unop_p - SSE1 unops in packed form.
3100 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3102 let Predicates = [HasAVX] in {
3103 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3104 !strconcat("v", OpcodeStr,
3105 "ps\t{$src, $dst|$dst, $src}"),
3106 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3108 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3109 !strconcat("v", OpcodeStr,
3110 "ps\t{$src, $dst|$dst, $src}"),
3111 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3113 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3114 !strconcat("v", OpcodeStr,
3115 "ps\t{$src, $dst|$dst, $src}"),
3116 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3117 itins.rr>, VEX, VEX_L;
3118 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3119 !strconcat("v", OpcodeStr,
3120 "ps\t{$src, $dst|$dst, $src}"),
3121 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3122 itins.rm>, VEX, VEX_L;
3125 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3126 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3127 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3128 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3129 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3130 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3133 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3134 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3135 Intrinsic V4F32Int, Intrinsic V8F32Int,
3137 let Predicates = [HasAVX] in {
3138 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3139 !strconcat("v", OpcodeStr,
3140 "ps\t{$src, $dst|$dst, $src}"),
3141 [(set VR128:$dst, (V4F32Int VR128:$src))],
3143 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3144 !strconcat("v", OpcodeStr,
3145 "ps\t{$src, $dst|$dst, $src}"),
3146 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3148 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3149 !strconcat("v", OpcodeStr,
3150 "ps\t{$src, $dst|$dst, $src}"),
3151 [(set VR256:$dst, (V8F32Int VR256:$src))],
3152 itins.rr>, VEX, VEX_L;
3153 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3155 !strconcat("v", OpcodeStr,
3156 "ps\t{$src, $dst|$dst, $src}"),
3157 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3158 itins.rm>, VEX, VEX_L;
3161 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3162 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3163 [(set VR128:$dst, (V4F32Int VR128:$src))],
3165 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3166 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3167 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3171 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3172 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3173 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3174 let Predicates = [HasAVX], hasSideEffects = 0 in {
3175 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3176 (ins FR64:$src1, FR64:$src2),
3177 !strconcat("v", OpcodeStr,
3178 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3179 []>, VEX_4V, VEX_LIG;
3180 let mayLoad = 1 in {
3181 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3182 (ins FR64:$src1,f64mem:$src2),
3183 !strconcat("v", OpcodeStr,
3184 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3185 []>, VEX_4V, VEX_LIG;
3186 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3187 (ins VR128:$src1, sdmem:$src2),
3188 !strconcat("v", OpcodeStr,
3189 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3190 []>, VEX_4V, VEX_LIG;
3194 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3195 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3196 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3197 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3198 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3199 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3200 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3201 Requires<[UseSSE2, OptForSize]>;
3202 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3203 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3204 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3205 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3206 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3207 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3210 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3211 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3212 SDNode OpNode, OpndItins itins> {
3213 let Predicates = [HasAVX] in {
3214 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3215 !strconcat("v", OpcodeStr,
3216 "pd\t{$src, $dst|$dst, $src}"),
3217 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3219 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3220 !strconcat("v", OpcodeStr,
3221 "pd\t{$src, $dst|$dst, $src}"),
3222 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3224 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3225 !strconcat("v", OpcodeStr,
3226 "pd\t{$src, $dst|$dst, $src}"),
3227 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3228 itins.rr>, VEX, VEX_L;
3229 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3230 !strconcat("v", OpcodeStr,
3231 "pd\t{$src, $dst|$dst, $src}"),
3232 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3233 itins.rm>, VEX, VEX_L;
3236 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3237 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3238 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3239 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3240 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3241 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3245 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3247 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
3248 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3250 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
3252 // Reciprocal approximations. Note that these typically require refinement
3253 // in order to obtain suitable precision.
3254 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3255 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
3256 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3257 int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
3258 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3259 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3260 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3261 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3263 def : Pat<(f32 (fsqrt FR32:$src)),
3264 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3265 def : Pat<(f32 (fsqrt (load addr:$src))),
3266 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3267 Requires<[HasAVX, OptForSize]>;
3268 def : Pat<(f64 (fsqrt FR64:$src)),
3269 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3270 def : Pat<(f64 (fsqrt (load addr:$src))),
3271 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3272 Requires<[HasAVX, OptForSize]>;
3274 def : Pat<(f32 (X86frsqrt FR32:$src)),
3275 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3276 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3277 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3278 Requires<[HasAVX, OptForSize]>;
3280 def : Pat<(f32 (X86frcp FR32:$src)),
3281 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3282 def : Pat<(f32 (X86frcp (load addr:$src))),
3283 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3284 Requires<[HasAVX, OptForSize]>;
3286 let Predicates = [HasAVX] in {
3287 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3288 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3289 (COPY_TO_REGCLASS VR128:$src, FR32)),
3291 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3292 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3294 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3295 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3296 (COPY_TO_REGCLASS VR128:$src, FR64)),
3298 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3299 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3301 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3302 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3303 (COPY_TO_REGCLASS VR128:$src, FR32)),
3305 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3306 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3308 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3309 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3310 (COPY_TO_REGCLASS VR128:$src, FR32)),
3312 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3313 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3316 // Reciprocal approximations. Note that these typically require refinement
3317 // in order to obtain suitable precision.
3318 let Predicates = [UseSSE1] in {
3319 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3320 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3321 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3322 (RCPSSr_Int VR128:$src, VR128:$src)>;
3325 // There is no f64 version of the reciprocal approximation instructions.
3327 //===----------------------------------------------------------------------===//
3328 // SSE 1 & 2 - Non-temporal stores
3329 //===----------------------------------------------------------------------===//
3331 let AddedComplexity = 400 in { // Prefer non-temporal versions
3332 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3333 (ins f128mem:$dst, VR128:$src),
3334 "movntps\t{$src, $dst|$dst, $src}",
3335 [(alignednontemporalstore (v4f32 VR128:$src),
3337 IIC_SSE_MOVNT>, VEX;
3338 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3339 (ins f128mem:$dst, VR128:$src),
3340 "movntpd\t{$src, $dst|$dst, $src}",
3341 [(alignednontemporalstore (v2f64 VR128:$src),
3343 IIC_SSE_MOVNT>, VEX;
3345 let ExeDomain = SSEPackedInt in
3346 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3347 (ins f128mem:$dst, VR128:$src),
3348 "movntdq\t{$src, $dst|$dst, $src}",
3349 [(alignednontemporalstore (v2i64 VR128:$src),
3351 IIC_SSE_MOVNT>, VEX;
3353 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3354 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3356 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3357 (ins f256mem:$dst, VR256:$src),
3358 "movntps\t{$src, $dst|$dst, $src}",
3359 [(alignednontemporalstore (v8f32 VR256:$src),
3361 IIC_SSE_MOVNT>, VEX, VEX_L;
3362 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3363 (ins f256mem:$dst, VR256:$src),
3364 "movntpd\t{$src, $dst|$dst, $src}",
3365 [(alignednontemporalstore (v4f64 VR256:$src),
3367 IIC_SSE_MOVNT>, VEX, VEX_L;
3368 let ExeDomain = SSEPackedInt in
3369 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3370 (ins f256mem:$dst, VR256:$src),
3371 "movntdq\t{$src, $dst|$dst, $src}",
3372 [(alignednontemporalstore (v4i64 VR256:$src),
3374 IIC_SSE_MOVNT>, VEX, VEX_L;
3377 let AddedComplexity = 400 in { // Prefer non-temporal versions
3378 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3379 "movntps\t{$src, $dst|$dst, $src}",
3380 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3382 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3383 "movntpd\t{$src, $dst|$dst, $src}",
3384 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3387 let ExeDomain = SSEPackedInt in
3388 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3389 "movntdq\t{$src, $dst|$dst, $src}",
3390 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3393 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3394 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3396 // There is no AVX form for instructions below this point
3397 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3398 "movnti{l}\t{$src, $dst|$dst, $src}",
3399 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3401 TB, Requires<[HasSSE2]>;
3402 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3403 "movnti{q}\t{$src, $dst|$dst, $src}",
3404 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3406 TB, Requires<[HasSSE2]>;
3409 //===----------------------------------------------------------------------===//
3410 // SSE 1 & 2 - Prefetch and memory fence
3411 //===----------------------------------------------------------------------===//
3413 // Prefetch intrinsic.
3414 let Predicates = [HasSSE1] in {
3415 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3416 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3417 IIC_SSE_PREFETCH>, TB;
3418 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3419 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3420 IIC_SSE_PREFETCH>, TB;
3421 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3422 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3423 IIC_SSE_PREFETCH>, TB;
3424 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3425 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3426 IIC_SSE_PREFETCH>, TB;
3430 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3431 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3432 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3434 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3435 // was introduced with SSE2, it's backward compatible.
3436 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3438 // Load, store, and memory fence
3439 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3440 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3441 TB, Requires<[HasSSE1]>;
3442 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3443 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3444 TB, Requires<[HasSSE2]>;
3445 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3446 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3447 TB, Requires<[HasSSE2]>;
3449 def : Pat<(X86SFence), (SFENCE)>;
3450 def : Pat<(X86LFence), (LFENCE)>;
3451 def : Pat<(X86MFence), (MFENCE)>;
3453 //===----------------------------------------------------------------------===//
3454 // SSE 1 & 2 - Load/Store XCSR register
3455 //===----------------------------------------------------------------------===//
3457 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3458 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3459 IIC_SSE_LDMXCSR>, VEX;
3460 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3461 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3462 IIC_SSE_STMXCSR>, VEX;
3464 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3465 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3467 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3468 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3471 //===---------------------------------------------------------------------===//
3472 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3473 //===---------------------------------------------------------------------===//
3475 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3477 let neverHasSideEffects = 1 in {
3478 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3479 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3481 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3482 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3484 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3485 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3487 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3488 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3493 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3494 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3495 "movdqa\t{$src, $dst|$dst, $src}", [],
3498 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3499 "movdqa\t{$src, $dst|$dst, $src}", [],
3500 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3501 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3502 "movdqu\t{$src, $dst|$dst, $src}", [],
3505 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3506 "movdqu\t{$src, $dst|$dst, $src}", [],
3507 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3510 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3511 neverHasSideEffects = 1 in {
3512 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3513 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3515 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3516 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3518 let Predicates = [HasAVX] in {
3519 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3520 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3522 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3523 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3528 let mayStore = 1, neverHasSideEffects = 1 in {
3529 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3530 (ins i128mem:$dst, VR128:$src),
3531 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3533 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3534 (ins i256mem:$dst, VR256:$src),
3535 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3537 let Predicates = [HasAVX] in {
3538 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3539 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3541 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3542 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3547 let neverHasSideEffects = 1 in
3548 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3549 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3551 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3552 "movdqu\t{$src, $dst|$dst, $src}",
3553 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3556 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3557 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3558 "movdqa\t{$src, $dst|$dst, $src}", [],
3561 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3562 "movdqu\t{$src, $dst|$dst, $src}",
3563 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3566 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3567 neverHasSideEffects = 1 in {
3568 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3569 "movdqa\t{$src, $dst|$dst, $src}",
3570 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3572 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3573 "movdqu\t{$src, $dst|$dst, $src}",
3574 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3576 XS, Requires<[UseSSE2]>;
3579 let mayStore = 1 in {
3580 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3581 "movdqa\t{$src, $dst|$dst, $src}",
3582 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3584 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3585 "movdqu\t{$src, $dst|$dst, $src}",
3586 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3588 XS, Requires<[UseSSE2]>;
3591 } // ExeDomain = SSEPackedInt
3593 let Predicates = [HasAVX] in {
3594 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3595 (VMOVDQUmr addr:$dst, VR128:$src)>;
3596 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3597 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3599 let Predicates = [UseSSE2] in
3600 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3601 (MOVDQUmr addr:$dst, VR128:$src)>;
3603 //===---------------------------------------------------------------------===//
3604 // SSE2 - Packed Integer Arithmetic Instructions
3605 //===---------------------------------------------------------------------===//
3607 def SSE_PMADD : OpndItins<
3608 IIC_SSE_PMADD, IIC_SSE_PMADD
3611 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3613 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3614 RegisterClass RC, PatFrag memop_frag,
3615 X86MemOperand x86memop,
3617 bit IsCommutable = 0,
3619 let isCommutable = IsCommutable in
3620 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3621 (ins RC:$src1, RC:$src2),
3623 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3624 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3625 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3626 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3627 (ins RC:$src1, x86memop:$src2),
3629 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3630 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3631 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3635 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3636 Intrinsic IntId256, OpndItins itins,
3637 bit IsCommutable = 0> {
3638 let Predicates = [HasAVX] in
3639 defm V#NAME : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3640 VR128, memopv2i64, i128mem, itins,
3641 IsCommutable, 0>, VEX_4V;
3643 let Constraints = "$src1 = $dst" in
3644 defm NAME : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3645 i128mem, itins, IsCommutable, 1>;
3647 let Predicates = [HasAVX2] in
3648 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3649 VR256, memopv4i64, i256mem, itins,
3650 IsCommutable, 0>, VEX_4V, VEX_L;
3653 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3654 string OpcodeStr, SDNode OpNode,
3655 SDNode OpNode2, RegisterClass RC,
3656 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3657 ShiftOpndItins itins,
3659 // src2 is always 128-bit
3660 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3661 (ins RC:$src1, VR128:$src2),
3663 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3664 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3665 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3667 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3668 (ins RC:$src1, i128mem:$src2),
3670 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3671 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3672 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3673 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3674 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3675 (ins RC:$src1, i32i8imm:$src2),
3677 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3678 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3679 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3682 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3683 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3684 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3685 PatFrag memop_frag, X86MemOperand x86memop,
3687 bit IsCommutable = 0, bit Is2Addr = 1> {
3688 let isCommutable = IsCommutable in
3689 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3690 (ins RC:$src1, RC:$src2),
3692 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3693 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3694 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3695 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3696 (ins RC:$src1, x86memop:$src2),
3698 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3699 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3700 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3701 (bitconvert (memop_frag addr:$src2)))))]>;
3703 } // ExeDomain = SSEPackedInt
3705 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3706 SSE_INTALU_ITINS_P, 1>;
3707 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3708 SSE_INTALU_ITINS_P, 1>;
3709 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3710 SSE_INTALU_ITINS_P, 1>;
3711 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3712 SSE_INTALUQ_ITINS_P, 1>;
3713 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3714 SSE_INTMUL_ITINS_P, 1>;
3715 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3716 SSE_INTALU_ITINS_P, 0>;
3717 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3718 SSE_INTALU_ITINS_P, 0>;
3719 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3720 SSE_INTALU_ITINS_P, 0>;
3721 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3722 SSE_INTALUQ_ITINS_P, 0>;
3723 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3724 SSE_INTALU_ITINS_P, 0>;
3725 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3726 SSE_INTALU_ITINS_P, 0>;
3727 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3728 SSE_INTALU_ITINS_P, 1>;
3729 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3730 SSE_INTALU_ITINS_P, 1>;
3731 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3732 SSE_INTALU_ITINS_P, 1>;
3733 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3734 SSE_INTALU_ITINS_P, 1>;
3737 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3738 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3739 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3740 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3741 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3742 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3743 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3744 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3745 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3746 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3747 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3748 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3749 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3750 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3751 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3752 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3753 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3754 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3755 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3756 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3757 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3758 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3759 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3760 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3762 let Predicates = [HasAVX] in
3763 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3764 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3766 let Predicates = [HasAVX2] in
3767 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3768 VR256, memopv4i64, i256mem,
3769 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3770 let Constraints = "$src1 = $dst" in
3771 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3772 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3774 //===---------------------------------------------------------------------===//
3775 // SSE2 - Packed Integer Logical Instructions
3776 //===---------------------------------------------------------------------===//
3778 let Predicates = [HasAVX] in {
3779 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3780 VR128, v8i16, v8i16, bc_v8i16,
3781 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3782 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3783 VR128, v4i32, v4i32, bc_v4i32,
3784 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3785 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3786 VR128, v2i64, v2i64, bc_v2i64,
3787 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3789 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3790 VR128, v8i16, v8i16, bc_v8i16,
3791 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3792 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3793 VR128, v4i32, v4i32, bc_v4i32,
3794 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3795 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3796 VR128, v2i64, v2i64, bc_v2i64,
3797 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3799 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3800 VR128, v8i16, v8i16, bc_v8i16,
3801 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3802 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3803 VR128, v4i32, v4i32, bc_v4i32,
3804 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3806 let ExeDomain = SSEPackedInt in {
3807 // 128-bit logical shifts.
3808 def VPSLLDQri : PDIi8<0x73, MRM7r,
3809 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3810 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3812 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3814 def VPSRLDQri : PDIi8<0x73, MRM3r,
3815 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3816 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3818 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3820 // PSRADQri doesn't exist in SSE[1-3].
3822 } // Predicates = [HasAVX]
3824 let Predicates = [HasAVX2] in {
3825 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3826 VR256, v16i16, v8i16, bc_v8i16,
3827 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3828 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3829 VR256, v8i32, v4i32, bc_v4i32,
3830 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3831 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3832 VR256, v4i64, v2i64, bc_v2i64,
3833 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3835 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3836 VR256, v16i16, v8i16, bc_v8i16,
3837 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3838 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3839 VR256, v8i32, v4i32, bc_v4i32,
3840 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3841 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3842 VR256, v4i64, v2i64, bc_v2i64,
3843 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3845 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3846 VR256, v16i16, v8i16, bc_v8i16,
3847 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3848 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3849 VR256, v8i32, v4i32, bc_v4i32,
3850 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3852 let ExeDomain = SSEPackedInt in {
3853 // 256-bit logical shifts.
3854 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3855 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3856 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3858 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3860 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3861 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3862 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3864 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3866 // PSRADQYri doesn't exist in SSE[1-3].
3868 } // Predicates = [HasAVX2]
3870 let Constraints = "$src1 = $dst" in {
3871 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3872 VR128, v8i16, v8i16, bc_v8i16,
3873 SSE_INTSHIFT_ITINS_P>;
3874 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3875 VR128, v4i32, v4i32, bc_v4i32,
3876 SSE_INTSHIFT_ITINS_P>;
3877 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3878 VR128, v2i64, v2i64, bc_v2i64,
3879 SSE_INTSHIFT_ITINS_P>;
3881 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3882 VR128, v8i16, v8i16, bc_v8i16,
3883 SSE_INTSHIFT_ITINS_P>;
3884 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3885 VR128, v4i32, v4i32, bc_v4i32,
3886 SSE_INTSHIFT_ITINS_P>;
3887 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3888 VR128, v2i64, v2i64, bc_v2i64,
3889 SSE_INTSHIFT_ITINS_P>;
3891 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3892 VR128, v8i16, v8i16, bc_v8i16,
3893 SSE_INTSHIFT_ITINS_P>;
3894 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3895 VR128, v4i32, v4i32, bc_v4i32,
3896 SSE_INTSHIFT_ITINS_P>;
3898 let ExeDomain = SSEPackedInt in {
3899 // 128-bit logical shifts.
3900 def PSLLDQri : PDIi8<0x73, MRM7r,
3901 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3902 "pslldq\t{$src2, $dst|$dst, $src2}",
3904 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3905 def PSRLDQri : PDIi8<0x73, MRM3r,
3906 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3907 "psrldq\t{$src2, $dst|$dst, $src2}",
3909 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3910 // PSRADQri doesn't exist in SSE[1-3].
3912 } // Constraints = "$src1 = $dst"
3914 let Predicates = [HasAVX] in {
3915 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3916 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3917 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3918 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3919 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3920 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3922 // Shift up / down and insert zero's.
3923 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3924 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3925 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3926 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3929 let Predicates = [HasAVX2] in {
3930 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3931 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3932 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3933 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3936 let Predicates = [UseSSE2] in {
3937 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3938 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3939 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3940 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3941 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3942 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3944 // Shift up / down and insert zero's.
3945 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3946 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3947 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3948 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3951 //===---------------------------------------------------------------------===//
3952 // SSE2 - Packed Integer Comparison Instructions
3953 //===---------------------------------------------------------------------===//
3955 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
3956 SSE_INTALU_ITINS_P, 1>;
3957 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
3958 SSE_INTALU_ITINS_P, 1>;
3959 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
3960 SSE_INTALU_ITINS_P, 1>;
3961 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
3962 SSE_INTALU_ITINS_P, 0>;
3963 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
3964 SSE_INTALU_ITINS_P, 0>;
3965 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
3966 SSE_INTALU_ITINS_P, 0>;
3968 //===---------------------------------------------------------------------===//
3969 // SSE2 - Packed Integer Pack Instructions
3970 //===---------------------------------------------------------------------===//
3972 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3973 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
3974 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3975 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
3976 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3977 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
3979 //===---------------------------------------------------------------------===//
3980 // SSE2 - Packed Integer Shuffle Instructions
3981 //===---------------------------------------------------------------------===//
3983 let ExeDomain = SSEPackedInt in {
3984 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt128, ValueType vt256,
3986 let Predicates = [HasAVX] in {
3987 def V#NAME#ri : Ii8<0x70, MRMSrcReg, (outs VR128:$dst),
3988 (ins VR128:$src1, i8imm:$src2),
3989 !strconcat("v", OpcodeStr,
3990 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3992 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
3993 IIC_SSE_PSHUF>, VEX;
3994 def V#NAME#mi : Ii8<0x70, MRMSrcMem, (outs VR128:$dst),
3995 (ins i128mem:$src1, i8imm:$src2),
3996 !strconcat("v", OpcodeStr,
3997 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3999 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4000 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX;
4003 let Predicates = [HasAVX2] in {
4004 def V#NAME#Yri : Ii8<0x70, MRMSrcReg, (outs VR256:$dst),
4005 (ins VR256:$src1, i8imm:$src2),
4006 !strconcat("v", OpcodeStr,
4007 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4009 (vt256 (OpNode VR256:$src1, (i8 imm:$src2))))],
4010 IIC_SSE_PSHUF>, VEX, VEX_L;
4011 def V#NAME#Ymi : Ii8<0x70, MRMSrcMem, (outs VR256:$dst),
4012 (ins i256mem:$src1, i8imm:$src2),
4013 !strconcat("v", OpcodeStr,
4014 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4016 (vt256 (OpNode (bitconvert (memopv4i64 addr:$src1)),
4017 (i8 imm:$src2))))], IIC_SSE_PSHUF>, VEX, VEX_L;
4020 let Predicates = [UseSSE2] in {
4021 def ri : Ii8<0x70, MRMSrcReg,
4022 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4023 !strconcat(OpcodeStr,
4024 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4026 (vt128 (OpNode VR128:$src1, (i8 imm:$src2))))],
4028 def mi : Ii8<0x70, MRMSrcMem,
4029 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4030 !strconcat(OpcodeStr,
4031 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4033 (vt128 (OpNode (bitconvert (memopv2i64 addr:$src1)),
4034 (i8 imm:$src2))))], IIC_SSE_PSHUF>;
4037 } // ExeDomain = SSEPackedInt
4039 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, v8i32, X86PShufd>, TB, OpSize;
4040 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, v16i16, X86PShufhw>, XS;
4041 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, v16i16, X86PShuflw>, XD;
4043 let Predicates = [HasAVX] in {
4044 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4045 (VPSHUFDmi addr:$src1, imm:$imm)>;
4046 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4047 (VPSHUFDri VR128:$src1, imm:$imm)>;
4050 let Predicates = [UseSSE2] in {
4051 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
4052 (PSHUFDmi addr:$src1, imm:$imm)>;
4053 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4054 (PSHUFDri VR128:$src1, imm:$imm)>;
4057 //===---------------------------------------------------------------------===//
4058 // SSE2 - Packed Integer Unpack Instructions
4059 //===---------------------------------------------------------------------===//
4061 let ExeDomain = SSEPackedInt in {
4062 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4063 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4064 def rr : PDI<opc, MRMSrcReg,
4065 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4067 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4068 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4069 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4071 def rm : PDI<opc, MRMSrcMem,
4072 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4074 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4075 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4076 [(set VR128:$dst, (OpNode VR128:$src1,
4077 (bc_frag (memopv2i64
4082 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4083 SDNode OpNode, PatFrag bc_frag> {
4084 def Yrr : PDI<opc, MRMSrcReg,
4085 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4086 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4087 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4088 def Yrm : PDI<opc, MRMSrcMem,
4089 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4090 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4091 [(set VR256:$dst, (OpNode VR256:$src1,
4092 (bc_frag (memopv4i64 addr:$src2))))]>;
4095 let Predicates = [HasAVX] in {
4096 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4097 bc_v16i8, 0>, VEX_4V;
4098 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4099 bc_v8i16, 0>, VEX_4V;
4100 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4101 bc_v4i32, 0>, VEX_4V;
4102 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4103 bc_v2i64, 0>, VEX_4V;
4105 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4106 bc_v16i8, 0>, VEX_4V;
4107 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4108 bc_v8i16, 0>, VEX_4V;
4109 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4110 bc_v4i32, 0>, VEX_4V;
4111 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4112 bc_v2i64, 0>, VEX_4V;
4115 let Predicates = [HasAVX2] in {
4116 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4117 bc_v32i8>, VEX_4V, VEX_L;
4118 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4119 bc_v16i16>, VEX_4V, VEX_L;
4120 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4121 bc_v8i32>, VEX_4V, VEX_L;
4122 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4123 bc_v4i64>, VEX_4V, VEX_L;
4125 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4126 bc_v32i8>, VEX_4V, VEX_L;
4127 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4128 bc_v16i16>, VEX_4V, VEX_L;
4129 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4130 bc_v8i32>, VEX_4V, VEX_L;
4131 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4132 bc_v4i64>, VEX_4V, VEX_L;
4135 let Constraints = "$src1 = $dst" in {
4136 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4138 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4140 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4142 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4145 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4147 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4149 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4151 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4154 } // ExeDomain = SSEPackedInt
4156 //===---------------------------------------------------------------------===//
4157 // SSE2 - Packed Integer Extract and Insert
4158 //===---------------------------------------------------------------------===//
4160 let ExeDomain = SSEPackedInt in {
4161 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4162 def rri : Ii8<0xC4, MRMSrcReg,
4163 (outs VR128:$dst), (ins VR128:$src1,
4164 GR32:$src2, i32i8imm:$src3),
4166 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4167 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4169 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4170 def rmi : Ii8<0xC4, MRMSrcMem,
4171 (outs VR128:$dst), (ins VR128:$src1,
4172 i16mem:$src2, i32i8imm:$src3),
4174 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4175 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4177 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4178 imm:$src3))], IIC_SSE_PINSRW>;
4182 let Predicates = [HasAVX] in
4183 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4184 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4185 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4186 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4187 imm:$src2))]>, TB, OpSize, VEX;
4188 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4189 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4190 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4191 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4192 imm:$src2))], IIC_SSE_PEXTRW>;
4195 let Predicates = [HasAVX] in {
4196 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4197 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4198 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4199 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4200 []>, TB, OpSize, VEX_4V;
4203 let Constraints = "$src1 = $dst" in
4204 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4206 } // ExeDomain = SSEPackedInt
4208 //===---------------------------------------------------------------------===//
4209 // SSE2 - Packed Mask Creation
4210 //===---------------------------------------------------------------------===//
4212 let ExeDomain = SSEPackedInt in {
4214 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4215 "pmovmskb\t{$src, $dst|$dst, $src}",
4216 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4217 IIC_SSE_MOVMSK>, VEX;
4218 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4219 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4221 let Predicates = [HasAVX2] in {
4222 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4223 "pmovmskb\t{$src, $dst|$dst, $src}",
4224 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4225 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4226 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4229 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4230 "pmovmskb\t{$src, $dst|$dst, $src}",
4231 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4234 } // ExeDomain = SSEPackedInt
4236 //===---------------------------------------------------------------------===//
4237 // SSE2 - Conditional Store
4238 //===---------------------------------------------------------------------===//
4240 let ExeDomain = SSEPackedInt in {
4243 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4244 (ins VR128:$src, VR128:$mask),
4245 "maskmovdqu\t{$mask, $src|$src, $mask}",
4246 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4247 IIC_SSE_MASKMOV>, VEX;
4249 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4250 (ins VR128:$src, VR128:$mask),
4251 "maskmovdqu\t{$mask, $src|$src, $mask}",
4252 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4253 IIC_SSE_MASKMOV>, VEX;
4256 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4257 "maskmovdqu\t{$mask, $src|$src, $mask}",
4258 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4261 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4262 "maskmovdqu\t{$mask, $src|$src, $mask}",
4263 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4266 } // ExeDomain = SSEPackedInt
4268 //===---------------------------------------------------------------------===//
4269 // SSE2 - Move Doubleword
4270 //===---------------------------------------------------------------------===//
4272 //===---------------------------------------------------------------------===//
4273 // Move Int Doubleword to Packed Double Int
4275 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4276 "movd\t{$src, $dst|$dst, $src}",
4278 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4280 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4281 "movd\t{$src, $dst|$dst, $src}",
4283 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4286 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4287 "mov{d|q}\t{$src, $dst|$dst, $src}",
4289 (v2i64 (scalar_to_vector GR64:$src)))],
4290 IIC_SSE_MOVDQ>, VEX;
4291 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4292 "mov{d|q}\t{$src, $dst|$dst, $src}",
4293 [(set FR64:$dst, (bitconvert GR64:$src))],
4294 IIC_SSE_MOVDQ>, VEX;
4296 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4297 "movd\t{$src, $dst|$dst, $src}",
4299 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4300 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4301 "movd\t{$src, $dst|$dst, $src}",
4303 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4305 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4306 "mov{d|q}\t{$src, $dst|$dst, $src}",
4308 (v2i64 (scalar_to_vector GR64:$src)))],
4310 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4311 "mov{d|q}\t{$src, $dst|$dst, $src}",
4312 [(set FR64:$dst, (bitconvert GR64:$src))],
4315 //===---------------------------------------------------------------------===//
4316 // Move Int Doubleword to Single Scalar
4318 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4319 "movd\t{$src, $dst|$dst, $src}",
4320 [(set FR32:$dst, (bitconvert GR32:$src))],
4321 IIC_SSE_MOVDQ>, VEX;
4323 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4324 "movd\t{$src, $dst|$dst, $src}",
4325 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4328 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4329 "movd\t{$src, $dst|$dst, $src}",
4330 [(set FR32:$dst, (bitconvert GR32:$src))],
4333 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4334 "movd\t{$src, $dst|$dst, $src}",
4335 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4338 //===---------------------------------------------------------------------===//
4339 // Move Packed Doubleword Int to Packed Double Int
4341 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4342 "movd\t{$src, $dst|$dst, $src}",
4343 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4344 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4345 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4346 (ins i32mem:$dst, VR128:$src),
4347 "movd\t{$src, $dst|$dst, $src}",
4348 [(store (i32 (vector_extract (v4i32 VR128:$src),
4349 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4351 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4352 "movd\t{$src, $dst|$dst, $src}",
4353 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4354 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4355 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4356 "movd\t{$src, $dst|$dst, $src}",
4357 [(store (i32 (vector_extract (v4i32 VR128:$src),
4358 (iPTR 0))), addr:$dst)],
4361 //===---------------------------------------------------------------------===//
4362 // Move Packed Doubleword Int first element to Doubleword Int
4364 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4365 "vmov{d|q}\t{$src, $dst|$dst, $src}",
4366 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4369 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4371 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4372 "mov{d|q}\t{$src, $dst|$dst, $src}",
4373 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4377 //===---------------------------------------------------------------------===//
4378 // Bitcast FR64 <-> GR64
4380 let Predicates = [HasAVX] in
4381 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4382 "vmovq\t{$src, $dst|$dst, $src}",
4383 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4385 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4386 "mov{d|q}\t{$src, $dst|$dst, $src}",
4387 [(set GR64:$dst, (bitconvert FR64:$src))],
4388 IIC_SSE_MOVDQ>, VEX;
4389 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4390 "movq\t{$src, $dst|$dst, $src}",
4391 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4392 IIC_SSE_MOVDQ>, VEX;
4394 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4395 "movq\t{$src, $dst|$dst, $src}",
4396 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4398 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4399 "mov{d|q}\t{$src, $dst|$dst, $src}",
4400 [(set GR64:$dst, (bitconvert FR64:$src))],
4402 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4403 "movq\t{$src, $dst|$dst, $src}",
4404 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4407 //===---------------------------------------------------------------------===//
4408 // Move Scalar Single to Double Int
4410 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4411 "movd\t{$src, $dst|$dst, $src}",
4412 [(set GR32:$dst, (bitconvert FR32:$src))],
4413 IIC_SSE_MOVD_ToGP>, VEX;
4414 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4415 "movd\t{$src, $dst|$dst, $src}",
4416 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4417 IIC_SSE_MOVDQ>, VEX;
4418 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4419 "movd\t{$src, $dst|$dst, $src}",
4420 [(set GR32:$dst, (bitconvert FR32:$src))],
4422 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4423 "movd\t{$src, $dst|$dst, $src}",
4424 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4427 //===---------------------------------------------------------------------===//
4428 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4430 let AddedComplexity = 15 in {
4431 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4432 "movd\t{$src, $dst|$dst, $src}",
4433 [(set VR128:$dst, (v4i32 (X86vzmovl
4434 (v4i32 (scalar_to_vector GR32:$src)))))],
4435 IIC_SSE_MOVDQ>, VEX;
4436 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4437 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4438 [(set VR128:$dst, (v2i64 (X86vzmovl
4439 (v2i64 (scalar_to_vector GR64:$src)))))],
4443 let AddedComplexity = 15 in {
4444 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4445 "movd\t{$src, $dst|$dst, $src}",
4446 [(set VR128:$dst, (v4i32 (X86vzmovl
4447 (v4i32 (scalar_to_vector GR32:$src)))))],
4449 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4450 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4451 [(set VR128:$dst, (v2i64 (X86vzmovl
4452 (v2i64 (scalar_to_vector GR64:$src)))))],
4456 let AddedComplexity = 20 in {
4457 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4460 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4461 (loadi32 addr:$src))))))],
4462 IIC_SSE_MOVDQ>, VEX;
4463 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4464 "movd\t{$src, $dst|$dst, $src}",
4466 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4467 (loadi32 addr:$src))))))],
4471 let Predicates = [HasAVX] in {
4472 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4473 let AddedComplexity = 20 in {
4474 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4475 (VMOVZDI2PDIrm addr:$src)>;
4476 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4477 (VMOVZDI2PDIrm addr:$src)>;
4479 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4480 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4481 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4482 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4483 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4484 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4485 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4488 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4489 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4490 (MOVZDI2PDIrm addr:$src)>;
4491 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4492 (MOVZDI2PDIrm addr:$src)>;
4495 // These are the correct encodings of the instructions so that we know how to
4496 // read correct assembly, even though we continue to emit the wrong ones for
4497 // compatibility with Darwin's buggy assembler.
4498 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4499 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4500 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4501 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4502 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4503 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4504 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4505 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4506 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4507 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4508 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4509 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4511 //===---------------------------------------------------------------------===//
4512 // SSE2 - Move Quadword
4513 //===---------------------------------------------------------------------===//
4515 //===---------------------------------------------------------------------===//
4516 // Move Quadword Int to Packed Quadword Int
4518 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4519 "vmovq\t{$src, $dst|$dst, $src}",
4521 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4522 VEX, Requires<[HasAVX]>;
4523 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4524 "movq\t{$src, $dst|$dst, $src}",
4526 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4528 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4530 //===---------------------------------------------------------------------===//
4531 // Move Packed Quadword Int to Quadword Int
4533 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4534 "movq\t{$src, $dst|$dst, $src}",
4535 [(store (i64 (vector_extract (v2i64 VR128:$src),
4536 (iPTR 0))), addr:$dst)],
4537 IIC_SSE_MOVDQ>, VEX;
4538 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4539 "movq\t{$src, $dst|$dst, $src}",
4540 [(store (i64 (vector_extract (v2i64 VR128:$src),
4541 (iPTR 0))), addr:$dst)],
4544 //===---------------------------------------------------------------------===//
4545 // Store / copy lower 64-bits of a XMM register.
4547 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4548 "movq\t{$src, $dst|$dst, $src}",
4549 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4550 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4551 "movq\t{$src, $dst|$dst, $src}",
4552 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4555 let AddedComplexity = 20 in
4556 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4557 "vmovq\t{$src, $dst|$dst, $src}",
4559 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4560 (loadi64 addr:$src))))))],
4562 XS, VEX, Requires<[HasAVX]>;
4564 let AddedComplexity = 20 in
4565 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4566 "movq\t{$src, $dst|$dst, $src}",
4568 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4569 (loadi64 addr:$src))))))],
4571 XS, Requires<[UseSSE2]>;
4573 let Predicates = [HasAVX], AddedComplexity = 20 in {
4574 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4575 (VMOVZQI2PQIrm addr:$src)>;
4576 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4577 (VMOVZQI2PQIrm addr:$src)>;
4578 def : Pat<(v2i64 (X86vzload addr:$src)),
4579 (VMOVZQI2PQIrm addr:$src)>;
4582 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4583 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4584 (MOVZQI2PQIrm addr:$src)>;
4585 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4586 (MOVZQI2PQIrm addr:$src)>;
4587 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4590 let Predicates = [HasAVX] in {
4591 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4592 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4593 def : Pat<(v4i64 (X86vzload addr:$src)),
4594 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4597 //===---------------------------------------------------------------------===//
4598 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4599 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4601 let AddedComplexity = 15 in
4602 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4603 "vmovq\t{$src, $dst|$dst, $src}",
4604 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4606 XS, VEX, Requires<[HasAVX]>;
4607 let AddedComplexity = 15 in
4608 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4609 "movq\t{$src, $dst|$dst, $src}",
4610 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4612 XS, Requires<[UseSSE2]>;
4614 let AddedComplexity = 20 in
4615 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4616 "vmovq\t{$src, $dst|$dst, $src}",
4617 [(set VR128:$dst, (v2i64 (X86vzmovl
4618 (loadv2i64 addr:$src))))],
4620 XS, VEX, Requires<[HasAVX]>;
4621 let AddedComplexity = 20 in {
4622 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4623 "movq\t{$src, $dst|$dst, $src}",
4624 [(set VR128:$dst, (v2i64 (X86vzmovl
4625 (loadv2i64 addr:$src))))],
4627 XS, Requires<[UseSSE2]>;
4630 let AddedComplexity = 20 in {
4631 let Predicates = [HasAVX] in {
4632 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4633 (VMOVZPQILo2PQIrm addr:$src)>;
4634 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4635 (VMOVZPQILo2PQIrr VR128:$src)>;
4637 let Predicates = [UseSSE2] in {
4638 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4639 (MOVZPQILo2PQIrm addr:$src)>;
4640 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4641 (MOVZPQILo2PQIrr VR128:$src)>;
4645 // Instructions to match in the assembler
4646 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4647 "movq\t{$src, $dst|$dst, $src}", [],
4648 IIC_SSE_MOVDQ>, VEX, VEX_W;
4649 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4650 "movq\t{$src, $dst|$dst, $src}", [],
4651 IIC_SSE_MOVDQ>, VEX, VEX_W;
4652 // Recognize "movd" with GR64 destination, but encode as a "movq"
4653 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4654 "movd\t{$src, $dst|$dst, $src}", [],
4655 IIC_SSE_MOVDQ>, VEX, VEX_W;
4657 // Instructions for the disassembler
4658 // xr = XMM register
4661 let Predicates = [HasAVX] in
4662 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4663 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4664 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4665 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4667 //===---------------------------------------------------------------------===//
4668 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4669 //===---------------------------------------------------------------------===//
4670 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4671 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4672 X86MemOperand x86memop> {
4673 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4674 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4675 [(set RC:$dst, (vt (OpNode RC:$src)))],
4677 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4678 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4679 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4683 let Predicates = [HasAVX] in {
4684 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4685 v4f32, VR128, memopv4f32, f128mem>, VEX;
4686 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4687 v4f32, VR128, memopv4f32, f128mem>, VEX;
4688 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4689 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4690 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4691 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4693 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4694 memopv4f32, f128mem>;
4695 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4696 memopv4f32, f128mem>;
4698 let Predicates = [HasAVX] in {
4699 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4700 (VMOVSHDUPrr VR128:$src)>;
4701 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4702 (VMOVSHDUPrm addr:$src)>;
4703 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4704 (VMOVSLDUPrr VR128:$src)>;
4705 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4706 (VMOVSLDUPrm addr:$src)>;
4707 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4708 (VMOVSHDUPYrr VR256:$src)>;
4709 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4710 (VMOVSHDUPYrm addr:$src)>;
4711 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4712 (VMOVSLDUPYrr VR256:$src)>;
4713 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4714 (VMOVSLDUPYrm addr:$src)>;
4717 let Predicates = [UseSSE3] in {
4718 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4719 (MOVSHDUPrr VR128:$src)>;
4720 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4721 (MOVSHDUPrm addr:$src)>;
4722 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4723 (MOVSLDUPrr VR128:$src)>;
4724 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4725 (MOVSLDUPrm addr:$src)>;
4728 //===---------------------------------------------------------------------===//
4729 // SSE3 - Replicate Double FP - MOVDDUP
4730 //===---------------------------------------------------------------------===//
4732 multiclass sse3_replicate_dfp<string OpcodeStr> {
4733 let neverHasSideEffects = 1 in
4734 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4735 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4736 [], IIC_SSE_MOV_LH>;
4737 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4738 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4741 (scalar_to_vector (loadf64 addr:$src)))))],
4745 // FIXME: Merge with above classe when there're patterns for the ymm version
4746 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4747 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4748 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4749 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4750 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4751 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4754 (scalar_to_vector (loadf64 addr:$src)))))]>;
4757 let Predicates = [HasAVX] in {
4758 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4759 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4762 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4764 let Predicates = [HasAVX] in {
4765 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4766 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4767 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4768 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4769 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4770 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4771 def : Pat<(X86Movddup (bc_v2f64
4772 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4773 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4776 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4777 (VMOVDDUPYrm addr:$src)>;
4778 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4779 (VMOVDDUPYrm addr:$src)>;
4780 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4781 (VMOVDDUPYrm addr:$src)>;
4782 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4783 (VMOVDDUPYrr VR256:$src)>;
4786 let Predicates = [UseSSE3] in {
4787 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4788 (MOVDDUPrm addr:$src)>;
4789 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4790 (MOVDDUPrm addr:$src)>;
4791 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4792 (MOVDDUPrm addr:$src)>;
4793 def : Pat<(X86Movddup (bc_v2f64
4794 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4795 (MOVDDUPrm addr:$src)>;
4798 //===---------------------------------------------------------------------===//
4799 // SSE3 - Move Unaligned Integer
4800 //===---------------------------------------------------------------------===//
4802 let Predicates = [HasAVX] in {
4803 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4804 "vlddqu\t{$src, $dst|$dst, $src}",
4805 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4806 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4807 "vlddqu\t{$src, $dst|$dst, $src}",
4808 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4811 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4812 "lddqu\t{$src, $dst|$dst, $src}",
4813 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4816 //===---------------------------------------------------------------------===//
4817 // SSE3 - Arithmetic
4818 //===---------------------------------------------------------------------===//
4820 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4821 X86MemOperand x86memop, OpndItins itins,
4823 def rr : I<0xD0, MRMSrcReg,
4824 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4826 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4827 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4828 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4829 def rm : I<0xD0, MRMSrcMem,
4830 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4832 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4833 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4834 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4837 let Predicates = [HasAVX] in {
4838 let ExeDomain = SSEPackedSingle in {
4839 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4840 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4841 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4842 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4844 let ExeDomain = SSEPackedDouble in {
4845 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4846 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4847 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4848 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4851 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4852 let ExeDomain = SSEPackedSingle in
4853 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4854 f128mem, SSE_ALU_F32P>, TB, XD;
4855 let ExeDomain = SSEPackedDouble in
4856 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4857 f128mem, SSE_ALU_F64P>, TB, OpSize;
4860 //===---------------------------------------------------------------------===//
4861 // SSE3 Instructions
4862 //===---------------------------------------------------------------------===//
4865 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4866 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4867 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4869 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4871 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4873 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4875 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4876 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4877 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4878 IIC_SSE_HADDSUB_RM>;
4880 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4881 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4882 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4884 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4885 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4886 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4888 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4890 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4891 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4892 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4893 IIC_SSE_HADDSUB_RM>;
4896 let Predicates = [HasAVX] in {
4897 let ExeDomain = SSEPackedSingle in {
4898 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4899 X86fhadd, 0>, VEX_4V;
4900 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4901 X86fhsub, 0>, VEX_4V;
4902 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4903 X86fhadd, 0>, VEX_4V, VEX_L;
4904 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4905 X86fhsub, 0>, VEX_4V, VEX_L;
4907 let ExeDomain = SSEPackedDouble in {
4908 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4909 X86fhadd, 0>, VEX_4V;
4910 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4911 X86fhsub, 0>, VEX_4V;
4912 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4913 X86fhadd, 0>, VEX_4V, VEX_L;
4914 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4915 X86fhsub, 0>, VEX_4V, VEX_L;
4919 let Constraints = "$src1 = $dst" in {
4920 let ExeDomain = SSEPackedSingle in {
4921 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4922 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4924 let ExeDomain = SSEPackedDouble in {
4925 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4926 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4930 //===---------------------------------------------------------------------===//
4931 // SSSE3 - Packed Absolute Instructions
4932 //===---------------------------------------------------------------------===//
4935 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4936 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4937 Intrinsic IntId128> {
4938 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4940 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4941 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
4944 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4946 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4949 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
4953 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4954 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4955 Intrinsic IntId256> {
4956 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4958 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4959 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4962 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4964 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4967 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4970 let Predicates = [HasAVX] in {
4971 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4972 int_x86_ssse3_pabs_b_128>, VEX;
4973 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4974 int_x86_ssse3_pabs_w_128>, VEX;
4975 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4976 int_x86_ssse3_pabs_d_128>, VEX;
4979 let Predicates = [HasAVX2] in {
4980 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4981 int_x86_avx2_pabs_b>, VEX, VEX_L;
4982 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4983 int_x86_avx2_pabs_w>, VEX, VEX_L;
4984 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4985 int_x86_avx2_pabs_d>, VEX, VEX_L;
4988 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4989 int_x86_ssse3_pabs_b_128>;
4990 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4991 int_x86_ssse3_pabs_w_128>;
4992 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4993 int_x86_ssse3_pabs_d_128>;
4995 //===---------------------------------------------------------------------===//
4996 // SSSE3 - Packed Binary Operator Instructions
4997 //===---------------------------------------------------------------------===//
4999 def SSE_PHADDSUBD : OpndItins<
5000 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
5002 def SSE_PHADDSUBSW : OpndItins<
5003 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
5005 def SSE_PHADDSUBW : OpndItins<
5006 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
5008 def SSE_PSHUFB : OpndItins<
5009 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
5011 def SSE_PSIGN : OpndItins<
5012 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
5014 def SSE_PMULHRSW : OpndItins<
5015 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
5018 /// SS3I_binop_rm - Simple SSSE3 bin op
5019 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5020 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
5021 X86MemOperand x86memop, OpndItins itins,
5023 let isCommutable = 1 in
5024 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
5025 (ins RC:$src1, RC:$src2),
5027 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5029 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
5031 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
5032 (ins RC:$src1, x86memop:$src2),
5034 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5035 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5037 (OpVT (OpNode RC:$src1,
5038 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
5041 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5042 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5043 Intrinsic IntId128, OpndItins itins,
5045 let isCommutable = 1 in
5046 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5047 (ins VR128:$src1, VR128:$src2),
5049 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5050 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5051 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5053 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5054 (ins VR128:$src1, i128mem:$src2),
5056 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5057 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5059 (IntId128 VR128:$src1,
5060 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5063 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5064 Intrinsic IntId256> {
5065 let isCommutable = 1 in
5066 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5067 (ins VR256:$src1, VR256:$src2),
5068 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5069 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5071 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5072 (ins VR256:$src1, i256mem:$src2),
5073 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5075 (IntId256 VR256:$src1,
5076 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5079 let ImmT = NoImm, Predicates = [HasAVX] in {
5080 let isCommutable = 0 in {
5081 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5082 memopv2i64, i128mem,
5083 SSE_PHADDSUBW, 0>, VEX_4V;
5084 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5085 memopv2i64, i128mem,
5086 SSE_PHADDSUBD, 0>, VEX_4V;
5087 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5088 memopv2i64, i128mem,
5089 SSE_PHADDSUBW, 0>, VEX_4V;
5090 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5091 memopv2i64, i128mem,
5092 SSE_PHADDSUBD, 0>, VEX_4V;
5093 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5094 memopv2i64, i128mem,
5095 SSE_PSIGN, 0>, VEX_4V;
5096 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5097 memopv2i64, i128mem,
5098 SSE_PSIGN, 0>, VEX_4V;
5099 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5100 memopv2i64, i128mem,
5101 SSE_PSIGN, 0>, VEX_4V;
5102 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5103 memopv2i64, i128mem,
5104 SSE_PSHUFB, 0>, VEX_4V;
5105 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5106 int_x86_ssse3_phadd_sw_128,
5107 SSE_PHADDSUBSW, 0>, VEX_4V;
5108 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5109 int_x86_ssse3_phsub_sw_128,
5110 SSE_PHADDSUBSW, 0>, VEX_4V;
5111 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5112 int_x86_ssse3_pmadd_ub_sw_128,
5113 SSE_PMADD, 0>, VEX_4V;
5115 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5116 int_x86_ssse3_pmul_hr_sw_128,
5117 SSE_PMULHRSW, 0>, VEX_4V;
5120 let ImmT = NoImm, Predicates = [HasAVX2] in {
5121 let isCommutable = 0 in {
5122 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5123 memopv4i64, i256mem,
5124 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5125 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5126 memopv4i64, i256mem,
5127 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5128 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5129 memopv4i64, i256mem,
5130 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5131 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5132 memopv4i64, i256mem,
5133 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5134 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5135 memopv4i64, i256mem,
5136 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5137 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5138 memopv4i64, i256mem,
5139 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5140 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5141 memopv4i64, i256mem,
5142 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5143 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5144 memopv4i64, i256mem,
5145 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5146 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5147 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5148 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5149 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5150 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5151 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5153 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5154 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5157 // None of these have i8 immediate fields.
5158 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5159 let isCommutable = 0 in {
5160 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5161 memopv2i64, i128mem, SSE_PHADDSUBW>;
5162 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5163 memopv2i64, i128mem, SSE_PHADDSUBD>;
5164 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5165 memopv2i64, i128mem, SSE_PHADDSUBW>;
5166 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5167 memopv2i64, i128mem, SSE_PHADDSUBD>;
5168 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5169 memopv2i64, i128mem, SSE_PSIGN>;
5170 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5171 memopv2i64, i128mem, SSE_PSIGN>;
5172 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5173 memopv2i64, i128mem, SSE_PSIGN>;
5174 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5175 memopv2i64, i128mem, SSE_PSHUFB>;
5176 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5177 int_x86_ssse3_phadd_sw_128,
5179 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5180 int_x86_ssse3_phsub_sw_128,
5182 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5183 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5185 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5186 int_x86_ssse3_pmul_hr_sw_128,
5190 //===---------------------------------------------------------------------===//
5191 // SSSE3 - Packed Align Instruction Patterns
5192 //===---------------------------------------------------------------------===//
5194 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5195 let neverHasSideEffects = 1 in {
5196 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5197 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5199 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5201 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5202 [], IIC_SSE_PALIGNR>, OpSize;
5204 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5205 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5207 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5209 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5210 [], IIC_SSE_PALIGNR>, OpSize;
5214 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5215 let neverHasSideEffects = 1 in {
5216 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5217 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5219 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5222 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5223 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5225 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5230 let Predicates = [HasAVX] in
5231 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5232 let Predicates = [HasAVX2] in
5233 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
5234 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5235 defm PALIGN : ssse3_palign<"palignr">;
5237 let Predicates = [HasAVX2] in {
5238 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5239 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5240 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5241 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5242 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5243 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5244 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5245 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5248 let Predicates = [HasAVX] in {
5249 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5250 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5251 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5252 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5253 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5254 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5255 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5256 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5259 let Predicates = [UseSSSE3] in {
5260 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5261 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5262 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5263 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5264 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5265 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5266 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5267 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5270 //===---------------------------------------------------------------------===//
5271 // SSSE3 - Thread synchronization
5272 //===---------------------------------------------------------------------===//
5274 let usesCustomInserter = 1 in {
5275 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5276 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5277 Requires<[HasSSE3]>;
5280 let Uses = [EAX, ECX, EDX] in
5281 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5282 TB, Requires<[HasSSE3]>;
5283 let Uses = [ECX, EAX] in
5284 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5285 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5286 TB, Requires<[HasSSE3]>;
5288 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5289 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5291 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5292 Requires<[In32BitMode]>;
5293 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5294 Requires<[In64BitMode]>;
5296 //===----------------------------------------------------------------------===//
5297 // SSE4.1 - Packed Move with Sign/Zero Extend
5298 //===----------------------------------------------------------------------===//
5300 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5301 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5302 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5303 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5305 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5306 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5308 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5312 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5314 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5315 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5316 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5318 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5319 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5320 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5323 let Predicates = [HasAVX] in {
5324 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5326 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5328 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5330 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5332 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5334 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5338 let Predicates = [HasAVX2] in {
5339 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5340 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5341 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5342 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5343 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5344 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5345 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5346 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5347 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5348 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5349 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5350 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5353 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5354 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5355 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5356 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5357 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5358 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5360 let Predicates = [HasAVX] in {
5361 // Common patterns involving scalar load.
5362 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5363 (VPMOVSXBWrm addr:$src)>;
5364 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5365 (VPMOVSXBWrm addr:$src)>;
5366 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5367 (VPMOVSXBWrm addr:$src)>;
5369 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5370 (VPMOVSXWDrm addr:$src)>;
5371 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5372 (VPMOVSXWDrm addr:$src)>;
5373 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5374 (VPMOVSXWDrm addr:$src)>;
5376 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5377 (VPMOVSXDQrm addr:$src)>;
5378 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5379 (VPMOVSXDQrm addr:$src)>;
5380 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5381 (VPMOVSXDQrm addr:$src)>;
5383 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5384 (VPMOVZXBWrm addr:$src)>;
5385 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5386 (VPMOVZXBWrm addr:$src)>;
5387 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5388 (VPMOVZXBWrm addr:$src)>;
5390 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5391 (VPMOVZXWDrm addr:$src)>;
5392 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5393 (VPMOVZXWDrm addr:$src)>;
5394 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5395 (VPMOVZXWDrm addr:$src)>;
5397 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5398 (VPMOVZXDQrm addr:$src)>;
5399 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5400 (VPMOVZXDQrm addr:$src)>;
5401 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5402 (VPMOVZXDQrm addr:$src)>;
5405 let Predicates = [UseSSE41] in {
5406 // Common patterns involving scalar load.
5407 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5408 (PMOVSXBWrm addr:$src)>;
5409 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5410 (PMOVSXBWrm addr:$src)>;
5411 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5412 (PMOVSXBWrm addr:$src)>;
5414 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5415 (PMOVSXWDrm addr:$src)>;
5416 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5417 (PMOVSXWDrm addr:$src)>;
5418 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5419 (PMOVSXWDrm addr:$src)>;
5421 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5422 (PMOVSXDQrm addr:$src)>;
5423 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5424 (PMOVSXDQrm addr:$src)>;
5425 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5426 (PMOVSXDQrm addr:$src)>;
5428 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5429 (PMOVZXBWrm addr:$src)>;
5430 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5431 (PMOVZXBWrm addr:$src)>;
5432 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5433 (PMOVZXBWrm addr:$src)>;
5435 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5436 (PMOVZXWDrm addr:$src)>;
5437 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5438 (PMOVZXWDrm addr:$src)>;
5439 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5440 (PMOVZXWDrm addr:$src)>;
5442 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5443 (PMOVZXDQrm addr:$src)>;
5444 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5445 (PMOVZXDQrm addr:$src)>;
5446 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5447 (PMOVZXDQrm addr:$src)>;
5450 let Predicates = [HasAVX2] in {
5451 let AddedComplexity = 15 in {
5452 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5453 (VPMOVZXDQYrr VR128:$src)>;
5454 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5455 (VPMOVZXWDYrr VR128:$src)>;
5458 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5459 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5462 let Predicates = [HasAVX] in {
5463 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5464 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5467 let Predicates = [UseSSE41] in {
5468 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5469 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5473 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5474 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5475 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5476 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5478 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5479 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5481 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5485 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5487 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5488 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5489 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5491 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5492 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5494 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5498 let Predicates = [HasAVX] in {
5499 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5501 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5503 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5505 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5509 let Predicates = [HasAVX2] in {
5510 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5511 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5512 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5513 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5514 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5515 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5516 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5517 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5520 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5521 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5522 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5523 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5525 let Predicates = [HasAVX] in {
5526 // Common patterns involving scalar load
5527 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5528 (VPMOVSXBDrm addr:$src)>;
5529 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5530 (VPMOVSXWQrm addr:$src)>;
5532 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5533 (VPMOVZXBDrm addr:$src)>;
5534 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5535 (VPMOVZXWQrm addr:$src)>;
5538 let Predicates = [UseSSE41] in {
5539 // Common patterns involving scalar load
5540 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5541 (PMOVSXBDrm addr:$src)>;
5542 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5543 (PMOVSXWQrm addr:$src)>;
5545 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5546 (PMOVZXBDrm addr:$src)>;
5547 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5548 (PMOVZXWQrm addr:$src)>;
5551 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5552 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5553 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5554 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5556 // Expecting a i16 load any extended to i32 value.
5557 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5558 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5559 [(set VR128:$dst, (IntId (bitconvert
5560 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5564 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5566 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5567 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5568 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5570 // Expecting a i16 load any extended to i32 value.
5571 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5572 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5573 [(set VR256:$dst, (IntId (bitconvert
5574 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5578 let Predicates = [HasAVX] in {
5579 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5581 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5584 let Predicates = [HasAVX2] in {
5585 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5586 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5587 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5588 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5590 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5591 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5593 let Predicates = [HasAVX2] in {
5594 def : Pat<(v16i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWYrr VR128:$src)>;
5595 def : Pat<(v8i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDYrr VR128:$src)>;
5596 def : Pat<(v4i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQYrr VR128:$src)>;
5598 def : Pat<(v8i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5599 def : Pat<(v4i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQYrr VR128:$src)>;
5601 def : Pat<(v4i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5603 def : Pat<(v16i16 (X86vsext (v32i8 VR256:$src))),
5604 (VPMOVSXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5605 def : Pat<(v8i32 (X86vsext (v32i8 VR256:$src))),
5606 (VPMOVSXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5607 def : Pat<(v4i64 (X86vsext (v32i8 VR256:$src))),
5608 (VPMOVSXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5610 def : Pat<(v8i32 (X86vsext (v16i16 VR256:$src))),
5611 (VPMOVSXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5612 def : Pat<(v4i64 (X86vsext (v16i16 VR256:$src))),
5613 (VPMOVSXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5615 def : Pat<(v4i64 (X86vsext (v8i32 VR256:$src))),
5616 (VPMOVSXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5618 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5619 (VPMOVSXWDYrm addr:$src)>;
5620 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5621 (VPMOVSXDQYrm addr:$src)>;
5623 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5624 (scalar_to_vector (loadi64 addr:$src))))))),
5625 (VPMOVSXBDYrm addr:$src)>;
5626 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5627 (scalar_to_vector (loadf64 addr:$src))))))),
5628 (VPMOVSXBDYrm addr:$src)>;
5630 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5631 (scalar_to_vector (loadi64 addr:$src))))))),
5632 (VPMOVSXWQYrm addr:$src)>;
5633 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5634 (scalar_to_vector (loadf64 addr:$src))))))),
5635 (VPMOVSXWQYrm addr:$src)>;
5637 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5638 (scalar_to_vector (loadi32 addr:$src))))))),
5639 (VPMOVSXBQYrm addr:$src)>;
5642 let Predicates = [HasAVX] in {
5643 // Common patterns involving scalar load
5644 def : Pat<(int_x86_sse41_pmovsxbq
5645 (bitconvert (v4i32 (X86vzmovl
5646 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5647 (VPMOVSXBQrm addr:$src)>;
5649 def : Pat<(int_x86_sse41_pmovzxbq
5650 (bitconvert (v4i32 (X86vzmovl
5651 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5652 (VPMOVZXBQrm addr:$src)>;
5655 let Predicates = [UseSSE41] in {
5656 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (PMOVSXBWrr VR128:$src)>;
5657 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (PMOVSXBDrr VR128:$src)>;
5658 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (PMOVSXBQrr VR128:$src)>;
5660 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5661 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (PMOVSXWQrr VR128:$src)>;
5663 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5665 // Common patterns involving scalar load
5666 def : Pat<(int_x86_sse41_pmovsxbq
5667 (bitconvert (v4i32 (X86vzmovl
5668 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5669 (PMOVSXBQrm addr:$src)>;
5671 def : Pat<(int_x86_sse41_pmovzxbq
5672 (bitconvert (v4i32 (X86vzmovl
5673 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5674 (PMOVZXBQrm addr:$src)>;
5676 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5677 (scalar_to_vector (loadi64 addr:$src))))))),
5678 (PMOVSXWDrm addr:$src)>;
5679 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5680 (scalar_to_vector (loadf64 addr:$src))))))),
5681 (PMOVSXWDrm addr:$src)>;
5682 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5683 (scalar_to_vector (loadi32 addr:$src))))))),
5684 (PMOVSXBDrm addr:$src)>;
5685 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5686 (scalar_to_vector (loadi32 addr:$src))))))),
5687 (PMOVSXWQrm addr:$src)>;
5688 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5689 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5690 (PMOVSXBQrm addr:$src)>;
5691 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5692 (scalar_to_vector (loadi64 addr:$src))))))),
5693 (PMOVSXDQrm addr:$src)>;
5694 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5695 (scalar_to_vector (loadf64 addr:$src))))))),
5696 (PMOVSXDQrm addr:$src)>;
5697 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5698 (scalar_to_vector (loadi64 addr:$src))))))),
5699 (PMOVSXBWrm addr:$src)>;
5700 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5701 (scalar_to_vector (loadf64 addr:$src))))))),
5702 (PMOVSXBWrm addr:$src)>;
5705 let Predicates = [HasAVX2] in {
5706 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5707 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5708 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5710 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5711 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5713 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5715 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5716 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5717 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5718 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5719 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5720 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5722 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5723 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5724 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5725 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5727 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5728 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5731 let Predicates = [HasAVX] in {
5732 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5733 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5734 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5736 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5737 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5739 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5741 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5742 (VPMOVZXBWrm addr:$src)>;
5743 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5744 (VPMOVZXBWrm addr:$src)>;
5745 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5746 (VPMOVZXBDrm addr:$src)>;
5747 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5748 (VPMOVZXBQrm addr:$src)>;
5750 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5751 (VPMOVZXWDrm addr:$src)>;
5752 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5753 (VPMOVZXWDrm addr:$src)>;
5754 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5755 (VPMOVZXWQrm addr:$src)>;
5757 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5758 (VPMOVZXDQrm addr:$src)>;
5759 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5760 (VPMOVZXDQrm addr:$src)>;
5761 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5762 (VPMOVZXDQrm addr:$src)>;
5764 def : Pat<(v8i16 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBWrr VR128:$src)>;
5765 def : Pat<(v4i32 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBDrr VR128:$src)>;
5766 def : Pat<(v2i64 (X86vsext (v16i8 VR128:$src))), (VPMOVSXBQrr VR128:$src)>;
5768 def : Pat<(v4i32 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5769 def : Pat<(v2i64 (X86vsext (v8i16 VR128:$src))), (VPMOVSXWQrr VR128:$src)>;
5771 def : Pat<(v2i64 (X86vsext (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5773 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5774 (scalar_to_vector (loadi64 addr:$src))))))),
5775 (VPMOVSXWDrm addr:$src)>;
5776 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5777 (scalar_to_vector (loadi64 addr:$src))))))),
5778 (VPMOVSXDQrm addr:$src)>;
5779 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5780 (scalar_to_vector (loadf64 addr:$src))))))),
5781 (VPMOVSXWDrm addr:$src)>;
5782 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5783 (scalar_to_vector (loadf64 addr:$src))))))),
5784 (VPMOVSXDQrm addr:$src)>;
5785 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5786 (scalar_to_vector (loadi64 addr:$src))))))),
5787 (VPMOVSXBWrm addr:$src)>;
5788 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5789 (scalar_to_vector (loadf64 addr:$src))))))),
5790 (VPMOVSXBWrm addr:$src)>;
5792 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5793 (scalar_to_vector (loadi32 addr:$src))))))),
5794 (VPMOVSXBDrm addr:$src)>;
5795 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5796 (scalar_to_vector (loadi32 addr:$src))))))),
5797 (VPMOVSXWQrm addr:$src)>;
5798 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5799 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5800 (VPMOVSXBQrm addr:$src)>;
5803 let Predicates = [UseSSE41] in {
5804 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5805 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5806 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5808 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5809 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5811 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5813 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5814 (PMOVZXBWrm addr:$src)>;
5815 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5816 (PMOVZXBWrm addr:$src)>;
5817 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5818 (PMOVZXBDrm addr:$src)>;
5819 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5820 (PMOVZXBQrm addr:$src)>;
5822 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5823 (PMOVZXWDrm addr:$src)>;
5824 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5825 (PMOVZXWDrm addr:$src)>;
5826 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5827 (PMOVZXWQrm addr:$src)>;
5829 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5830 (PMOVZXDQrm addr:$src)>;
5831 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5832 (PMOVZXDQrm addr:$src)>;
5833 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5834 (PMOVZXDQrm addr:$src)>;
5837 //===----------------------------------------------------------------------===//
5838 // SSE4.1 - Extract Instructions
5839 //===----------------------------------------------------------------------===//
5841 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5842 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5843 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5844 (ins VR128:$src1, i32i8imm:$src2),
5845 !strconcat(OpcodeStr,
5846 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5847 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5849 let neverHasSideEffects = 1, mayStore = 1 in
5850 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5851 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5852 !strconcat(OpcodeStr,
5853 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5856 // There's an AssertZext in the way of writing the store pattern
5857 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5860 let Predicates = [HasAVX] in {
5861 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5862 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5863 (ins VR128:$src1, i32i8imm:$src2),
5864 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5867 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5870 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5871 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5872 let neverHasSideEffects = 1, mayStore = 1 in
5873 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5874 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5875 !strconcat(OpcodeStr,
5876 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5879 // There's an AssertZext in the way of writing the store pattern
5880 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5883 let Predicates = [HasAVX] in
5884 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5886 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5889 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5890 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5891 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5892 (ins VR128:$src1, i32i8imm:$src2),
5893 !strconcat(OpcodeStr,
5894 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5896 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5897 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5898 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5899 !strconcat(OpcodeStr,
5900 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5901 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5902 addr:$dst)]>, OpSize;
5905 let Predicates = [HasAVX] in
5906 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5908 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5910 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5911 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5912 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5913 (ins VR128:$src1, i32i8imm:$src2),
5914 !strconcat(OpcodeStr,
5915 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5917 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5918 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5919 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5920 !strconcat(OpcodeStr,
5921 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5922 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5923 addr:$dst)]>, OpSize, REX_W;
5926 let Predicates = [HasAVX] in
5927 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5929 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5931 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5933 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5934 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5935 (ins VR128:$src1, i32i8imm:$src2),
5936 !strconcat(OpcodeStr,
5937 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5939 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5941 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5942 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5943 !strconcat(OpcodeStr,
5944 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5945 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5946 addr:$dst)]>, OpSize;
5949 let ExeDomain = SSEPackedSingle in {
5950 let Predicates = [HasAVX] in {
5951 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5952 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5953 (ins VR128:$src1, i32i8imm:$src2),
5954 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5957 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5960 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5961 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5964 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5966 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5969 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5970 Requires<[UseSSE41]>;
5972 //===----------------------------------------------------------------------===//
5973 // SSE4.1 - Insert Instructions
5974 //===----------------------------------------------------------------------===//
5976 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5977 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5978 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5980 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5982 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5984 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5985 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5986 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5988 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5990 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5992 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5993 imm:$src3))]>, OpSize;
5996 let Predicates = [HasAVX] in
5997 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5998 let Constraints = "$src1 = $dst" in
5999 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
6001 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
6002 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6003 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
6005 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6007 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6009 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
6011 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6012 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
6014 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6016 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6018 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
6019 imm:$src3)))]>, OpSize;
6022 let Predicates = [HasAVX] in
6023 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
6024 let Constraints = "$src1 = $dst" in
6025 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
6027 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
6028 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6029 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
6031 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6033 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6035 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
6037 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6038 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
6040 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6042 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6044 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
6045 imm:$src3)))]>, OpSize;
6048 let Predicates = [HasAVX] in
6049 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6050 let Constraints = "$src1 = $dst" in
6051 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6053 // insertps has a few different modes, there's the first two here below which
6054 // are optimized inserts that won't zero arbitrary elements in the destination
6055 // vector. The next one matches the intrinsic and could zero arbitrary elements
6056 // in the target vector.
6057 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6058 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6059 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6061 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6063 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6065 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6067 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6068 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6070 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6072 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6074 (X86insrtps VR128:$src1,
6075 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6076 imm:$src3))]>, OpSize;
6079 let ExeDomain = SSEPackedSingle in {
6080 let Predicates = [HasAVX] in
6081 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6082 let Constraints = "$src1 = $dst" in
6083 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6086 //===----------------------------------------------------------------------===//
6087 // SSE4.1 - Round Instructions
6088 //===----------------------------------------------------------------------===//
6090 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6091 X86MemOperand x86memop, RegisterClass RC,
6092 PatFrag mem_frag32, PatFrag mem_frag64,
6093 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6094 let ExeDomain = SSEPackedSingle in {
6095 // Intrinsic operation, reg.
6096 // Vector intrinsic operation, reg
6097 def PSr : SS4AIi8<opcps, MRMSrcReg,
6098 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6099 !strconcat(OpcodeStr,
6100 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6101 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6104 // Vector intrinsic operation, mem
6105 def PSm : SS4AIi8<opcps, MRMSrcMem,
6106 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6107 !strconcat(OpcodeStr,
6108 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6110 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6112 } // ExeDomain = SSEPackedSingle
6114 let ExeDomain = SSEPackedDouble in {
6115 // Vector intrinsic operation, reg
6116 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6117 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6118 !strconcat(OpcodeStr,
6119 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6120 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6123 // Vector intrinsic operation, mem
6124 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6125 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6126 !strconcat(OpcodeStr,
6127 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6129 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6131 } // ExeDomain = SSEPackedDouble
6134 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6137 Intrinsic F64Int, bit Is2Addr = 1> {
6138 let ExeDomain = GenericDomain in {
6140 let hasSideEffects = 0 in
6141 def SSr : SS4AIi8<opcss, MRMSrcReg,
6142 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6144 !strconcat(OpcodeStr,
6145 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6146 !strconcat(OpcodeStr,
6147 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6150 // Intrinsic operation, reg.
6151 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6152 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6154 !strconcat(OpcodeStr,
6155 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6156 !strconcat(OpcodeStr,
6157 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6158 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6161 // Intrinsic operation, mem.
6162 def SSm : SS4AIi8<opcss, MRMSrcMem,
6163 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6165 !strconcat(OpcodeStr,
6166 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6167 !strconcat(OpcodeStr,
6168 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6170 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6174 let hasSideEffects = 0 in
6175 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6176 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6178 !strconcat(OpcodeStr,
6179 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6180 !strconcat(OpcodeStr,
6181 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6184 // Intrinsic operation, reg.
6185 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6186 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6188 !strconcat(OpcodeStr,
6189 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6190 !strconcat(OpcodeStr,
6191 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6192 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6195 // Intrinsic operation, mem.
6196 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6197 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6199 !strconcat(OpcodeStr,
6200 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6201 !strconcat(OpcodeStr,
6202 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6204 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6206 } // ExeDomain = GenericDomain
6209 // FP round - roundss, roundps, roundsd, roundpd
6210 let Predicates = [HasAVX] in {
6212 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6213 memopv4f32, memopv2f64,
6214 int_x86_sse41_round_ps,
6215 int_x86_sse41_round_pd>, VEX;
6216 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6217 memopv8f32, memopv4f64,
6218 int_x86_avx_round_ps_256,
6219 int_x86_avx_round_pd_256>, VEX, VEX_L;
6220 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6221 int_x86_sse41_round_ss,
6222 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6224 def : Pat<(ffloor FR32:$src),
6225 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6226 def : Pat<(f64 (ffloor FR64:$src)),
6227 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6228 def : Pat<(f32 (fnearbyint FR32:$src)),
6229 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6230 def : Pat<(f64 (fnearbyint FR64:$src)),
6231 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6232 def : Pat<(f32 (fceil FR32:$src)),
6233 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6234 def : Pat<(f64 (fceil FR64:$src)),
6235 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6236 def : Pat<(f32 (frint FR32:$src)),
6237 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6238 def : Pat<(f64 (frint FR64:$src)),
6239 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6240 def : Pat<(f32 (ftrunc FR32:$src)),
6241 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6242 def : Pat<(f64 (ftrunc FR64:$src)),
6243 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6245 def : Pat<(v4f32 (ffloor VR128:$src)),
6246 (VROUNDPSr VR128:$src, (i32 0x1))>;
6247 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6248 (VROUNDPSr VR128:$src, (i32 0xC))>;
6249 def : Pat<(v4f32 (fceil VR128:$src)),
6250 (VROUNDPSr VR128:$src, (i32 0x2))>;
6251 def : Pat<(v4f32 (frint VR128:$src)),
6252 (VROUNDPSr VR128:$src, (i32 0x4))>;
6253 def : Pat<(v4f32 (ftrunc VR128:$src)),
6254 (VROUNDPSr VR128:$src, (i32 0x3))>;
6256 def : Pat<(v2f64 (ffloor VR128:$src)),
6257 (VROUNDPDr VR128:$src, (i32 0x1))>;
6258 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6259 (VROUNDPDr VR128:$src, (i32 0xC))>;
6260 def : Pat<(v2f64 (fceil VR128:$src)),
6261 (VROUNDPDr VR128:$src, (i32 0x2))>;
6262 def : Pat<(v2f64 (frint VR128:$src)),
6263 (VROUNDPDr VR128:$src, (i32 0x4))>;
6264 def : Pat<(v2f64 (ftrunc VR128:$src)),
6265 (VROUNDPDr VR128:$src, (i32 0x3))>;
6267 def : Pat<(v8f32 (ffloor VR256:$src)),
6268 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6269 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6270 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6271 def : Pat<(v8f32 (fceil VR256:$src)),
6272 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6273 def : Pat<(v8f32 (frint VR256:$src)),
6274 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6275 def : Pat<(v8f32 (ftrunc VR256:$src)),
6276 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6278 def : Pat<(v4f64 (ffloor VR256:$src)),
6279 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6280 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6281 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6282 def : Pat<(v4f64 (fceil VR256:$src)),
6283 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6284 def : Pat<(v4f64 (frint VR256:$src)),
6285 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6286 def : Pat<(v4f64 (ftrunc VR256:$src)),
6287 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6290 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6291 memopv4f32, memopv2f64,
6292 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6293 let Constraints = "$src1 = $dst" in
6294 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6295 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6297 let Predicates = [UseSSE41] in {
6298 def : Pat<(ffloor FR32:$src),
6299 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6300 def : Pat<(f64 (ffloor FR64:$src)),
6301 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6302 def : Pat<(f32 (fnearbyint FR32:$src)),
6303 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6304 def : Pat<(f64 (fnearbyint FR64:$src)),
6305 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6306 def : Pat<(f32 (fceil FR32:$src)),
6307 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6308 def : Pat<(f64 (fceil FR64:$src)),
6309 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6310 def : Pat<(f32 (frint FR32:$src)),
6311 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6312 def : Pat<(f64 (frint FR64:$src)),
6313 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6314 def : Pat<(f32 (ftrunc FR32:$src)),
6315 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6316 def : Pat<(f64 (ftrunc FR64:$src)),
6317 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6319 def : Pat<(v4f32 (ffloor VR128:$src)),
6320 (ROUNDPSr VR128:$src, (i32 0x1))>;
6321 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6322 (ROUNDPSr VR128:$src, (i32 0xC))>;
6323 def : Pat<(v4f32 (fceil VR128:$src)),
6324 (ROUNDPSr VR128:$src, (i32 0x2))>;
6325 def : Pat<(v4f32 (frint VR128:$src)),
6326 (ROUNDPSr VR128:$src, (i32 0x4))>;
6327 def : Pat<(v4f32 (ftrunc VR128:$src)),
6328 (ROUNDPSr VR128:$src, (i32 0x3))>;
6330 def : Pat<(v2f64 (ffloor VR128:$src)),
6331 (ROUNDPDr VR128:$src, (i32 0x1))>;
6332 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6333 (ROUNDPDr VR128:$src, (i32 0xC))>;
6334 def : Pat<(v2f64 (fceil VR128:$src)),
6335 (ROUNDPDr VR128:$src, (i32 0x2))>;
6336 def : Pat<(v2f64 (frint VR128:$src)),
6337 (ROUNDPDr VR128:$src, (i32 0x4))>;
6338 def : Pat<(v2f64 (ftrunc VR128:$src)),
6339 (ROUNDPDr VR128:$src, (i32 0x3))>;
6342 //===----------------------------------------------------------------------===//
6343 // SSE4.1 - Packed Bit Test
6344 //===----------------------------------------------------------------------===//
6346 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6347 // the intel intrinsic that corresponds to this.
6348 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6349 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6350 "vptest\t{$src2, $src1|$src1, $src2}",
6351 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6353 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6354 "vptest\t{$src2, $src1|$src1, $src2}",
6355 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6358 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6359 "vptest\t{$src2, $src1|$src1, $src2}",
6360 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6362 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6363 "vptest\t{$src2, $src1|$src1, $src2}",
6364 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6368 let Defs = [EFLAGS] in {
6369 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6370 "ptest\t{$src2, $src1|$src1, $src2}",
6371 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6373 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6374 "ptest\t{$src2, $src1|$src1, $src2}",
6375 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6379 // The bit test instructions below are AVX only
6380 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6381 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6382 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6383 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6384 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6385 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6386 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6387 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6391 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6392 let ExeDomain = SSEPackedSingle in {
6393 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6394 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6397 let ExeDomain = SSEPackedDouble in {
6398 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6399 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6404 //===----------------------------------------------------------------------===//
6405 // SSE4.1 - Misc Instructions
6406 //===----------------------------------------------------------------------===//
6408 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6409 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6410 "popcnt{w}\t{$src, $dst|$dst, $src}",
6411 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6413 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6414 "popcnt{w}\t{$src, $dst|$dst, $src}",
6415 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6416 (implicit EFLAGS)]>, OpSize, XS;
6418 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6419 "popcnt{l}\t{$src, $dst|$dst, $src}",
6420 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6422 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6423 "popcnt{l}\t{$src, $dst|$dst, $src}",
6424 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6425 (implicit EFLAGS)]>, XS;
6427 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6428 "popcnt{q}\t{$src, $dst|$dst, $src}",
6429 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6431 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6432 "popcnt{q}\t{$src, $dst|$dst, $src}",
6433 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6434 (implicit EFLAGS)]>, XS;
6439 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6440 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6441 Intrinsic IntId128> {
6442 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6444 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6445 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6446 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6448 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6451 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6454 let Predicates = [HasAVX] in
6455 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6456 int_x86_sse41_phminposuw>, VEX;
6457 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6458 int_x86_sse41_phminposuw>;
6460 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6461 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6462 Intrinsic IntId128, bit Is2Addr = 1> {
6463 let isCommutable = 1 in
6464 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6465 (ins VR128:$src1, VR128:$src2),
6467 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6469 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6470 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6471 (ins VR128:$src1, i128mem:$src2),
6473 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6476 (IntId128 VR128:$src1,
6477 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6480 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6481 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6482 Intrinsic IntId256> {
6483 let isCommutable = 1 in
6484 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6485 (ins VR256:$src1, VR256:$src2),
6486 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6487 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6488 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6489 (ins VR256:$src1, i256mem:$src2),
6490 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6492 (IntId256 VR256:$src1,
6493 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6497 /// SS48I_binop_rm - Simple SSE41 binary operator.
6498 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6499 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6500 X86MemOperand x86memop, bit Is2Addr = 1> {
6501 let isCommutable = 1 in
6502 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6503 (ins RC:$src1, RC:$src2),
6505 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6506 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6507 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6508 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6509 (ins RC:$src1, x86memop:$src2),
6511 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6512 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6514 (OpVT (OpNode RC:$src1,
6515 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6518 let Predicates = [HasAVX] in {
6519 let isCommutable = 0 in
6520 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6522 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6523 memopv2i64, i128mem, 0>, VEX_4V;
6524 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6525 memopv2i64, i128mem, 0>, VEX_4V;
6526 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6527 memopv2i64, i128mem, 0>, VEX_4V;
6528 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6529 memopv2i64, i128mem, 0>, VEX_4V;
6530 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6531 memopv2i64, i128mem, 0>, VEX_4V;
6532 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6533 memopv2i64, i128mem, 0>, VEX_4V;
6534 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6535 memopv2i64, i128mem, 0>, VEX_4V;
6536 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6537 memopv2i64, i128mem, 0>, VEX_4V;
6538 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6542 let Predicates = [HasAVX2] in {
6543 let isCommutable = 0 in
6544 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6545 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6546 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6547 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6548 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6549 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6550 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6551 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6552 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6553 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6554 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6555 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6556 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6557 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6558 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6559 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6560 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6561 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6562 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6563 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6566 let Constraints = "$src1 = $dst" in {
6567 let isCommutable = 0 in
6568 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6569 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6570 memopv2i64, i128mem>;
6571 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6572 memopv2i64, i128mem>;
6573 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6574 memopv2i64, i128mem>;
6575 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6576 memopv2i64, i128mem>;
6577 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6578 memopv2i64, i128mem>;
6579 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6580 memopv2i64, i128mem>;
6581 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6582 memopv2i64, i128mem>;
6583 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6584 memopv2i64, i128mem>;
6585 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6588 let Predicates = [HasAVX] in {
6589 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6590 memopv2i64, i128mem, 0>, VEX_4V;
6591 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6592 memopv2i64, i128mem, 0>, VEX_4V;
6594 let Predicates = [HasAVX2] in {
6595 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6596 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6597 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6598 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6601 let Constraints = "$src1 = $dst" in {
6602 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6603 memopv2i64, i128mem>;
6604 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6605 memopv2i64, i128mem>;
6608 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6609 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6610 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6611 X86MemOperand x86memop, bit Is2Addr = 1> {
6612 let isCommutable = 1 in
6613 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6614 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6616 !strconcat(OpcodeStr,
6617 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6618 !strconcat(OpcodeStr,
6619 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6620 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6622 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6623 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6625 !strconcat(OpcodeStr,
6626 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6627 !strconcat(OpcodeStr,
6628 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6631 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6635 let Predicates = [HasAVX] in {
6636 let isCommutable = 0 in {
6637 let ExeDomain = SSEPackedSingle in {
6638 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6639 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6640 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6641 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6642 f256mem, 0>, VEX_4V, VEX_L;
6644 let ExeDomain = SSEPackedDouble in {
6645 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6646 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6647 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6648 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6649 f256mem, 0>, VEX_4V, VEX_L;
6651 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6652 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6653 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6654 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6656 let ExeDomain = SSEPackedSingle in
6657 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6658 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6659 let ExeDomain = SSEPackedDouble in
6660 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6661 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6662 let ExeDomain = SSEPackedSingle in
6663 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6664 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6667 let Predicates = [HasAVX2] in {
6668 let isCommutable = 0 in {
6669 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6670 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6671 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6672 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6676 let Constraints = "$src1 = $dst" in {
6677 let isCommutable = 0 in {
6678 let ExeDomain = SSEPackedSingle in
6679 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6680 VR128, memopv4f32, f128mem>;
6681 let ExeDomain = SSEPackedDouble in
6682 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6683 VR128, memopv2f64, f128mem>;
6684 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6685 VR128, memopv2i64, i128mem>;
6686 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6687 VR128, memopv2i64, i128mem>;
6689 let ExeDomain = SSEPackedSingle in
6690 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6691 VR128, memopv4f32, f128mem>;
6692 let ExeDomain = SSEPackedDouble in
6693 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6694 VR128, memopv2f64, f128mem>;
6697 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6698 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6699 RegisterClass RC, X86MemOperand x86memop,
6700 PatFrag mem_frag, Intrinsic IntId> {
6701 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6702 (ins RC:$src1, RC:$src2, RC:$src3),
6703 !strconcat(OpcodeStr,
6704 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6705 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6706 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6708 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6709 (ins RC:$src1, x86memop:$src2, RC:$src3),
6710 !strconcat(OpcodeStr,
6711 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6713 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6715 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6718 let Predicates = [HasAVX] in {
6719 let ExeDomain = SSEPackedDouble in {
6720 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6721 memopv2f64, int_x86_sse41_blendvpd>;
6722 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6723 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6724 } // ExeDomain = SSEPackedDouble
6725 let ExeDomain = SSEPackedSingle in {
6726 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6727 memopv4f32, int_x86_sse41_blendvps>;
6728 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6729 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6730 } // ExeDomain = SSEPackedSingle
6731 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6732 memopv2i64, int_x86_sse41_pblendvb>;
6735 let Predicates = [HasAVX2] in {
6736 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6737 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6740 let Predicates = [HasAVX] in {
6741 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6742 (v16i8 VR128:$src2))),
6743 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6744 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6745 (v4i32 VR128:$src2))),
6746 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6747 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6748 (v4f32 VR128:$src2))),
6749 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6750 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6751 (v2i64 VR128:$src2))),
6752 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6753 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6754 (v2f64 VR128:$src2))),
6755 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6756 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6757 (v8i32 VR256:$src2))),
6758 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6759 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6760 (v8f32 VR256:$src2))),
6761 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6762 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6763 (v4i64 VR256:$src2))),
6764 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6765 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6766 (v4f64 VR256:$src2))),
6767 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6769 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6771 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6772 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6774 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6776 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6778 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6779 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6781 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6782 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6784 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6787 let Predicates = [HasAVX2] in {
6788 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6789 (v32i8 VR256:$src2))),
6790 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6791 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6793 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6796 /// SS41I_ternary_int - SSE 4.1 ternary operator
6797 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6798 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6799 X86MemOperand x86memop, Intrinsic IntId> {
6800 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6801 (ins VR128:$src1, VR128:$src2),
6802 !strconcat(OpcodeStr,
6803 "\t{$src2, $dst|$dst, $src2}"),
6804 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6807 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6808 (ins VR128:$src1, x86memop:$src2),
6809 !strconcat(OpcodeStr,
6810 "\t{$src2, $dst|$dst, $src2}"),
6813 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6817 let ExeDomain = SSEPackedDouble in
6818 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6819 int_x86_sse41_blendvpd>;
6820 let ExeDomain = SSEPackedSingle in
6821 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6822 int_x86_sse41_blendvps>;
6823 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6824 int_x86_sse41_pblendvb>;
6826 // Aliases with the implicit xmm0 argument
6827 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6828 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6829 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6830 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6831 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6832 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6833 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6834 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6835 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6836 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6837 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6838 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6840 let Predicates = [UseSSE41] in {
6841 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6842 (v16i8 VR128:$src2))),
6843 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6844 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6845 (v4i32 VR128:$src2))),
6846 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6847 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6848 (v4f32 VR128:$src2))),
6849 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6850 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6851 (v2i64 VR128:$src2))),
6852 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6853 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6854 (v2f64 VR128:$src2))),
6855 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6857 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6859 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6860 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6862 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6863 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6865 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6869 let Predicates = [HasAVX] in
6870 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6871 "vmovntdqa\t{$src, $dst|$dst, $src}",
6872 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6874 let Predicates = [HasAVX2] in
6875 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6876 "vmovntdqa\t{$src, $dst|$dst, $src}",
6877 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6879 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6880 "movntdqa\t{$src, $dst|$dst, $src}",
6881 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6884 //===----------------------------------------------------------------------===//
6885 // SSE4.2 - Compare Instructions
6886 //===----------------------------------------------------------------------===//
6888 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6889 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6890 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6891 X86MemOperand x86memop, bit Is2Addr = 1> {
6892 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6893 (ins RC:$src1, RC:$src2),
6895 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6896 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6897 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6899 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6900 (ins RC:$src1, x86memop:$src2),
6902 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6905 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6908 let Predicates = [HasAVX] in
6909 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6910 memopv2i64, i128mem, 0>, VEX_4V;
6912 let Predicates = [HasAVX2] in
6913 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6914 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6916 let Constraints = "$src1 = $dst" in
6917 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6918 memopv2i64, i128mem>;
6920 //===----------------------------------------------------------------------===//
6921 // SSE4.2 - String/text Processing Instructions
6922 //===----------------------------------------------------------------------===//
6924 // Packed Compare Implicit Length Strings, Return Mask
6925 multiclass pseudo_pcmpistrm<string asm> {
6926 def REG : PseudoI<(outs VR128:$dst),
6927 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6928 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6930 def MEM : PseudoI<(outs VR128:$dst),
6931 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6932 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
6933 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6936 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6937 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6938 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
6941 multiclass pcmpistrm_SS42AI<string asm> {
6942 def rr : SS42AI<0x62, MRMSrcReg, (outs),
6943 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6944 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6947 def rm :SS42AI<0x62, MRMSrcMem, (outs),
6948 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6949 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6953 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6954 let Predicates = [HasAVX] in
6955 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
6956 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
6959 // Packed Compare Explicit Length Strings, Return Mask
6960 multiclass pseudo_pcmpestrm<string asm> {
6961 def REG : PseudoI<(outs VR128:$dst),
6962 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6963 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6964 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6965 def MEM : PseudoI<(outs VR128:$dst),
6966 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6967 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
6968 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
6971 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6972 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6973 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
6976 multiclass SS42AI_pcmpestrm<string asm> {
6977 def rr : SS42AI<0x60, MRMSrcReg, (outs),
6978 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6979 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6982 def rm : SS42AI<0x60, MRMSrcMem, (outs),
6983 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6984 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6988 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6989 let Predicates = [HasAVX] in
6990 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
6991 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
6994 // Packed Compare Implicit Length Strings, Return Index
6995 multiclass pseudo_pcmpistri<string asm> {
6996 def REG : PseudoI<(outs GR32:$dst),
6997 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6998 [(set GR32:$dst, EFLAGS,
6999 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
7000 def MEM : PseudoI<(outs GR32:$dst),
7001 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7002 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
7003 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
7006 let Defs = [EFLAGS], usesCustomInserter = 1 in {
7007 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
7008 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
7011 multiclass SS42AI_pcmpistri<string asm> {
7012 def rr : SS42AI<0x63, MRMSrcReg, (outs),
7013 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7014 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7017 def rm : SS42AI<0x63, MRMSrcMem, (outs),
7018 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7019 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
7023 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
7024 let Predicates = [HasAVX] in
7025 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
7026 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
7029 // Packed Compare Explicit Length Strings, Return Index
7030 multiclass pseudo_pcmpestri<string asm> {
7031 def REG : PseudoI<(outs GR32:$dst),
7032 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7033 [(set GR32:$dst, EFLAGS,
7034 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
7035 def MEM : PseudoI<(outs GR32:$dst),
7036 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7037 [(set GR32:$dst, EFLAGS,
7038 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
7042 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
7043 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
7044 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
7047 multiclass SS42AI_pcmpestri<string asm> {
7048 def rr : SS42AI<0x61, MRMSrcReg, (outs),
7049 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
7050 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7053 def rm : SS42AI<0x61, MRMSrcMem, (outs),
7054 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
7055 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
7059 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
7060 let Predicates = [HasAVX] in
7061 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
7062 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
7065 //===----------------------------------------------------------------------===//
7066 // SSE4.2 - CRC Instructions
7067 //===----------------------------------------------------------------------===//
7069 // No CRC instructions have AVX equivalents
7071 // crc intrinsic instruction
7072 // This set of instructions are only rm, the only difference is the size
7074 let Constraints = "$src1 = $dst" in {
7075 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
7076 (ins GR32:$src1, i8mem:$src2),
7077 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7079 (int_x86_sse42_crc32_32_8 GR32:$src1,
7080 (load addr:$src2)))]>;
7081 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
7082 (ins GR32:$src1, GR8:$src2),
7083 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7085 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
7086 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7087 (ins GR32:$src1, i16mem:$src2),
7088 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7090 (int_x86_sse42_crc32_32_16 GR32:$src1,
7091 (load addr:$src2)))]>,
7093 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7094 (ins GR32:$src1, GR16:$src2),
7095 "crc32{w} \t{$src2, $src1|$src1, $src2}",
7097 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
7099 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
7100 (ins GR32:$src1, i32mem:$src2),
7101 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7103 (int_x86_sse42_crc32_32_32 GR32:$src1,
7104 (load addr:$src2)))]>;
7105 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7106 (ins GR32:$src1, GR32:$src2),
7107 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7109 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7110 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7111 (ins GR64:$src1, i8mem:$src2),
7112 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7114 (int_x86_sse42_crc32_64_8 GR64:$src1,
7115 (load addr:$src2)))]>,
7117 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7118 (ins GR64:$src1, GR8:$src2),
7119 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7121 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7123 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7124 (ins GR64:$src1, i64mem:$src2),
7125 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7127 (int_x86_sse42_crc32_64_64 GR64:$src1,
7128 (load addr:$src2)))]>,
7130 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7131 (ins GR64:$src1, GR64:$src2),
7132 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7134 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7138 //===----------------------------------------------------------------------===//
7139 // AES-NI Instructions
7140 //===----------------------------------------------------------------------===//
7142 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7143 Intrinsic IntId128, bit Is2Addr = 1> {
7144 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7145 (ins VR128:$src1, VR128:$src2),
7147 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7148 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7149 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7151 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7152 (ins VR128:$src1, i128mem:$src2),
7154 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7155 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7157 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7160 // Perform One Round of an AES Encryption/Decryption Flow
7161 let Predicates = [HasAVX, HasAES] in {
7162 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7163 int_x86_aesni_aesenc, 0>, VEX_4V;
7164 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7165 int_x86_aesni_aesenclast, 0>, VEX_4V;
7166 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7167 int_x86_aesni_aesdec, 0>, VEX_4V;
7168 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7169 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7172 let Constraints = "$src1 = $dst" in {
7173 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7174 int_x86_aesni_aesenc>;
7175 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7176 int_x86_aesni_aesenclast>;
7177 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7178 int_x86_aesni_aesdec>;
7179 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7180 int_x86_aesni_aesdeclast>;
7183 // Perform the AES InvMixColumn Transformation
7184 let Predicates = [HasAVX, HasAES] in {
7185 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7187 "vaesimc\t{$src1, $dst|$dst, $src1}",
7189 (int_x86_aesni_aesimc VR128:$src1))]>,
7191 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7192 (ins i128mem:$src1),
7193 "vaesimc\t{$src1, $dst|$dst, $src1}",
7194 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7197 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7199 "aesimc\t{$src1, $dst|$dst, $src1}",
7201 (int_x86_aesni_aesimc VR128:$src1))]>,
7203 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7204 (ins i128mem:$src1),
7205 "aesimc\t{$src1, $dst|$dst, $src1}",
7206 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7209 // AES Round Key Generation Assist
7210 let Predicates = [HasAVX, HasAES] in {
7211 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7212 (ins VR128:$src1, i8imm:$src2),
7213 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7215 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7217 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7218 (ins i128mem:$src1, i8imm:$src2),
7219 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7221 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7224 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7225 (ins VR128:$src1, i8imm:$src2),
7226 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7228 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7230 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7231 (ins i128mem:$src1, i8imm:$src2),
7232 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7234 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7237 //===----------------------------------------------------------------------===//
7238 // PCLMUL Instructions
7239 //===----------------------------------------------------------------------===//
7241 // AVX carry-less Multiplication instructions
7242 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7243 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7244 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7246 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7248 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7249 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7250 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7251 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7252 (memopv2i64 addr:$src2), imm:$src3))]>;
7254 // Carry-less Multiplication instructions
7255 let Constraints = "$src1 = $dst" in {
7256 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7257 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7258 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7260 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7262 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7263 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7264 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7265 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7266 (memopv2i64 addr:$src2), imm:$src3))]>;
7267 } // Constraints = "$src1 = $dst"
7270 multiclass pclmul_alias<string asm, int immop> {
7271 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7272 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7274 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7275 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7277 def : InstAlias<!strconcat("vpclmul", asm,
7278 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7279 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7281 def : InstAlias<!strconcat("vpclmul", asm,
7282 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7283 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7285 defm : pclmul_alias<"hqhq", 0x11>;
7286 defm : pclmul_alias<"hqlq", 0x01>;
7287 defm : pclmul_alias<"lqhq", 0x10>;
7288 defm : pclmul_alias<"lqlq", 0x00>;
7290 //===----------------------------------------------------------------------===//
7291 // SSE4A Instructions
7292 //===----------------------------------------------------------------------===//
7294 let Predicates = [HasSSE4A] in {
7296 let Constraints = "$src = $dst" in {
7297 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7298 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7299 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7300 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7301 imm:$idx))]>, TB, OpSize;
7302 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7303 (ins VR128:$src, VR128:$mask),
7304 "extrq\t{$mask, $src|$src, $mask}",
7305 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7306 VR128:$mask))]>, TB, OpSize;
7308 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7309 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7310 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7311 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7312 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7313 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7314 (ins VR128:$src, VR128:$mask),
7315 "insertq\t{$mask, $src|$src, $mask}",
7316 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7317 VR128:$mask))]>, XD;
7320 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7321 "movntss\t{$src, $dst|$dst, $src}",
7322 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7324 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7325 "movntsd\t{$src, $dst|$dst, $src}",
7326 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7329 //===----------------------------------------------------------------------===//
7331 //===----------------------------------------------------------------------===//
7333 //===----------------------------------------------------------------------===//
7334 // VBROADCAST - Load from memory and broadcast to all elements of the
7335 // destination operand
7337 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7338 X86MemOperand x86memop, Intrinsic Int> :
7339 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7340 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7341 [(set RC:$dst, (Int addr:$src))]>, VEX;
7343 // AVX2 adds register forms
7344 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7346 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7347 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7348 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7350 let ExeDomain = SSEPackedSingle in {
7351 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7352 int_x86_avx_vbroadcast_ss>;
7353 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7354 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7356 let ExeDomain = SSEPackedDouble in
7357 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7358 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7359 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7360 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7362 let ExeDomain = SSEPackedSingle in {
7363 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7364 int_x86_avx2_vbroadcast_ss_ps>;
7365 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7366 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7368 let ExeDomain = SSEPackedDouble in
7369 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7370 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7372 let Predicates = [HasAVX2] in
7373 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7374 int_x86_avx2_vbroadcasti128>, VEX_L;
7376 let Predicates = [HasAVX] in
7377 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7378 (VBROADCASTF128 addr:$src)>;
7381 //===----------------------------------------------------------------------===//
7382 // VINSERTF128 - Insert packed floating-point values
7384 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7385 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7386 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7387 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7390 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7391 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7392 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7396 let Predicates = [HasAVX] in {
7397 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7399 (VINSERTF128rr VR256:$src1, VR128:$src2,
7400 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7401 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7403 (VINSERTF128rr VR256:$src1, VR128:$src2,
7404 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7406 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7408 (VINSERTF128rm VR256:$src1, addr:$src2,
7409 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7410 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7412 (VINSERTF128rm VR256:$src1, addr:$src2,
7413 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7416 let Predicates = [HasAVX1Only] in {
7417 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7419 (VINSERTF128rr VR256:$src1, VR128:$src2,
7420 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7421 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7423 (VINSERTF128rr VR256:$src1, VR128:$src2,
7424 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7425 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7427 (VINSERTF128rr VR256:$src1, VR128:$src2,
7428 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7429 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7431 (VINSERTF128rr VR256:$src1, VR128:$src2,
7432 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7434 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7436 (VINSERTF128rm VR256:$src1, addr:$src2,
7437 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7438 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7439 (bc_v4i32 (memopv2i64 addr:$src2)),
7441 (VINSERTF128rm VR256:$src1, addr:$src2,
7442 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7443 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7444 (bc_v16i8 (memopv2i64 addr:$src2)),
7446 (VINSERTF128rm VR256:$src1, addr:$src2,
7447 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7448 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7449 (bc_v8i16 (memopv2i64 addr:$src2)),
7451 (VINSERTF128rm VR256:$src1, addr:$src2,
7452 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7455 //===----------------------------------------------------------------------===//
7456 // VEXTRACTF128 - Extract packed floating-point values
7458 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7459 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7460 (ins VR256:$src1, i8imm:$src2),
7461 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7464 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7465 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7466 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7471 let Predicates = [HasAVX] in {
7472 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7473 (v4f32 (VEXTRACTF128rr
7474 (v8f32 VR256:$src1),
7475 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7476 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7477 (v2f64 (VEXTRACTF128rr
7478 (v4f64 VR256:$src1),
7479 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7481 def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
7482 (iPTR imm))), addr:$dst),
7483 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7484 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7485 def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
7486 (iPTR imm))), addr:$dst),
7487 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7488 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7491 let Predicates = [HasAVX1Only] in {
7492 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7493 (v2i64 (VEXTRACTF128rr
7494 (v4i64 VR256:$src1),
7495 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7496 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7497 (v4i32 (VEXTRACTF128rr
7498 (v8i32 VR256:$src1),
7499 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7500 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7501 (v8i16 (VEXTRACTF128rr
7502 (v16i16 VR256:$src1),
7503 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7504 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7505 (v16i8 (VEXTRACTF128rr
7506 (v32i8 VR256:$src1),
7507 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7509 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7510 (iPTR imm))), addr:$dst),
7511 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7512 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7513 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7514 (iPTR imm))), addr:$dst),
7515 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7516 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7517 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7518 (iPTR imm))), addr:$dst),
7519 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7520 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7521 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7522 (iPTR imm))), addr:$dst),
7523 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7524 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7527 //===----------------------------------------------------------------------===//
7528 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7530 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7531 Intrinsic IntLd, Intrinsic IntLd256,
7532 Intrinsic IntSt, Intrinsic IntSt256> {
7533 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7534 (ins VR128:$src1, f128mem:$src2),
7535 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7536 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7538 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7539 (ins VR256:$src1, f256mem:$src2),
7540 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7541 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7543 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7544 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7545 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7546 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7547 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7548 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7549 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7550 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7553 let ExeDomain = SSEPackedSingle in
7554 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7555 int_x86_avx_maskload_ps,
7556 int_x86_avx_maskload_ps_256,
7557 int_x86_avx_maskstore_ps,
7558 int_x86_avx_maskstore_ps_256>;
7559 let ExeDomain = SSEPackedDouble in
7560 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7561 int_x86_avx_maskload_pd,
7562 int_x86_avx_maskload_pd_256,
7563 int_x86_avx_maskstore_pd,
7564 int_x86_avx_maskstore_pd_256>;
7566 //===----------------------------------------------------------------------===//
7567 // VPERMIL - Permute Single and Double Floating-Point Values
7569 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7570 RegisterClass RC, X86MemOperand x86memop_f,
7571 X86MemOperand x86memop_i, PatFrag i_frag,
7572 Intrinsic IntVar, ValueType vt> {
7573 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7574 (ins RC:$src1, RC:$src2),
7575 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7576 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7577 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7578 (ins RC:$src1, x86memop_i:$src2),
7579 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7580 [(set RC:$dst, (IntVar RC:$src1,
7581 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7583 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7584 (ins RC:$src1, i8imm:$src2),
7585 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7586 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7587 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7588 (ins x86memop_f:$src1, i8imm:$src2),
7589 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7591 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7594 let ExeDomain = SSEPackedSingle in {
7595 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7596 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7597 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7598 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7600 let ExeDomain = SSEPackedDouble in {
7601 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7602 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7603 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7604 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7607 let Predicates = [HasAVX] in {
7608 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7609 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7610 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7611 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7612 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7614 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7615 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7616 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7618 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7619 (VPERMILPDri VR128:$src1, imm:$imm)>;
7620 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7621 (VPERMILPDmi addr:$src1, imm:$imm)>;
7624 //===----------------------------------------------------------------------===//
7625 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7627 let ExeDomain = SSEPackedSingle in {
7628 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7629 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7630 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7631 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7632 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7633 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7634 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7635 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7636 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7637 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7640 let Predicates = [HasAVX] in {
7641 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7642 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7643 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7644 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7645 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7648 let Predicates = [HasAVX1Only] in {
7649 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7650 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7651 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7652 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7653 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7654 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7655 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7656 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7658 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7659 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7660 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7661 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7662 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7663 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7664 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7665 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7666 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7667 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7668 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7669 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7672 //===----------------------------------------------------------------------===//
7673 // VZERO - Zero YMM registers
7675 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7676 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7677 // Zero All YMM registers
7678 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7679 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7681 // Zero Upper bits of YMM registers
7682 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7683 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7686 //===----------------------------------------------------------------------===//
7687 // Half precision conversion instructions
7688 //===----------------------------------------------------------------------===//
7689 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7690 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7691 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7692 [(set RC:$dst, (Int VR128:$src))]>,
7694 let neverHasSideEffects = 1, mayLoad = 1 in
7695 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7696 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7699 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7700 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7701 (ins RC:$src1, i32i8imm:$src2),
7702 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7703 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7705 let neverHasSideEffects = 1, mayStore = 1 in
7706 def mr : Ii8<0x1D, MRMDestMem, (outs),
7707 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7708 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7712 let Predicates = [HasAVX, HasF16C] in {
7713 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7714 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7715 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7716 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7719 //===----------------------------------------------------------------------===//
7720 // AVX2 Instructions
7721 //===----------------------------------------------------------------------===//
7723 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7724 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7725 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7726 X86MemOperand x86memop> {
7727 let isCommutable = 1 in
7728 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7729 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7730 !strconcat(OpcodeStr,
7731 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7732 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7734 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7735 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7736 !strconcat(OpcodeStr,
7737 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7740 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7744 let isCommutable = 0 in {
7745 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7746 VR128, memopv2i64, i128mem>;
7747 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7748 VR256, memopv4i64, i256mem>, VEX_L;
7751 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7753 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7754 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7756 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7758 //===----------------------------------------------------------------------===//
7759 // VPBROADCAST - Load from memory and broadcast to all elements of the
7760 // destination operand
7762 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7763 X86MemOperand x86memop, PatFrag ld_frag,
7764 Intrinsic Int128, Intrinsic Int256> {
7765 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7766 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7767 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7768 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7769 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7771 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7772 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7773 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7774 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7775 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7776 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7778 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7782 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7783 int_x86_avx2_pbroadcastb_128,
7784 int_x86_avx2_pbroadcastb_256>;
7785 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7786 int_x86_avx2_pbroadcastw_128,
7787 int_x86_avx2_pbroadcastw_256>;
7788 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7789 int_x86_avx2_pbroadcastd_128,
7790 int_x86_avx2_pbroadcastd_256>;
7791 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7792 int_x86_avx2_pbroadcastq_128,
7793 int_x86_avx2_pbroadcastq_256>;
7795 let Predicates = [HasAVX2] in {
7796 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7797 (VPBROADCASTBrm addr:$src)>;
7798 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7799 (VPBROADCASTBYrm addr:$src)>;
7800 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7801 (VPBROADCASTWrm addr:$src)>;
7802 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7803 (VPBROADCASTWYrm addr:$src)>;
7804 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7805 (VPBROADCASTDrm addr:$src)>;
7806 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7807 (VPBROADCASTDYrm addr:$src)>;
7808 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7809 (VPBROADCASTQrm addr:$src)>;
7810 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7811 (VPBROADCASTQYrm addr:$src)>;
7813 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7814 (VPBROADCASTBrr VR128:$src)>;
7815 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7816 (VPBROADCASTBYrr VR128:$src)>;
7817 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7818 (VPBROADCASTWrr VR128:$src)>;
7819 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7820 (VPBROADCASTWYrr VR128:$src)>;
7821 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7822 (VPBROADCASTDrr VR128:$src)>;
7823 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7824 (VPBROADCASTDYrr VR128:$src)>;
7825 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7826 (VPBROADCASTQrr VR128:$src)>;
7827 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7828 (VPBROADCASTQYrr VR128:$src)>;
7829 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7830 (VBROADCASTSSrr VR128:$src)>;
7831 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7832 (VBROADCASTSSYrr VR128:$src)>;
7833 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7834 (VPBROADCASTQrr VR128:$src)>;
7835 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7836 (VBROADCASTSDYrr VR128:$src)>;
7838 // Provide fallback in case the load node that is used in the patterns above
7839 // is used by additional users, which prevents the pattern selection.
7840 let AddedComplexity = 20 in {
7841 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7842 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7843 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7844 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7845 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7846 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7848 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7849 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7850 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7851 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7852 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7853 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7857 // AVX1 broadcast patterns
7858 let Predicates = [HasAVX1Only] in {
7859 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7860 (VBROADCASTSSYrm addr:$src)>;
7861 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7862 (VBROADCASTSDYrm addr:$src)>;
7863 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7864 (VBROADCASTSSrm addr:$src)>;
7867 let Predicates = [HasAVX] in {
7868 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7869 (VBROADCASTSSYrm addr:$src)>;
7870 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7871 (VBROADCASTSDYrm addr:$src)>;
7872 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7873 (VBROADCASTSSrm addr:$src)>;
7875 // Provide fallback in case the load node that is used in the patterns above
7876 // is used by additional users, which prevents the pattern selection.
7877 let AddedComplexity = 20 in {
7878 // 128bit broadcasts:
7879 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7880 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7881 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7882 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7883 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7884 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7885 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7886 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7887 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7888 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7890 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7891 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7892 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7893 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7894 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7895 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7896 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7897 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7898 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7899 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7903 //===----------------------------------------------------------------------===//
7904 // VPERM - Permute instructions
7907 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7909 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7910 (ins VR256:$src1, VR256:$src2),
7911 !strconcat(OpcodeStr,
7912 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7914 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
7916 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7917 (ins VR256:$src1, i256mem:$src2),
7918 !strconcat(OpcodeStr,
7919 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7921 (OpVT (X86VPermv VR256:$src1,
7922 (bitconvert (mem_frag addr:$src2)))))]>,
7926 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7927 let ExeDomain = SSEPackedSingle in
7928 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7930 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7932 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7933 (ins VR256:$src1, i8imm:$src2),
7934 !strconcat(OpcodeStr,
7935 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7937 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
7939 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7940 (ins i256mem:$src1, i8imm:$src2),
7941 !strconcat(OpcodeStr,
7942 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7944 (OpVT (X86VPermi (mem_frag addr:$src1),
7945 (i8 imm:$src2))))]>, VEX, VEX_L;
7948 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7949 let ExeDomain = SSEPackedDouble in
7950 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7952 //===----------------------------------------------------------------------===//
7953 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7955 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7956 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7957 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7958 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7959 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7960 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7961 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7962 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7963 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7964 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7966 let Predicates = [HasAVX2] in {
7967 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7968 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7969 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7970 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7971 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7972 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7974 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7976 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7977 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7978 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7979 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7980 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7982 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7986 //===----------------------------------------------------------------------===//
7987 // VINSERTI128 - Insert packed integer values
7989 let neverHasSideEffects = 1 in {
7990 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7991 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7992 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7995 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7996 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7997 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
8001 let Predicates = [HasAVX2] in {
8002 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
8004 (VINSERTI128rr VR256:$src1, VR128:$src2,
8005 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8006 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
8008 (VINSERTI128rr VR256:$src1, VR128:$src2,
8009 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8010 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
8012 (VINSERTI128rr VR256:$src1, VR128:$src2,
8013 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8014 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
8016 (VINSERTI128rr VR256:$src1, VR128:$src2,
8017 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8019 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
8021 (VINSERTI128rm VR256:$src1, addr:$src2,
8022 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8023 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
8024 (bc_v4i32 (memopv2i64 addr:$src2)),
8026 (VINSERTI128rm VR256:$src1, addr:$src2,
8027 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8028 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
8029 (bc_v16i8 (memopv2i64 addr:$src2)),
8031 (VINSERTI128rm VR256:$src1, addr:$src2,
8032 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8033 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
8034 (bc_v8i16 (memopv2i64 addr:$src2)),
8036 (VINSERTI128rm VR256:$src1, addr:$src2,
8037 (INSERT_get_vinsertf128_imm VR256:$ins))>;
8040 //===----------------------------------------------------------------------===//
8041 // VEXTRACTI128 - Extract packed integer values
8043 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
8044 (ins VR256:$src1, i8imm:$src2),
8045 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
8047 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
8049 let neverHasSideEffects = 1, mayStore = 1 in
8050 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
8051 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
8052 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
8055 let Predicates = [HasAVX2] in {
8056 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8057 (v2i64 (VEXTRACTI128rr
8058 (v4i64 VR256:$src1),
8059 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8060 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8061 (v4i32 (VEXTRACTI128rr
8062 (v8i32 VR256:$src1),
8063 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8064 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8065 (v8i16 (VEXTRACTI128rr
8066 (v16i16 VR256:$src1),
8067 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8068 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
8069 (v16i8 (VEXTRACTI128rr
8070 (v32i8 VR256:$src1),
8071 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
8073 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
8074 (iPTR imm))), addr:$dst),
8075 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8076 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8077 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
8078 (iPTR imm))), addr:$dst),
8079 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8080 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8081 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
8082 (iPTR imm))), addr:$dst),
8083 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8084 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8085 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
8086 (iPTR imm))), addr:$dst),
8087 (VEXTRACTI128mr addr:$dst, VR256:$src1,
8088 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
8091 //===----------------------------------------------------------------------===//
8092 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
8094 multiclass avx2_pmovmask<string OpcodeStr,
8095 Intrinsic IntLd128, Intrinsic IntLd256,
8096 Intrinsic IntSt128, Intrinsic IntSt256> {
8097 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
8098 (ins VR128:$src1, i128mem:$src2),
8099 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8100 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
8101 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
8102 (ins VR256:$src1, i256mem:$src2),
8103 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8104 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
8106 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8107 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8108 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8109 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8110 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8111 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8112 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8113 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8116 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8117 int_x86_avx2_maskload_d,
8118 int_x86_avx2_maskload_d_256,
8119 int_x86_avx2_maskstore_d,
8120 int_x86_avx2_maskstore_d_256>;
8121 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8122 int_x86_avx2_maskload_q,
8123 int_x86_avx2_maskload_q_256,
8124 int_x86_avx2_maskstore_q,
8125 int_x86_avx2_maskstore_q_256>, VEX_W;
8128 //===----------------------------------------------------------------------===//
8129 // Variable Bit Shifts
8131 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8132 ValueType vt128, ValueType vt256> {
8133 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8134 (ins VR128:$src1, VR128:$src2),
8135 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8137 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8139 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8140 (ins VR128:$src1, i128mem:$src2),
8141 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8143 (vt128 (OpNode VR128:$src1,
8144 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8146 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8147 (ins VR256:$src1, VR256:$src2),
8148 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8150 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8152 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8153 (ins VR256:$src1, i256mem:$src2),
8154 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8156 (vt256 (OpNode VR256:$src1,
8157 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8161 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8162 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8163 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8164 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8165 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8167 //===----------------------------------------------------------------------===//
8168 // VGATHER - GATHER Operations
8169 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8170 X86MemOperand memop128, X86MemOperand memop256> {
8171 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8172 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8173 !strconcat(OpcodeStr,
8174 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8176 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8177 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8178 !strconcat(OpcodeStr,
8179 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8180 []>, VEX_4VOp3, VEX_L;
8183 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8184 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8185 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8186 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8187 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8188 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8189 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8190 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8191 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;