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:$src))],
1109 //===----------------------------------------------------------------------===//
1110 // SSE 1 & 2 - Move Low packed FP Instructions
1111 //===----------------------------------------------------------------------===//
1113 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
1114 SDNode psnode, SDNode pdnode, string base_opc,
1115 string asm_opr, InstrItinClass itin> {
1116 def PSrm : PI<opc, MRMSrcMem,
1117 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
1118 !strconcat(base_opc, "s", asm_opr),
1121 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1122 itin, SSEPackedSingle>, TB;
1124 def PDrm : PI<opc, MRMSrcMem,
1125 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1126 !strconcat(base_opc, "d", asm_opr),
1127 [(set RC:$dst, (v2f64 (pdnode RC:$src1,
1128 (scalar_to_vector (loadf64 addr:$src2)))))],
1129 itin, SSEPackedDouble>, TB, OpSize;
1132 let AddedComplexity = 20 in {
1133 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1134 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1135 IIC_SSE_MOV_LH>, VEX_4V;
1137 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1138 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, X86Movlps, X86Movlpd, "movlp",
1139 "\t{$src2, $dst|$dst, $src2}",
1143 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1144 "movlps\t{$src, $dst|$dst, $src}",
1145 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1146 (iPTR 0))), addr:$dst)],
1147 IIC_SSE_MOV_LH>, VEX;
1148 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1149 "movlpd\t{$src, $dst|$dst, $src}",
1150 [(store (f64 (vector_extract (v2f64 VR128:$src),
1151 (iPTR 0))), addr:$dst)],
1152 IIC_SSE_MOV_LH>, VEX;
1153 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1154 "movlps\t{$src, $dst|$dst, $src}",
1155 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1156 (iPTR 0))), addr:$dst)],
1158 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1159 "movlpd\t{$src, $dst|$dst, $src}",
1160 [(store (f64 (vector_extract (v2f64 VR128:$src),
1161 (iPTR 0))), addr:$dst)],
1164 let Predicates = [HasAVX] in {
1165 // Shuffle with VMOVLPS
1166 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1167 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1168 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1169 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1171 // Shuffle with VMOVLPD
1172 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1173 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1174 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1175 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1178 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1180 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1181 def : Pat<(store (v4i32 (X86Movlps
1182 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1183 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1184 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1186 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1187 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1189 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1192 let Predicates = [UseSSE1] in {
1193 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1194 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1195 (iPTR 0))), addr:$src1),
1196 (MOVLPSmr addr:$src1, VR128:$src2)>;
1198 // Shuffle with MOVLPS
1199 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1200 (MOVLPSrm VR128:$src1, addr:$src2)>;
1201 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1202 (MOVLPSrm VR128:$src1, addr:$src2)>;
1203 def : Pat<(X86Movlps VR128:$src1,
1204 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1205 (MOVLPSrm VR128:$src1, addr:$src2)>;
1208 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1210 (MOVLPSmr addr:$src1, VR128:$src2)>;
1211 def : Pat<(store (v4i32 (X86Movlps
1212 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1214 (MOVLPSmr addr:$src1, VR128:$src2)>;
1217 let Predicates = [UseSSE2] in {
1218 // Shuffle with MOVLPD
1219 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1220 (MOVLPDrm VR128:$src1, addr:$src2)>;
1221 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1222 (MOVLPDrm VR128:$src1, addr:$src2)>;
1225 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1227 (MOVLPDmr addr:$src1, VR128:$src2)>;
1228 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1230 (MOVLPDmr addr:$src1, VR128:$src2)>;
1233 //===----------------------------------------------------------------------===//
1234 // SSE 1 & 2 - Move Hi packed FP Instructions
1235 //===----------------------------------------------------------------------===//
1237 let AddedComplexity = 20 in {
1238 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1239 "\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1240 IIC_SSE_MOV_LH>, VEX_4V;
1242 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1243 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, X86Movlhps, X86Movlhpd, "movhp",
1244 "\t{$src2, $dst|$dst, $src2}",
1248 // v2f64 extract element 1 is always custom lowered to unpack high to low
1249 // and extract element 0 so the non-store version isn't too horrible.
1250 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1251 "movhps\t{$src, $dst|$dst, $src}",
1252 [(store (f64 (vector_extract
1253 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1254 (bc_v2f64 (v4f32 VR128:$src))),
1255 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1256 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1257 "movhpd\t{$src, $dst|$dst, $src}",
1258 [(store (f64 (vector_extract
1259 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1260 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>, VEX;
1261 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1262 "movhps\t{$src, $dst|$dst, $src}",
1263 [(store (f64 (vector_extract
1264 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1265 (bc_v2f64 (v4f32 VR128:$src))),
1266 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1267 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1268 "movhpd\t{$src, $dst|$dst, $src}",
1269 [(store (f64 (vector_extract
1270 (v2f64 (X86Unpckh VR128:$src, VR128:$src)),
1271 (iPTR 0))), addr:$dst)], IIC_SSE_MOV_LH>;
1273 let Predicates = [HasAVX] in {
1275 def : Pat<(X86Movlhps VR128:$src1,
1276 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1277 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1278 def : Pat<(X86Movlhps VR128:$src1,
1279 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1280 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1282 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1283 // is during lowering, where it's not possible to recognize the load fold
1284 // cause it has two uses through a bitcast. One use disappears at isel time
1285 // and the fold opportunity reappears.
1286 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1287 (scalar_to_vector (loadf64 addr:$src2)))),
1288 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1291 let Predicates = [UseSSE1] in {
1293 def : Pat<(X86Movlhps VR128:$src1,
1294 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1295 (MOVHPSrm VR128:$src1, addr:$src2)>;
1296 def : Pat<(X86Movlhps VR128:$src1,
1297 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1298 (MOVHPSrm VR128:$src1, addr:$src2)>;
1301 let Predicates = [UseSSE2] in {
1302 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1303 // is during lowering, where it's not possible to recognize the load fold
1304 // cause it has two uses through a bitcast. One use disappears at isel time
1305 // and the fold opportunity reappears.
1306 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1307 (scalar_to_vector (loadf64 addr:$src2)))),
1308 (MOVHPDrm VR128:$src1, addr:$src2)>;
1311 //===----------------------------------------------------------------------===//
1312 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1313 //===----------------------------------------------------------------------===//
1315 let AddedComplexity = 20 in {
1316 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1317 (ins VR128:$src1, VR128:$src2),
1318 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1320 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1323 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1324 (ins VR128:$src1, VR128:$src2),
1325 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1327 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1331 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1332 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1333 (ins VR128:$src1, VR128:$src2),
1334 "movlhps\t{$src2, $dst|$dst, $src2}",
1336 (v4f32 (X86Movlhps VR128:$src1, VR128:$src2)))],
1338 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1339 (ins VR128:$src1, VR128:$src2),
1340 "movhlps\t{$src2, $dst|$dst, $src2}",
1342 (v4f32 (X86Movhlps VR128:$src1, VR128:$src2)))],
1346 let Predicates = [HasAVX] in {
1348 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1349 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1350 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1351 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1354 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1355 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1358 let Predicates = [UseSSE1] in {
1360 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1361 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1362 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1363 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1366 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1367 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1370 //===----------------------------------------------------------------------===//
1371 // SSE 1 & 2 - Conversion Instructions
1372 //===----------------------------------------------------------------------===//
1374 def SSE_CVT_PD : OpndItins<
1375 IIC_SSE_CVT_PD_RR, IIC_SSE_CVT_PD_RM
1378 def SSE_CVT_PS : OpndItins<
1379 IIC_SSE_CVT_PS_RR, IIC_SSE_CVT_PS_RM
1382 def SSE_CVT_Scalar : OpndItins<
1383 IIC_SSE_CVT_Scalar_RR, IIC_SSE_CVT_Scalar_RM
1386 def SSE_CVT_SS2SI_32 : OpndItins<
1387 IIC_SSE_CVT_SS2SI32_RR, IIC_SSE_CVT_SS2SI32_RM
1390 def SSE_CVT_SS2SI_64 : OpndItins<
1391 IIC_SSE_CVT_SS2SI64_RR, IIC_SSE_CVT_SS2SI64_RM
1394 def SSE_CVT_SD2SI : OpndItins<
1395 IIC_SSE_CVT_SD2SI_RR, IIC_SSE_CVT_SD2SI_RM
1398 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1399 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1400 string asm, OpndItins itins> {
1401 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1402 [(set DstRC:$dst, (OpNode SrcRC:$src))],
1404 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1405 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))],
1409 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1410 X86MemOperand x86memop, string asm, Domain d,
1412 let neverHasSideEffects = 1 in {
1413 def rr : I<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1416 def rm : I<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1421 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1422 X86MemOperand x86memop, string asm> {
1423 let neverHasSideEffects = 1 in {
1424 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1425 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1427 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1428 (ins DstRC:$src1, x86memop:$src),
1429 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1430 } // neverHasSideEffects = 1
1433 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1434 "cvttss2si\t{$src, $dst|$dst, $src}",
1437 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1438 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1440 XS, VEX, VEX_W, VEX_LIG;
1441 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1442 "cvttsd2si\t{$src, $dst|$dst, $src}",
1445 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1446 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1448 XD, VEX, VEX_W, VEX_LIG;
1450 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1451 // register, but the same isn't true when only using memory operands,
1452 // provide other assembly "l" and "q" forms to address this explicitly
1453 // where appropriate to do so.
1454 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">,
1455 XS, VEX_4V, VEX_LIG;
1456 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">,
1457 XS, VEX_4V, VEX_W, VEX_LIG;
1458 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">,
1459 XD, VEX_4V, VEX_LIG;
1460 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">,
1461 XD, VEX_4V, VEX_W, VEX_LIG;
1463 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1464 (VCVTSI2SDrr FR64:$dst, FR64:$src1, GR32:$src)>;
1465 def : InstAlias<"vcvtsi2sd{l}\t{$src, $src1, $dst|$dst, $src1, $src}",
1466 (VCVTSI2SDrm FR64:$dst, FR64:$src1, i32mem:$src)>;
1468 let Predicates = [HasAVX] in {
1469 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1470 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1471 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1472 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1473 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1474 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1475 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1476 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1478 def : Pat<(f32 (sint_to_fp GR32:$src)),
1479 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1480 def : Pat<(f32 (sint_to_fp GR64:$src)),
1481 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1482 def : Pat<(f64 (sint_to_fp GR32:$src)),
1483 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1484 def : Pat<(f64 (sint_to_fp GR64:$src)),
1485 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1488 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1489 "cvttss2si\t{$src, $dst|$dst, $src}",
1490 SSE_CVT_SS2SI_32>, XS;
1491 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1492 "cvttss2si{q}\t{$src, $dst|$dst, $src}",
1493 SSE_CVT_SS2SI_64>, XS, REX_W;
1494 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1495 "cvttsd2si\t{$src, $dst|$dst, $src}",
1497 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1498 "cvttsd2si{q}\t{$src, $dst|$dst, $src}",
1499 SSE_CVT_SD2SI>, XD, REX_W;
1500 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1501 "cvtsi2ss\t{$src, $dst|$dst, $src}",
1502 SSE_CVT_Scalar>, XS;
1503 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1504 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}",
1505 SSE_CVT_Scalar>, XS, REX_W;
1506 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1507 "cvtsi2sd\t{$src, $dst|$dst, $src}",
1508 SSE_CVT_Scalar>, XD;
1509 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1510 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}",
1511 SSE_CVT_Scalar>, XD, REX_W;
1513 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1514 // and/or XMM operand(s).
1516 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1517 Intrinsic Int, Operand memop, ComplexPattern mem_cpat,
1518 string asm, OpndItins itins> {
1519 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1520 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1521 [(set DstRC:$dst, (Int SrcRC:$src))], itins.rr>;
1522 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins memop:$src),
1523 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1524 [(set DstRC:$dst, (Int mem_cpat:$src))], itins.rm>;
1527 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1528 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1529 PatFrag ld_frag, string asm, OpndItins itins,
1531 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1533 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1534 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1535 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))],
1537 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1538 (ins DstRC:$src1, x86memop:$src2),
1540 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1541 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1542 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))],
1546 defm VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32,
1547 int_x86_sse2_cvtsd2si, sdmem, sse_load_f64, "cvtsd2si{l}",
1548 SSE_CVT_SD2SI>, XD, VEX, VEX_LIG;
1549 defm VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1550 int_x86_sse2_cvtsd2si64, sdmem, sse_load_f64, "cvtsd2si{q}",
1551 SSE_CVT_SD2SI>, XD, VEX, VEX_W, VEX_LIG;
1553 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1554 sdmem, sse_load_f64, "cvtsd2si{l}", SSE_CVT_SD2SI>, XD;
1555 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1556 sdmem, sse_load_f64, "cvtsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1559 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1560 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss",
1561 SSE_CVT_Scalar, 0>, XS, VEX_4V;
1562 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1563 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss{q}",
1564 SSE_CVT_Scalar, 0>, XS, VEX_4V,
1566 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1567 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd",
1568 SSE_CVT_Scalar, 0>, XD, VEX_4V;
1569 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1570 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd{q}",
1571 SSE_CVT_Scalar, 0>, XD,
1574 let Constraints = "$src1 = $dst" in {
1575 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1576 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1577 "cvtsi2ss", SSE_CVT_Scalar>, XS;
1578 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1579 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1580 "cvtsi2ss{q}", SSE_CVT_Scalar>, XS, REX_W;
1581 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1582 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1583 "cvtsi2sd", SSE_CVT_Scalar>, XD;
1584 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1585 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1586 "cvtsi2sd{q}", SSE_CVT_Scalar>, XD, REX_W;
1591 // Aliases for intrinsics
1592 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1593 ssmem, sse_load_f32, "cvttss2si",
1594 SSE_CVT_SS2SI_32>, XS, VEX;
1595 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1596 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1597 "cvttss2si{q}", SSE_CVT_SS2SI_64>,
1599 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1600 sdmem, sse_load_f64, "cvttsd2si",
1601 SSE_CVT_SD2SI>, XD, VEX;
1602 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1603 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1604 "cvttsd2si{q}", SSE_CVT_SD2SI>,
1606 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1607 ssmem, sse_load_f32, "cvttss2si",
1608 SSE_CVT_SS2SI_32>, XS;
1609 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1610 int_x86_sse_cvttss2si64, ssmem, sse_load_f32,
1611 "cvttss2si{q}", SSE_CVT_SS2SI_64>, XS, REX_W;
1612 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1613 sdmem, sse_load_f64, "cvttsd2si",
1615 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1616 int_x86_sse2_cvttsd2si64, sdmem, sse_load_f64,
1617 "cvttsd2si{q}", SSE_CVT_SD2SI>, XD, REX_W;
1619 defm VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1620 ssmem, sse_load_f32, "cvtss2si{l}",
1621 SSE_CVT_SS2SI_32>, XS, VEX, VEX_LIG;
1622 defm VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1623 ssmem, sse_load_f32, "cvtss2si{q}",
1624 SSE_CVT_SS2SI_64>, XS, VEX, VEX_W, VEX_LIG;
1626 defm CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
1627 ssmem, sse_load_f32, "cvtss2si{l}",
1628 SSE_CVT_SS2SI_32>, XS;
1629 defm CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
1630 ssmem, sse_load_f32, "cvtss2si{q}",
1631 SSE_CVT_SS2SI_64>, XS, REX_W;
1633 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1634 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1635 SSEPackedSingle, SSE_CVT_PS>,
1636 TB, VEX, Requires<[HasAVX]>;
1637 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, i256mem,
1638 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1639 SSEPackedSingle, SSE_CVT_PS>,
1640 TB, VEX, VEX_L, Requires<[HasAVX]>;
1642 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, i128mem,
1643 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1644 SSEPackedSingle, SSE_CVT_PS>,
1645 TB, Requires<[UseSSE2]>;
1649 // Convert scalar double to scalar single
1650 let neverHasSideEffects = 1 in {
1651 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1652 (ins FR64:$src1, FR64:$src2),
1653 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", [],
1654 IIC_SSE_CVT_Scalar_RR>, VEX_4V, VEX_LIG;
1656 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1657 (ins FR64:$src1, f64mem:$src2),
1658 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1659 [], IIC_SSE_CVT_Scalar_RM>,
1660 XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1663 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1666 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1667 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1668 [(set FR32:$dst, (fround FR64:$src))],
1669 IIC_SSE_CVT_Scalar_RR>;
1670 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1671 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1672 [(set FR32:$dst, (fround (loadf64 addr:$src)))],
1673 IIC_SSE_CVT_Scalar_RM>,
1675 Requires<[UseSSE2, OptForSize]>;
1677 def Int_VCVTSD2SSrr: I<0x5A, MRMSrcReg,
1678 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1679 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1681 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1682 IIC_SSE_CVT_Scalar_RR>, XD, VEX_4V, Requires<[HasAVX]>;
1683 def Int_VCVTSD2SSrm: I<0x5A, MRMSrcReg,
1684 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1685 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1686 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1687 VR128:$src1, sse_load_f64:$src2))],
1688 IIC_SSE_CVT_Scalar_RM>, XD, VEX_4V, Requires<[HasAVX]>;
1690 let Constraints = "$src1 = $dst" in {
1691 def Int_CVTSD2SSrr: I<0x5A, MRMSrcReg,
1692 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1693 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1695 (int_x86_sse2_cvtsd2ss VR128:$src1, VR128:$src2))],
1696 IIC_SSE_CVT_Scalar_RR>, XD, Requires<[UseSSE2]>;
1697 def Int_CVTSD2SSrm: I<0x5A, MRMSrcReg,
1698 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2),
1699 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1700 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss
1701 VR128:$src1, sse_load_f64:$src2))],
1702 IIC_SSE_CVT_Scalar_RM>, XD, Requires<[UseSSE2]>;
1705 // Convert scalar single to scalar double
1706 // SSE2 instructions with XS prefix
1707 let neverHasSideEffects = 1 in {
1708 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1709 (ins FR32:$src1, FR32:$src2),
1710 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1711 [], IIC_SSE_CVT_Scalar_RR>,
1712 XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1714 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1715 (ins FR32:$src1, f32mem:$src2),
1716 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1717 [], IIC_SSE_CVT_Scalar_RM>,
1718 XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1721 def : Pat<(f64 (fextend FR32:$src)),
1722 (VCVTSS2SDrr FR32:$src, FR32:$src)>, Requires<[HasAVX]>;
1723 def : Pat<(fextend (loadf32 addr:$src)),
1724 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>, Requires<[HasAVX]>;
1726 def : Pat<(extloadf32 addr:$src),
1727 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>,
1728 Requires<[HasAVX, OptForSize]>;
1729 def : Pat<(extloadf32 addr:$src),
1730 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (VMOVSSrm addr:$src))>,
1731 Requires<[HasAVX, OptForSpeed]>;
1733 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1734 "cvtss2sd\t{$src, $dst|$dst, $src}",
1735 [(set FR64:$dst, (fextend FR32:$src))],
1736 IIC_SSE_CVT_Scalar_RR>, XS,
1737 Requires<[UseSSE2]>;
1738 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1739 "cvtss2sd\t{$src, $dst|$dst, $src}",
1740 [(set FR64:$dst, (extloadf32 addr:$src))],
1741 IIC_SSE_CVT_Scalar_RM>, XS,
1742 Requires<[UseSSE2, OptForSize]>;
1744 // extload f32 -> f64. This matches load+fextend because we have a hack in
1745 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1747 // Since these loads aren't folded into the fextend, we have to match it
1749 def : Pat<(fextend (loadf32 addr:$src)),
1750 (CVTSS2SDrm addr:$src)>, Requires<[UseSSE2]>;
1751 def : Pat<(extloadf32 addr:$src),
1752 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[UseSSE2, OptForSpeed]>;
1754 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1755 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1756 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1758 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1759 IIC_SSE_CVT_Scalar_RR>, XS, VEX_4V, Requires<[HasAVX]>;
1760 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1761 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1762 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1764 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1765 IIC_SSE_CVT_Scalar_RM>, XS, VEX_4V, Requires<[HasAVX]>;
1766 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1767 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1768 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1769 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1771 (int_x86_sse2_cvtss2sd VR128:$src1, VR128:$src2))],
1772 IIC_SSE_CVT_Scalar_RR>, XS, Requires<[UseSSE2]>;
1773 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1774 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2),
1775 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1777 (int_x86_sse2_cvtss2sd VR128:$src1, sse_load_f32:$src2))],
1778 IIC_SSE_CVT_Scalar_RM>, XS, Requires<[UseSSE2]>;
1781 // Convert packed single/double fp to doubleword
1782 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1783 "cvtps2dq\t{$src, $dst|$dst, $src}",
1784 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1785 IIC_SSE_CVT_PS_RR>, VEX;
1786 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1787 "cvtps2dq\t{$src, $dst|$dst, $src}",
1789 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1790 IIC_SSE_CVT_PS_RM>, VEX;
1791 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1792 "cvtps2dq\t{$src, $dst|$dst, $src}",
1794 (int_x86_avx_cvt_ps2dq_256 VR256:$src))],
1795 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1796 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1797 "cvtps2dq\t{$src, $dst|$dst, $src}",
1799 (int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)))],
1800 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1801 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1802 "cvtps2dq\t{$src, $dst|$dst, $src}",
1803 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))],
1805 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1806 "cvtps2dq\t{$src, $dst|$dst, $src}",
1808 (int_x86_sse2_cvtps2dq (memopv4f32 addr:$src)))],
1812 // Convert Packed Double FP to Packed DW Integers
1813 let Predicates = [HasAVX] in {
1814 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1815 // register, but the same isn't true when using memory operands instead.
1816 // Provide other assembly rr and rm forms to address this explicitly.
1817 def VCVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1818 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1819 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1823 def : InstAlias<"vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1824 (VCVTPD2DQrr VR128:$dst, VR128:$src)>;
1825 def VCVTPD2DQXrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1826 "vcvtpd2dqx\t{$src, $dst|$dst, $src}",
1828 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))]>, VEX;
1831 def VCVTPD2DQYrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1832 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1834 (int_x86_avx_cvt_pd2dq_256 VR256:$src))]>, VEX, VEX_L;
1835 def VCVTPD2DQYrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1836 "vcvtpd2dq{y}\t{$src, $dst|$dst, $src}",
1838 (int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)))]>,
1840 def : InstAlias<"vcvtpd2dq\t{$src, $dst|$dst, $src}",
1841 (VCVTPD2DQYrr VR128:$dst, VR256:$src)>;
1844 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1845 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1847 (int_x86_sse2_cvtpd2dq (memopv2f64 addr:$src)))],
1849 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1850 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1851 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))],
1854 // Convert with truncation packed single/double fp to doubleword
1855 // SSE2 packed instructions with XS prefix
1856 def VCVTTPS2DQrr : VS2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1857 "cvttps2dq\t{$src, $dst|$dst, $src}",
1859 (int_x86_sse2_cvttps2dq VR128:$src))],
1860 IIC_SSE_CVT_PS_RR>, VEX;
1861 def VCVTTPS2DQrm : VS2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1862 "cvttps2dq\t{$src, $dst|$dst, $src}",
1863 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1864 (memopv4f32 addr:$src)))],
1865 IIC_SSE_CVT_PS_RM>, VEX;
1866 def VCVTTPS2DQYrr : VS2SI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1867 "cvttps2dq\t{$src, $dst|$dst, $src}",
1869 (int_x86_avx_cvtt_ps2dq_256 VR256:$src))],
1870 IIC_SSE_CVT_PS_RR>, VEX, VEX_L;
1871 def VCVTTPS2DQYrm : VS2SI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1872 "cvttps2dq\t{$src, $dst|$dst, $src}",
1873 [(set VR256:$dst, (int_x86_avx_cvtt_ps2dq_256
1874 (memopv8f32 addr:$src)))],
1875 IIC_SSE_CVT_PS_RM>, VEX, VEX_L;
1877 def CVTTPS2DQrr : S2SI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1878 "cvttps2dq\t{$src, $dst|$dst, $src}",
1879 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))],
1881 def CVTTPS2DQrm : S2SI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1882 "cvttps2dq\t{$src, $dst|$dst, $src}",
1884 (int_x86_sse2_cvttps2dq (memopv4f32 addr:$src)))],
1887 let Predicates = [HasAVX] in {
1888 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1889 (VCVTDQ2PSrr VR128:$src)>;
1890 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1891 (VCVTDQ2PSrm addr:$src)>;
1893 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1894 (VCVTDQ2PSrr VR128:$src)>;
1895 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1896 (VCVTDQ2PSrm addr:$src)>;
1898 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1899 (VCVTTPS2DQrr VR128:$src)>;
1900 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1901 (VCVTTPS2DQrm addr:$src)>;
1903 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1904 (VCVTDQ2PSYrr VR256:$src)>;
1905 def : Pat<(v8f32 (sint_to_fp (bc_v8i32 (memopv4i64 addr:$src)))),
1906 (VCVTDQ2PSYrm addr:$src)>;
1908 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1909 (VCVTTPS2DQYrr VR256:$src)>;
1910 def : Pat<(v8i32 (fp_to_sint (memopv8f32 addr:$src))),
1911 (VCVTTPS2DQYrm addr:$src)>;
1914 let Predicates = [UseSSE2] in {
1915 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1916 (CVTDQ2PSrr VR128:$src)>;
1917 def : Pat<(v4f32 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
1918 (CVTDQ2PSrm addr:$src)>;
1920 def : Pat<(int_x86_sse2_cvtdq2ps VR128:$src),
1921 (CVTDQ2PSrr VR128:$src)>;
1922 def : Pat<(int_x86_sse2_cvtdq2ps (bc_v4i32 (memopv2i64 addr:$src))),
1923 (CVTDQ2PSrm addr:$src)>;
1925 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1926 (CVTTPS2DQrr VR128:$src)>;
1927 def : Pat<(v4i32 (fp_to_sint (memopv4f32 addr:$src))),
1928 (CVTTPS2DQrm addr:$src)>;
1931 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1932 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1934 (int_x86_sse2_cvttpd2dq VR128:$src))],
1935 IIC_SSE_CVT_PD_RR>, VEX;
1937 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1938 // register, but the same isn't true when using memory operands instead.
1939 // Provide other assembly rr and rm forms to address this explicitly.
1942 def : InstAlias<"vcvttpd2dqx\t{$src, $dst|$dst, $src}",
1943 (VCVTTPD2DQrr VR128:$dst, VR128:$src)>;
1944 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1945 "cvttpd2dqx\t{$src, $dst|$dst, $src}",
1946 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1947 (memopv2f64 addr:$src)))],
1948 IIC_SSE_CVT_PD_RM>, VEX;
1951 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1952 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1954 (int_x86_avx_cvtt_pd2dq_256 VR256:$src))],
1955 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
1956 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1957 "cvttpd2dq{y}\t{$src, $dst|$dst, $src}",
1959 (int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)))],
1960 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
1961 def : InstAlias<"vcvttpd2dq\t{$src, $dst|$dst, $src}",
1962 (VCVTTPD2DQYrr VR128:$dst, VR256:$src)>;
1964 let Predicates = [HasAVX] in {
1965 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
1966 (VCVTTPD2DQYrr VR256:$src)>;
1967 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
1968 (VCVTTPD2DQYrm addr:$src)>;
1969 } // Predicates = [HasAVX]
1971 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1972 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1973 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))],
1975 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1976 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1977 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1978 (memopv2f64 addr:$src)))],
1981 // Convert packed single to packed double
1982 let Predicates = [HasAVX] in {
1983 // SSE2 instructions without OpSize prefix
1984 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1985 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1986 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
1987 IIC_SSE_CVT_PD_RR>, TB, VEX;
1988 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1989 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1990 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
1991 IIC_SSE_CVT_PD_RM>, TB, VEX;
1992 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1993 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1995 (int_x86_avx_cvt_ps2_pd_256 VR128:$src))],
1996 IIC_SSE_CVT_PD_RR>, TB, VEX, VEX_L;
1997 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1998 "vcvtps2pd\t{$src, $dst|$dst, $src}",
2000 (int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)))],
2001 IIC_SSE_CVT_PD_RM>, TB, VEX, VEX_L;
2004 let Predicates = [UseSSE2] in {
2005 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2006 "cvtps2pd\t{$src, $dst|$dst, $src}",
2007 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))],
2008 IIC_SSE_CVT_PD_RR>, TB;
2009 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2010 "cvtps2pd\t{$src, $dst|$dst, $src}",
2011 [(set VR128:$dst, (v2f64 (extloadv2f32 addr:$src)))],
2012 IIC_SSE_CVT_PD_RM>, TB;
2015 // Convert Packed DW Integers to Packed Double FP
2016 let Predicates = [HasAVX] in {
2017 let neverHasSideEffects = 1, mayLoad = 1 in
2018 def VCVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2019 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2021 def VCVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2022 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2024 (int_x86_sse2_cvtdq2pd VR128:$src))]>, VEX;
2025 def VCVTDQ2PDYrm : S2SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
2026 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2028 (int_x86_avx_cvtdq2_pd_256
2029 (bitconvert (memopv2i64 addr:$src))))]>, VEX, VEX_L;
2030 def VCVTDQ2PDYrr : S2SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
2031 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
2033 (int_x86_avx_cvtdq2_pd_256 VR128:$src))]>, VEX, VEX_L;
2036 let neverHasSideEffects = 1, mayLoad = 1 in
2037 def CVTDQ2PDrm : S2SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2038 "cvtdq2pd\t{$src, $dst|$dst, $src}", [],
2040 def CVTDQ2PDrr : S2SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2041 "cvtdq2pd\t{$src, $dst|$dst, $src}",
2042 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))],
2045 // AVX 256-bit register conversion intrinsics
2046 let Predicates = [HasAVX] in {
2047 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
2048 (VCVTDQ2PDYrr VR128:$src)>;
2049 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
2050 (VCVTDQ2PDYrm addr:$src)>;
2051 } // Predicates = [HasAVX]
2053 // Convert packed double to packed single
2054 // The assembler can recognize rr 256-bit instructions by seeing a ymm
2055 // register, but the same isn't true when using memory operands instead.
2056 // Provide other assembly rr and rm forms to address this explicitly.
2057 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2058 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2059 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2060 IIC_SSE_CVT_PD_RR>, VEX;
2063 def : InstAlias<"vcvtpd2psx\t{$src, $dst|$dst, $src}",
2064 (VCVTPD2PSrr VR128:$dst, VR128:$src)>;
2065 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2066 "cvtpd2psx\t{$src, $dst|$dst, $src}",
2068 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2069 IIC_SSE_CVT_PD_RM>, VEX;
2072 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
2073 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2075 (int_x86_avx_cvt_pd2_ps_256 VR256:$src))],
2076 IIC_SSE_CVT_PD_RR>, VEX, VEX_L;
2077 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
2078 "cvtpd2ps{y}\t{$src, $dst|$dst, $src}",
2080 (int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)))],
2081 IIC_SSE_CVT_PD_RM>, VEX, VEX_L;
2082 def : InstAlias<"vcvtpd2ps\t{$src, $dst|$dst, $src}",
2083 (VCVTPD2PSYrr VR128:$dst, VR256:$src)>;
2085 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2086 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2087 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))],
2089 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2090 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2092 (int_x86_sse2_cvtpd2ps (memopv2f64 addr:$src)))],
2096 // AVX 256-bit register conversion intrinsics
2097 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2098 // whenever possible to avoid declaring two versions of each one.
2099 let Predicates = [HasAVX] in {
2100 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2101 (VCVTDQ2PSYrr VR256:$src)>;
2102 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2103 (VCVTDQ2PSYrm addr:$src)>;
2105 // Match fround and fextend for 128/256-bit conversions
2106 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2107 (VCVTPD2PSrr VR128:$src)>;
2108 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2109 (VCVTPD2PSXrm addr:$src)>;
2110 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2111 (VCVTPD2PSYrr VR256:$src)>;
2112 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2113 (VCVTPD2PSYrm addr:$src)>;
2115 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2116 (VCVTPS2PDrr VR128:$src)>;
2117 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2118 (VCVTPS2PDYrr VR128:$src)>;
2119 def : Pat<(v4f64 (extloadv4f32 addr:$src)),
2120 (VCVTPS2PDYrm addr:$src)>;
2123 let Predicates = [UseSSE2] in {
2124 // Match fround and fextend for 128 conversions
2125 def : Pat<(v4f32 (X86vfpround (v2f64 VR128:$src))),
2126 (CVTPD2PSrr VR128:$src)>;
2127 def : Pat<(v4f32 (X86vfpround (memopv2f64 addr:$src))),
2128 (CVTPD2PSrm addr:$src)>;
2130 def : Pat<(v2f64 (X86vfpext (v4f32 VR128:$src))),
2131 (CVTPS2PDrr VR128:$src)>;
2134 //===----------------------------------------------------------------------===//
2135 // SSE 1 & 2 - Compare Instructions
2136 //===----------------------------------------------------------------------===//
2138 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2139 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2140 Operand CC, SDNode OpNode, ValueType VT,
2141 PatFrag ld_frag, string asm, string asm_alt,
2143 def rr : SIi8<0xC2, MRMSrcReg,
2144 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2145 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))],
2147 def rm : SIi8<0xC2, MRMSrcMem,
2148 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2149 [(set RC:$dst, (OpNode (VT RC:$src1),
2150 (ld_frag addr:$src2), imm:$cc))],
2153 // Accept explicit immediate argument form instead of comparison code.
2154 let neverHasSideEffects = 1 in {
2155 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2156 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, [],
2157 IIC_SSE_ALU_F32S_RR>;
2159 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2160 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, [],
2161 IIC_SSE_ALU_F32S_RM>;
2165 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, AVXCC, X86cmpss, f32, loadf32,
2166 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2167 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2169 XS, VEX_4V, VEX_LIG;
2170 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, AVXCC, X86cmpsd, f64, loadf64,
2171 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2172 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2173 SSE_ALU_F32S>, // same latency as 32 bit compare
2174 XD, VEX_4V, VEX_LIG;
2176 let Constraints = "$src1 = $dst" in {
2177 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, SSECC, X86cmpss, f32, loadf32,
2178 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2179 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}", SSE_ALU_F32S>,
2181 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, SSECC, X86cmpsd, f64, loadf64,
2182 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2183 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2184 SSE_ALU_F32S>, // same latency as 32 bit compare
2188 multiclass sse12_cmp_scalar_int<X86MemOperand x86memop, Operand CC,
2189 Intrinsic Int, string asm, OpndItins itins> {
2190 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2191 (ins VR128:$src1, VR128:$src, CC:$cc), asm,
2192 [(set VR128:$dst, (Int VR128:$src1,
2193 VR128:$src, imm:$cc))],
2195 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2196 (ins VR128:$src1, x86memop:$src, CC:$cc), asm,
2197 [(set VR128:$dst, (Int VR128:$src1,
2198 (load addr:$src), imm:$cc))],
2202 // Aliases to match intrinsics which expect XMM operand(s).
2203 defm Int_VCMPSS : sse12_cmp_scalar_int<f32mem, AVXCC, int_x86_sse_cmp_ss,
2204 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
2207 defm Int_VCMPSD : sse12_cmp_scalar_int<f64mem, AVXCC, int_x86_sse2_cmp_sd,
2208 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
2209 SSE_ALU_F32S>, // same latency as f32
2211 let Constraints = "$src1 = $dst" in {
2212 defm Int_CMPSS : sse12_cmp_scalar_int<f32mem, SSECC, int_x86_sse_cmp_ss,
2213 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
2215 defm Int_CMPSD : sse12_cmp_scalar_int<f64mem, SSECC, int_x86_sse2_cmp_sd,
2216 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
2217 SSE_ALU_F32S>, // same latency as f32
2222 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2223 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2224 ValueType vt, X86MemOperand x86memop,
2225 PatFrag ld_frag, string OpcodeStr, Domain d> {
2226 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2227 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2228 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))],
2229 IIC_SSE_COMIS_RR, d>;
2230 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2231 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2232 [(set EFLAGS, (OpNode (vt RC:$src1),
2233 (ld_frag addr:$src2)))],
2234 IIC_SSE_COMIS_RM, d>;
2237 let Defs = [EFLAGS] in {
2238 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2239 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2240 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2241 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2243 let Pattern = []<dag> in {
2244 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2245 "comiss", SSEPackedSingle>, TB, VEX,
2247 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2248 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2252 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2253 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2254 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2255 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2257 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2258 load, "comiss", SSEPackedSingle>, TB, VEX;
2259 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2260 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2261 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2262 "ucomiss", SSEPackedSingle>, TB;
2263 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2264 "ucomisd", SSEPackedDouble>, TB, OpSize;
2266 let Pattern = []<dag> in {
2267 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2268 "comiss", SSEPackedSingle>, TB;
2269 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2270 "comisd", SSEPackedDouble>, TB, OpSize;
2273 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2274 load, "ucomiss", SSEPackedSingle>, TB;
2275 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2276 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2278 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2279 "comiss", SSEPackedSingle>, TB;
2280 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2281 "comisd", SSEPackedDouble>, TB, OpSize;
2282 } // Defs = [EFLAGS]
2284 // sse12_cmp_packed - sse 1 & 2 compare packed instructions
2285 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2286 Operand CC, Intrinsic Int, string asm,
2287 string asm_alt, Domain d> {
2288 def rri : PIi8<0xC2, MRMSrcReg,
2289 (outs RC:$dst), (ins RC:$src1, RC:$src2, CC:$cc), asm,
2290 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))],
2291 IIC_SSE_CMPP_RR, d>;
2292 def rmi : PIi8<0xC2, MRMSrcMem,
2293 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, CC:$cc), asm,
2294 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))],
2295 IIC_SSE_CMPP_RM, d>;
2297 // Accept explicit immediate argument form instead of comparison code.
2298 let neverHasSideEffects = 1 in {
2299 def rri_alt : PIi8<0xC2, MRMSrcReg,
2300 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2301 asm_alt, [], IIC_SSE_CMPP_RR, d>;
2302 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2303 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, i8imm:$cc),
2304 asm_alt, [], IIC_SSE_CMPP_RM, d>;
2308 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse_cmp_ps,
2309 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2310 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2311 SSEPackedSingle>, TB, VEX_4V;
2312 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, AVXCC, int_x86_sse2_cmp_pd,
2313 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2314 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2315 SSEPackedDouble>, TB, OpSize, VEX_4V;
2316 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_ps_256,
2317 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2318 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2319 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2320 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, AVXCC, int_x86_avx_cmp_pd_256,
2321 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2322 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2323 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2324 let Constraints = "$src1 = $dst" in {
2325 defm CMPPS : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse_cmp_ps,
2326 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2327 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2328 SSEPackedSingle>, TB;
2329 defm CMPPD : sse12_cmp_packed<VR128, f128mem, SSECC, int_x86_sse2_cmp_pd,
2330 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2331 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2332 SSEPackedDouble>, TB, OpSize;
2335 let Predicates = [HasAVX] in {
2336 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2337 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2338 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2339 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2340 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2341 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2342 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2343 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2345 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2346 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2347 def : Pat<(v8i32 (X86cmpp (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2348 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2349 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2350 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2351 def : Pat<(v4i64 (X86cmpp (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2352 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2355 let Predicates = [UseSSE1] in {
2356 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2357 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2358 def : Pat<(v4i32 (X86cmpp (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2359 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2362 let Predicates = [UseSSE2] in {
2363 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2364 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2365 def : Pat<(v2i64 (X86cmpp (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2366 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2369 //===----------------------------------------------------------------------===//
2370 // SSE 1 & 2 - Shuffle Instructions
2371 //===----------------------------------------------------------------------===//
2373 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2374 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2375 ValueType vt, string asm, PatFrag mem_frag,
2376 Domain d, bit IsConvertibleToThreeAddress = 0> {
2377 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2378 (ins RC:$src1, x86memop:$src2, i8imm:$src3), asm,
2379 [(set RC:$dst, (vt (X86Shufp RC:$src1, (mem_frag addr:$src2),
2380 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2381 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2382 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2383 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2384 [(set RC:$dst, (vt (X86Shufp RC:$src1, RC:$src2,
2385 (i8 imm:$src3))))], IIC_SSE_SHUFP, d>;
2388 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2389 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2390 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2391 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2392 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2393 memopv8f32, SSEPackedSingle>, TB, VEX_4V, VEX_L;
2394 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2395 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2396 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2397 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2398 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2399 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2401 let Constraints = "$src1 = $dst" in {
2402 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2403 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2404 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2406 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2407 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2408 memopv2f64, SSEPackedDouble, 1 /* cvt to pshufd */>,
2412 let Predicates = [HasAVX] in {
2413 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2414 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2415 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2416 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2417 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2419 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2420 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2421 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2422 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2423 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2426 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2427 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2428 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2429 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2430 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2432 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2433 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2434 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2435 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2436 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2439 let Predicates = [UseSSE1] in {
2440 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2441 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2442 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2443 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2444 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2447 let Predicates = [UseSSE2] in {
2448 // Generic SHUFPD patterns
2449 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2450 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2451 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2452 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2453 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2456 //===----------------------------------------------------------------------===//
2457 // SSE 1 & 2 - Unpack Instructions
2458 //===----------------------------------------------------------------------===//
2460 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2461 multiclass sse12_unpack_interleave<bits<8> opc, SDNode OpNode, ValueType vt,
2462 PatFrag mem_frag, RegisterClass RC,
2463 X86MemOperand x86memop, string asm,
2465 def rr : PI<opc, MRMSrcReg,
2466 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2468 (vt (OpNode RC:$src1, RC:$src2)))],
2470 def rm : PI<opc, MRMSrcMem,
2471 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2473 (vt (OpNode RC:$src1,
2474 (mem_frag addr:$src2))))],
2478 defm VUNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2479 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2480 SSEPackedSingle>, TB, VEX_4V;
2481 defm VUNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2482 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2483 SSEPackedDouble>, TB, OpSize, VEX_4V;
2484 defm VUNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2485 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2486 SSEPackedSingle>, TB, VEX_4V;
2487 defm VUNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2488 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2489 SSEPackedDouble>, TB, OpSize, VEX_4V;
2491 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, X86Unpckh, v8f32, memopv8f32,
2492 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2493 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2494 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, X86Unpckh, v4f64, memopv4f64,
2495 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2496 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2497 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, X86Unpckl, v8f32, memopv8f32,
2498 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2499 SSEPackedSingle>, TB, VEX_4V, VEX_L;
2500 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, X86Unpckl, v4f64, memopv4f64,
2501 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2502 SSEPackedDouble>, TB, OpSize, VEX_4V, VEX_L;
2504 let Constraints = "$src1 = $dst" in {
2505 defm UNPCKHPS: sse12_unpack_interleave<0x15, X86Unpckh, v4f32, memopv4f32,
2506 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2507 SSEPackedSingle>, TB;
2508 defm UNPCKHPD: sse12_unpack_interleave<0x15, X86Unpckh, v2f64, memopv2f64,
2509 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2510 SSEPackedDouble>, TB, OpSize;
2511 defm UNPCKLPS: sse12_unpack_interleave<0x14, X86Unpckl, v4f32, memopv4f32,
2512 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2513 SSEPackedSingle>, TB;
2514 defm UNPCKLPD: sse12_unpack_interleave<0x14, X86Unpckl, v2f64, memopv2f64,
2515 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2516 SSEPackedDouble>, TB, OpSize;
2517 } // Constraints = "$src1 = $dst"
2519 let Predicates = [HasAVX1Only] in {
2520 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2521 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2522 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
2523 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2524 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
2525 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2526 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
2527 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2529 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
2530 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2531 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
2532 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2533 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
2534 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2535 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
2536 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2539 let Predicates = [HasAVX] in {
2540 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2541 // problem is during lowering, where it's not possible to recognize the load
2542 // fold cause it has two uses through a bitcast. One use disappears at isel
2543 // time and the fold opportunity reappears.
2544 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2545 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2548 let Predicates = [UseSSE2] in {
2549 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2550 // problem is during lowering, where it's not possible to recognize the load
2551 // fold cause it has two uses through a bitcast. One use disappears at isel
2552 // time and the fold opportunity reappears.
2553 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2554 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2557 //===----------------------------------------------------------------------===//
2558 // SSE 1 & 2 - Extract Floating-Point Sign mask
2559 //===----------------------------------------------------------------------===//
2561 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2562 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2564 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2565 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2566 [(set GR32:$dst, (Int RC:$src))], IIC_SSE_MOVMSK, d>;
2567 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2568 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [],
2569 IIC_SSE_MOVMSK, d>, REX_W;
2572 let Predicates = [HasAVX] in {
2573 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2574 "movmskps", SSEPackedSingle>, TB, VEX;
2575 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2576 "movmskpd", SSEPackedDouble>, TB,
2578 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2579 "movmskps", SSEPackedSingle>, TB,
2581 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2582 "movmskpd", SSEPackedDouble>, TB,
2585 def : Pat<(i32 (X86fgetsign FR32:$src)),
2586 (VMOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2587 def : Pat<(i64 (X86fgetsign FR32:$src)),
2588 (VMOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>;
2589 def : Pat<(i32 (X86fgetsign FR64:$src)),
2590 (VMOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2591 def : Pat<(i64 (X86fgetsign FR64:$src)),
2592 (VMOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>;
2595 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2596 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2597 SSEPackedSingle>, TB, VEX;
2598 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2599 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2600 SSEPackedDouble>, TB,
2602 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2603 "movmskps\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2604 SSEPackedSingle>, TB, VEX, VEX_L;
2605 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2606 "movmskpd\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK,
2607 SSEPackedDouble>, TB,
2611 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2612 SSEPackedSingle>, TB;
2613 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2614 SSEPackedDouble>, TB, OpSize;
2616 def : Pat<(i32 (X86fgetsign FR32:$src)),
2617 (MOVMSKPSrr32 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2618 Requires<[UseSSE1]>;
2619 def : Pat<(i64 (X86fgetsign FR32:$src)),
2620 (MOVMSKPSrr64 (COPY_TO_REGCLASS FR32:$src, VR128))>,
2621 Requires<[UseSSE1]>;
2622 def : Pat<(i32 (X86fgetsign FR64:$src)),
2623 (MOVMSKPDrr32 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2624 Requires<[UseSSE2]>;
2625 def : Pat<(i64 (X86fgetsign FR64:$src)),
2626 (MOVMSKPDrr64 (COPY_TO_REGCLASS FR64:$src, VR128))>,
2627 Requires<[UseSSE2]>;
2629 //===---------------------------------------------------------------------===//
2630 // SSE2 - Packed Integer Logical Instructions
2631 //===---------------------------------------------------------------------===//
2633 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2635 /// PDI_binop_rm - Simple SSE2 binary operator.
2636 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2637 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2638 X86MemOperand x86memop, OpndItins itins,
2639 bit IsCommutable, bit Is2Addr> {
2640 let isCommutable = IsCommutable in
2641 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2642 (ins RC:$src1, RC:$src2),
2644 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2645 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2646 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>;
2647 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2648 (ins RC:$src1, x86memop:$src2),
2650 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2651 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2652 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2653 (bitconvert (memop_frag addr:$src2)))))],
2656 } // ExeDomain = SSEPackedInt
2658 multiclass PDI_binop_all<bits<8> opc, string OpcodeStr, SDNode Opcode,
2659 ValueType OpVT128, ValueType OpVT256,
2660 OpndItins itins, bit IsCommutable = 0> {
2661 let Predicates = [HasAVX] in
2662 defm V#NAME# : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode, OpVT128,
2663 VR128, memopv2i64, i128mem, itins, IsCommutable, 0>, VEX_4V;
2665 let Constraints = "$src1 = $dst" in
2666 defm #NAME# : PDI_binop_rm<opc, OpcodeStr, Opcode, OpVT128, VR128,
2667 memopv2i64, i128mem, itins, IsCommutable, 1>;
2669 let Predicates = [HasAVX2] in
2670 defm V#NAME#Y : PDI_binop_rm<opc, !strconcat("v", OpcodeStr), Opcode,
2671 OpVT256, VR256, memopv4i64, i256mem, itins,
2672 IsCommutable, 0>, VEX_4V, VEX_L;
2675 // These are ordered here for pattern ordering requirements with the fp versions
2677 defm PAND : PDI_binop_all<0xDB, "pand", and, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2678 defm POR : PDI_binop_all<0xEB, "por", or, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2679 defm PXOR : PDI_binop_all<0xEF, "pxor", xor, v2i64, v4i64, SSE_BIT_ITINS_P, 1>;
2680 defm PANDN : PDI_binop_all<0xDF, "pandn", X86andnp, v2i64, v4i64,
2681 SSE_BIT_ITINS_P, 0>;
2683 //===----------------------------------------------------------------------===//
2684 // SSE 1 & 2 - Logical Instructions
2685 //===----------------------------------------------------------------------===//
2687 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2689 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2690 SDNode OpNode, OpndItins itins> {
2691 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2692 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, itins, 0>,
2695 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2696 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, itins, 0>,
2699 let Constraints = "$src1 = $dst" in {
2700 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2701 f32, f128mem, memopfsf32, SSEPackedSingle, itins>,
2704 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2705 f64, f128mem, memopfsf64, SSEPackedDouble, itins>,
2710 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2711 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand,
2713 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for,
2715 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor,
2718 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2719 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef,
2722 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2724 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2726 defm V#NAME#PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2727 !strconcat(OpcodeStr, "ps"), f256mem,
2728 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2729 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2730 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V, VEX_L;
2732 defm V#NAME#PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2733 !strconcat(OpcodeStr, "pd"), f256mem,
2734 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2735 (bc_v4i64 (v4f64 VR256:$src2))))],
2736 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2737 (memopv4i64 addr:$src2)))], 0>,
2738 TB, OpSize, VEX_4V, VEX_L;
2740 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2741 // are all promoted to v2i64, and the patterns are covered by the int
2742 // version. This is needed in SSE only, because v2i64 isn't supported on
2743 // SSE1, but only on SSE2.
2744 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2745 !strconcat(OpcodeStr, "ps"), f128mem, [],
2746 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2747 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2749 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2750 !strconcat(OpcodeStr, "pd"), f128mem,
2751 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2752 (bc_v2i64 (v2f64 VR128:$src2))))],
2753 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2754 (memopv2i64 addr:$src2)))], 0>,
2757 let Constraints = "$src1 = $dst" in {
2758 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2759 !strconcat(OpcodeStr, "ps"), f128mem,
2760 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2761 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2762 (memopv2i64 addr:$src2)))]>, TB;
2764 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2765 !strconcat(OpcodeStr, "pd"), f128mem,
2766 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2767 (bc_v2i64 (v2f64 VR128:$src2))))],
2768 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2769 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2773 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2774 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2775 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2776 let isCommutable = 0 in
2777 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2779 //===----------------------------------------------------------------------===//
2780 // SSE 1 & 2 - Arithmetic Instructions
2781 //===----------------------------------------------------------------------===//
2783 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2786 /// In addition, we also have a special variant of the scalar form here to
2787 /// represent the associated intrinsic operation. This form is unlike the
2788 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2789 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2791 /// These three forms can each be reg+reg or reg+mem.
2794 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2796 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2799 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2800 OpNode, FR32, f32mem,
2801 itins.s, Is2Addr>, XS;
2802 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2803 OpNode, FR64, f64mem,
2804 itins.d, Is2Addr>, XD;
2807 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr,
2808 SDNode OpNode, SizeItins itins> {
2809 let Predicates = [HasAVX] in {
2810 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2811 VR128, v4f32, f128mem, memopv4f32,
2812 SSEPackedSingle, itins.s, 0>, TB, VEX_4V;
2813 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2814 VR128, v2f64, f128mem, memopv2f64,
2815 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V;
2817 defm V#NAME#PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"),
2818 OpNode, VR256, v8f32, f256mem, memopv8f32,
2819 SSEPackedSingle, itins.s, 0>, TB, VEX_4V, VEX_L;
2820 defm V#NAME#PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"),
2821 OpNode, VR256, v4f64, f256mem, memopv4f64,
2822 SSEPackedDouble, itins.d, 0>, TB, OpSize, VEX_4V, VEX_L;
2825 let Constraints = "$src1 = $dst" in {
2826 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2827 v4f32, f128mem, memopv4f32, SSEPackedSingle,
2829 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2830 v2f64, f128mem, memopv2f64, SSEPackedDouble,
2831 itins.d, 1>, TB, OpSize;
2835 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2838 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2839 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32,
2840 itins.s, Is2Addr>, XS;
2841 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2842 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64,
2843 itins.d, Is2Addr>, XD;
2846 // Binary Arithmetic instructions
2847 defm ADD : basic_sse12_fp_binop_p<0x58, "add", fadd, SSE_ALU_ITINS_P>;
2848 defm MUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, SSE_MUL_ITINS_P>;
2849 let isCommutable = 0 in {
2850 defm SUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, SSE_ALU_ITINS_P>;
2851 defm DIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, SSE_DIV_ITINS_P>;
2852 defm MAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, SSE_ALU_ITINS_P>;
2853 defm MIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, SSE_ALU_ITINS_P>;
2856 let isCodeGenOnly = 1 in {
2857 defm MAXC: basic_sse12_fp_binop_p<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_P>;
2858 defm MINC: basic_sse12_fp_binop_p<0x5D, "min", X86fminc, SSE_ALU_ITINS_P>;
2861 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S, 0>,
2862 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S, 0>,
2864 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S, 0>,
2865 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S, 0>,
2868 let isCommutable = 0 in {
2869 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S, 0>,
2870 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S, 0>,
2872 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S, 0>,
2873 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S, 0>,
2875 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S, 0>,
2876 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S, 0>,
2878 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S, 0>,
2879 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S, 0>,
2883 let Constraints = "$src1 = $dst" in {
2884 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd, SSE_ALU_ITINS_S>,
2885 basic_sse12_fp_binop_s_int<0x58, "add", SSE_ALU_ITINS_S>;
2886 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, SSE_MUL_ITINS_S>,
2887 basic_sse12_fp_binop_s_int<0x59, "mul", SSE_MUL_ITINS_S>;
2889 let isCommutable = 0 in {
2890 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, SSE_ALU_ITINS_S>,
2891 basic_sse12_fp_binop_s_int<0x5C, "sub", SSE_ALU_ITINS_S>;
2892 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, SSE_DIV_ITINS_S>,
2893 basic_sse12_fp_binop_s_int<0x5E, "div", SSE_DIV_ITINS_S>;
2894 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, SSE_ALU_ITINS_S>,
2895 basic_sse12_fp_binop_s_int<0x5F, "max", SSE_ALU_ITINS_S>;
2896 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, SSE_ALU_ITINS_S>,
2897 basic_sse12_fp_binop_s_int<0x5D, "min", SSE_ALU_ITINS_S>;
2901 let isCodeGenOnly = 1 in {
2902 defm VMAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S, 0>,
2904 defm VMINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S, 0>,
2906 let Constraints = "$src1 = $dst" in {
2907 defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
2908 defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
2913 /// In addition, we also have a special variant of the scalar form here to
2914 /// represent the associated intrinsic operation. This form is unlike the
2915 /// plain scalar form, in that it takes an entire vector (instead of a
2916 /// scalar) and leaves the top elements undefined.
2918 /// And, we have a special variant form for a full-vector intrinsic form.
2920 def SSE_SQRTP : OpndItins<
2921 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2924 def SSE_SQRTS : OpndItins<
2925 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2928 def SSE_RCPP : OpndItins<
2929 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2932 def SSE_RCPS : OpndItins<
2933 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2936 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2937 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2938 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2939 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2940 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2941 [(set FR32:$dst, (OpNode FR32:$src))]>;
2942 // For scalar unary operations, fold a load into the operation
2943 // only in OptForSize mode. It eliminates an instruction, but it also
2944 // eliminates a whole-register clobber (the load), so it introduces a
2945 // partial register update condition.
2946 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2947 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2948 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2949 Requires<[UseSSE1, OptForSize]>;
2950 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2951 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2952 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2953 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2954 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2955 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2958 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2959 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2960 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2961 !strconcat(OpcodeStr,
2962 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2963 let mayLoad = 1 in {
2964 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2965 !strconcat(OpcodeStr,
2966 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2967 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2968 (ins VR128:$src1, ssmem:$src2),
2969 !strconcat(OpcodeStr,
2970 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2974 /// sse1_fp_unop_p - SSE1 unops in packed form.
2975 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2977 let Predicates = [HasAVX] in {
2978 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2979 !strconcat(!strconcat("v", OpcodeStr),
2980 "ps\t{$src, $dst|$dst, $src}"),
2981 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
2983 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2984 !strconcat(!strconcat("v", OpcodeStr),
2985 "ps\t{$src, $dst|$dst, $src}"),
2986 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
2988 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2989 !strconcat(!strconcat("v", OpcodeStr),
2990 "ps\t{$src, $dst|$dst, $src}"),
2991 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
2992 itins.rr>, VEX, VEX_L;
2993 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2994 !strconcat(!strconcat("v", OpcodeStr),
2995 "ps\t{$src, $dst|$dst, $src}"),
2996 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
2997 itins.rm>, VEX, VEX_L;
3000 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3001 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3002 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3003 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3004 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3005 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3008 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3009 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3010 Intrinsic V4F32Int, Intrinsic V8F32Int,
3012 let Predicates = [HasAVX] in {
3013 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3014 !strconcat(!strconcat("v", OpcodeStr),
3015 "ps\t{$src, $dst|$dst, $src}"),
3016 [(set VR128:$dst, (V4F32Int VR128:$src))],
3018 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3019 !strconcat(!strconcat("v", OpcodeStr),
3020 "ps\t{$src, $dst|$dst, $src}"),
3021 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3023 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3024 !strconcat(!strconcat("v", OpcodeStr),
3025 "ps\t{$src, $dst|$dst, $src}"),
3026 [(set VR256:$dst, (V8F32Int VR256:$src))],
3027 itins.rr>, VEX, VEX_L;
3028 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3030 !strconcat(!strconcat("v", OpcodeStr),
3031 "ps\t{$src, $dst|$dst, $src}"),
3032 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3033 itins.rm>, VEX, VEX_L;
3036 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3037 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3038 [(set VR128:$dst, (V4F32Int VR128:$src))],
3040 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3041 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3042 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3046 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3047 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3048 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3049 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3050 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3051 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3052 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3053 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3054 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3055 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3056 Requires<[UseSSE2, OptForSize]>;
3057 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3058 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3059 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3060 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3061 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3062 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3065 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3066 let hasSideEffects = 0 in
3067 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3068 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3069 !strconcat(OpcodeStr,
3070 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3071 let mayLoad = 1 in {
3072 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3073 !strconcat(OpcodeStr,
3074 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3075 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3076 (ins VR128:$src1, sdmem:$src2),
3077 !strconcat(OpcodeStr,
3078 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3082 /// sse2_fp_unop_p_new - SSE2 unops in vector forms.
3083 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3084 SDNode OpNode, OpndItins itins> {
3085 let Predicates = [HasAVX] in {
3086 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3087 !strconcat(!strconcat("v", OpcodeStr),
3088 "pd\t{$src, $dst|$dst, $src}"),
3089 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3091 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3092 !strconcat(!strconcat("v", OpcodeStr),
3093 "pd\t{$src, $dst|$dst, $src}"),
3094 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3096 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3097 !strconcat(!strconcat("v", OpcodeStr),
3098 "pd\t{$src, $dst|$dst, $src}"),
3099 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3100 itins.rr>, VEX, VEX_L;
3101 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3102 !strconcat(!strconcat("v", OpcodeStr),
3103 "pd\t{$src, $dst|$dst, $src}"),
3104 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3105 itins.rm>, VEX, VEX_L;
3108 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3109 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3110 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3111 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3112 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3113 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3116 defm SQRT : sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
3117 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
3118 defm RSQRT : sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
3119 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3120 int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
3121 defm RCP : sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3122 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3123 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3125 let Predicates = [HasAVX] in {
3127 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3128 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3130 // Reciprocal approximations. Note that these typically require refinement
3131 // in order to obtain suitable precision.
3132 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3133 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3136 def : Pat<(f32 (fsqrt FR32:$src)),
3137 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3138 def : Pat<(f32 (fsqrt (load addr:$src))),
3139 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3140 Requires<[HasAVX, OptForSize]>;
3141 def : Pat<(f64 (fsqrt FR64:$src)),
3142 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3143 def : Pat<(f64 (fsqrt (load addr:$src))),
3144 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3145 Requires<[HasAVX, OptForSize]>;
3147 def : Pat<(f32 (X86frsqrt FR32:$src)),
3148 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3149 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3150 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3151 Requires<[HasAVX, OptForSize]>;
3153 def : Pat<(f32 (X86frcp FR32:$src)),
3154 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3155 def : Pat<(f32 (X86frcp (load addr:$src))),
3156 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3157 Requires<[HasAVX, OptForSize]>;
3159 let Predicates = [HasAVX] in {
3160 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3161 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3162 (COPY_TO_REGCLASS VR128:$src, FR32)),
3164 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3165 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3167 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3168 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3169 (COPY_TO_REGCLASS VR128:$src, FR64)),
3171 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3172 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3174 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3175 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3176 (COPY_TO_REGCLASS VR128:$src, FR32)),
3178 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3179 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3181 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3182 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3183 (COPY_TO_REGCLASS VR128:$src, FR32)),
3185 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3186 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3190 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3192 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3195 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3196 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3197 Intrinsic F32Int, OpndItins itins> {
3198 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3199 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3200 [(set FR32:$dst, (OpNode FR32:$src))]>;
3201 // For scalar unary operations, fold a load into the operation
3202 // only in OptForSize mode. It eliminates an instruction, but it also
3203 // eliminates a whole-register clobber (the load), so it introduces a
3204 // partial register update condition.
3205 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3206 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3207 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3208 Requires<[UseSSE1, OptForSize]>;
3209 let Constraints = "$src1 = $dst" in {
3210 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3211 (ins VR128:$src1, VR128:$src2),
3212 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3214 let mayLoad = 1, hasSideEffects = 0 in
3215 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3216 (ins VR128:$src1, ssmem:$src2),
3217 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3222 // Reciprocal approximations. Note that these typically require refinement
3223 // in order to obtain suitable precision.
3224 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3226 let Predicates = [UseSSE1] in {
3227 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3228 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3231 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3233 let Predicates = [UseSSE1] in {
3234 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3235 (RCPSSr_Int VR128:$src, VR128:$src)>;
3238 // There is no f64 version of the reciprocal approximation instructions.
3240 //===----------------------------------------------------------------------===//
3241 // SSE 1 & 2 - Non-temporal stores
3242 //===----------------------------------------------------------------------===//
3244 let AddedComplexity = 400 in { // Prefer non-temporal versions
3245 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3246 (ins f128mem:$dst, VR128:$src),
3247 "movntps\t{$src, $dst|$dst, $src}",
3248 [(alignednontemporalstore (v4f32 VR128:$src),
3250 IIC_SSE_MOVNT>, VEX;
3251 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3252 (ins f128mem:$dst, VR128:$src),
3253 "movntpd\t{$src, $dst|$dst, $src}",
3254 [(alignednontemporalstore (v2f64 VR128:$src),
3256 IIC_SSE_MOVNT>, VEX;
3258 let ExeDomain = SSEPackedInt in
3259 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3260 (ins f128mem:$dst, VR128:$src),
3261 "movntdq\t{$src, $dst|$dst, $src}",
3262 [(alignednontemporalstore (v2i64 VR128:$src),
3264 IIC_SSE_MOVNT>, VEX;
3266 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3267 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3269 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3270 (ins f256mem:$dst, VR256:$src),
3271 "movntps\t{$src, $dst|$dst, $src}",
3272 [(alignednontemporalstore (v8f32 VR256:$src),
3274 IIC_SSE_MOVNT>, VEX, VEX_L;
3275 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3276 (ins f256mem:$dst, VR256:$src),
3277 "movntpd\t{$src, $dst|$dst, $src}",
3278 [(alignednontemporalstore (v4f64 VR256:$src),
3280 IIC_SSE_MOVNT>, VEX, VEX_L;
3281 let ExeDomain = SSEPackedInt in
3282 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3283 (ins f256mem:$dst, VR256:$src),
3284 "movntdq\t{$src, $dst|$dst, $src}",
3285 [(alignednontemporalstore (v4i64 VR256:$src),
3287 IIC_SSE_MOVNT>, VEX, VEX_L;
3290 let AddedComplexity = 400 in { // Prefer non-temporal versions
3291 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3292 "movntps\t{$src, $dst|$dst, $src}",
3293 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3295 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3296 "movntpd\t{$src, $dst|$dst, $src}",
3297 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3300 let ExeDomain = SSEPackedInt in
3301 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3302 "movntdq\t{$src, $dst|$dst, $src}",
3303 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3306 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3307 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3309 // There is no AVX form for instructions below this point
3310 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3311 "movnti{l}\t{$src, $dst|$dst, $src}",
3312 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3314 TB, Requires<[HasSSE2]>;
3315 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3316 "movnti{q}\t{$src, $dst|$dst, $src}",
3317 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3319 TB, Requires<[HasSSE2]>;
3322 //===----------------------------------------------------------------------===//
3323 // SSE 1 & 2 - Prefetch and memory fence
3324 //===----------------------------------------------------------------------===//
3326 // Prefetch intrinsic.
3327 let Predicates = [HasSSE1] in {
3328 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3329 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3330 IIC_SSE_PREFETCH>, TB;
3331 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3332 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3333 IIC_SSE_PREFETCH>, TB;
3334 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3335 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3336 IIC_SSE_PREFETCH>, TB;
3337 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3338 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3339 IIC_SSE_PREFETCH>, TB;
3343 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3344 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3345 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3347 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3348 // was introduced with SSE2, it's backward compatible.
3349 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3351 // Load, store, and memory fence
3352 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3353 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3354 TB, Requires<[HasSSE1]>;
3355 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3356 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3357 TB, Requires<[HasSSE2]>;
3358 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3359 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3360 TB, Requires<[HasSSE2]>;
3362 def : Pat<(X86SFence), (SFENCE)>;
3363 def : Pat<(X86LFence), (LFENCE)>;
3364 def : Pat<(X86MFence), (MFENCE)>;
3366 //===----------------------------------------------------------------------===//
3367 // SSE 1 & 2 - Load/Store XCSR register
3368 //===----------------------------------------------------------------------===//
3370 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3371 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3372 IIC_SSE_LDMXCSR>, VEX;
3373 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3374 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3375 IIC_SSE_STMXCSR>, VEX;
3377 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3378 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3380 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3381 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3384 //===---------------------------------------------------------------------===//
3385 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3386 //===---------------------------------------------------------------------===//
3388 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3390 let neverHasSideEffects = 1 in {
3391 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3392 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3394 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3395 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3397 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3398 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3400 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3401 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3406 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3407 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3408 "movdqa\t{$src, $dst|$dst, $src}", [],
3411 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3412 "movdqa\t{$src, $dst|$dst, $src}", [],
3413 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3414 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3415 "movdqu\t{$src, $dst|$dst, $src}", [],
3418 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3419 "movdqu\t{$src, $dst|$dst, $src}", [],
3420 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3423 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3424 neverHasSideEffects = 1 in {
3425 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3426 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3428 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3429 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3431 let Predicates = [HasAVX] in {
3432 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3433 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3435 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3436 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3441 let mayStore = 1, neverHasSideEffects = 1 in {
3442 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3443 (ins i128mem:$dst, VR128:$src),
3444 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3446 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3447 (ins i256mem:$dst, VR256:$src),
3448 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3450 let Predicates = [HasAVX] in {
3451 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3452 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3454 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3455 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3460 let neverHasSideEffects = 1 in
3461 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3462 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3464 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3465 "movdqu\t{$src, $dst|$dst, $src}",
3466 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3469 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3470 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3471 "movdqa\t{$src, $dst|$dst, $src}", [],
3474 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3475 "movdqu\t{$src, $dst|$dst, $src}",
3476 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3479 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3480 neverHasSideEffects = 1 in {
3481 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3482 "movdqa\t{$src, $dst|$dst, $src}",
3483 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3485 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3486 "movdqu\t{$src, $dst|$dst, $src}",
3487 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3489 XS, Requires<[UseSSE2]>;
3492 let mayStore = 1 in {
3493 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3494 "movdqa\t{$src, $dst|$dst, $src}",
3495 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3497 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3498 "movdqu\t{$src, $dst|$dst, $src}",
3499 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3501 XS, Requires<[UseSSE2]>;
3504 } // ExeDomain = SSEPackedInt
3506 let Predicates = [HasAVX] in {
3507 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3508 (VMOVDQUmr addr:$dst, VR128:$src)>;
3509 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3510 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3512 let Predicates = [UseSSE2] in
3513 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3514 (MOVDQUmr addr:$dst, VR128:$src)>;
3516 //===---------------------------------------------------------------------===//
3517 // SSE2 - Packed Integer Arithmetic Instructions
3518 //===---------------------------------------------------------------------===//
3520 def SSE_PMADD : OpndItins<
3521 IIC_SSE_PMADD, IIC_SSE_PMADD
3524 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3526 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3527 RegisterClass RC, PatFrag memop_frag,
3528 X86MemOperand x86memop,
3530 bit IsCommutable = 0,
3532 let isCommutable = IsCommutable in
3533 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3534 (ins RC:$src1, RC:$src2),
3536 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3537 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3538 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3539 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3540 (ins RC:$src1, x86memop:$src2),
3542 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3543 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3544 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3548 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3549 Intrinsic IntId256, OpndItins itins,
3550 bit IsCommutable = 0> {
3551 let Predicates = [HasAVX] in
3552 defm V#NAME# : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3553 VR128, memopv2i64, i128mem, itins,
3554 IsCommutable, 0>, VEX_4V;
3556 let Constraints = "$src1 = $dst" in
3557 defm #NAME# : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3558 i128mem, itins, IsCommutable, 1>;
3560 let Predicates = [HasAVX2] in
3561 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3562 VR256, memopv4i64, i256mem, itins,
3563 IsCommutable, 0>, VEX_4V, VEX_L;
3566 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3567 string OpcodeStr, SDNode OpNode,
3568 SDNode OpNode2, RegisterClass RC,
3569 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3570 ShiftOpndItins itins,
3572 // src2 is always 128-bit
3573 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3574 (ins RC:$src1, VR128:$src2),
3576 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3577 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3578 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3580 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3581 (ins RC:$src1, i128mem:$src2),
3583 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3584 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3585 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3586 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3587 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3588 (ins RC:$src1, i32i8imm:$src2),
3590 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3591 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3592 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3595 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3596 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3597 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3598 PatFrag memop_frag, X86MemOperand x86memop,
3600 bit IsCommutable = 0, bit Is2Addr = 1> {
3601 let isCommutable = IsCommutable in
3602 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3603 (ins RC:$src1, RC:$src2),
3605 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3606 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3607 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3608 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3609 (ins RC:$src1, x86memop:$src2),
3611 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3612 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3613 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3614 (bitconvert (memop_frag addr:$src2)))))]>;
3616 } // ExeDomain = SSEPackedInt
3618 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3619 SSE_INTALU_ITINS_P, 1>;
3620 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3621 SSE_INTALU_ITINS_P, 1>;
3622 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3623 SSE_INTALU_ITINS_P, 1>;
3624 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3625 SSE_INTALUQ_ITINS_P, 1>;
3626 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3627 SSE_INTMUL_ITINS_P, 1>;
3628 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3629 SSE_INTALU_ITINS_P, 0>;
3630 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3631 SSE_INTALU_ITINS_P, 0>;
3632 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3633 SSE_INTALU_ITINS_P, 0>;
3634 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3635 SSE_INTALUQ_ITINS_P, 0>;
3636 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3637 SSE_INTALU_ITINS_P, 0>;
3638 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3639 SSE_INTALU_ITINS_P, 0>;
3640 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3641 SSE_INTALU_ITINS_P, 1>;
3642 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3643 SSE_INTALU_ITINS_P, 1>;
3644 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3645 SSE_INTALU_ITINS_P, 1>;
3646 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3647 SSE_INTALU_ITINS_P, 1>;
3650 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3651 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3652 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3653 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3654 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3655 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3656 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3657 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3658 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3659 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3660 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3661 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3662 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3663 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3664 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3665 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3666 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3667 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3668 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3669 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3670 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3671 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3672 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3673 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3675 let Predicates = [HasAVX] in
3676 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3677 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3679 let Predicates = [HasAVX2] in
3680 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3681 VR256, memopv4i64, i256mem,
3682 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3683 let Constraints = "$src1 = $dst" in
3684 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3685 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3687 //===---------------------------------------------------------------------===//
3688 // SSE2 - Packed Integer Logical Instructions
3689 //===---------------------------------------------------------------------===//
3691 let Predicates = [HasAVX] in {
3692 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3693 VR128, v8i16, v8i16, bc_v8i16,
3694 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3695 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3696 VR128, v4i32, v4i32, bc_v4i32,
3697 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3698 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3699 VR128, v2i64, v2i64, bc_v2i64,
3700 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3702 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3703 VR128, v8i16, v8i16, bc_v8i16,
3704 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3705 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3706 VR128, v4i32, v4i32, bc_v4i32,
3707 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3708 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3709 VR128, v2i64, v2i64, bc_v2i64,
3710 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3712 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3713 VR128, v8i16, v8i16, bc_v8i16,
3714 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3715 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3716 VR128, v4i32, v4i32, bc_v4i32,
3717 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3719 let ExeDomain = SSEPackedInt in {
3720 // 128-bit logical shifts.
3721 def VPSLLDQri : PDIi8<0x73, MRM7r,
3722 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3723 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3725 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3727 def VPSRLDQri : PDIi8<0x73, MRM3r,
3728 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3729 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3731 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3733 // PSRADQri doesn't exist in SSE[1-3].
3735 } // Predicates = [HasAVX]
3737 let Predicates = [HasAVX2] in {
3738 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3739 VR256, v16i16, v8i16, bc_v8i16,
3740 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3741 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3742 VR256, v8i32, v4i32, bc_v4i32,
3743 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3744 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3745 VR256, v4i64, v2i64, bc_v2i64,
3746 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3748 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3749 VR256, v16i16, v8i16, bc_v8i16,
3750 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3751 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3752 VR256, v8i32, v4i32, bc_v4i32,
3753 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3754 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3755 VR256, v4i64, v2i64, bc_v2i64,
3756 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3758 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3759 VR256, v16i16, v8i16, bc_v8i16,
3760 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3761 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3762 VR256, v8i32, v4i32, bc_v4i32,
3763 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3765 let ExeDomain = SSEPackedInt in {
3766 // 256-bit logical shifts.
3767 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3768 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3769 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3771 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3773 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3774 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3775 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3777 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3779 // PSRADQYri doesn't exist in SSE[1-3].
3781 } // Predicates = [HasAVX2]
3783 let Constraints = "$src1 = $dst" in {
3784 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3785 VR128, v8i16, v8i16, bc_v8i16,
3786 SSE_INTSHIFT_ITINS_P>;
3787 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3788 VR128, v4i32, v4i32, bc_v4i32,
3789 SSE_INTSHIFT_ITINS_P>;
3790 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3791 VR128, v2i64, v2i64, bc_v2i64,
3792 SSE_INTSHIFT_ITINS_P>;
3794 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3795 VR128, v8i16, v8i16, bc_v8i16,
3796 SSE_INTSHIFT_ITINS_P>;
3797 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3798 VR128, v4i32, v4i32, bc_v4i32,
3799 SSE_INTSHIFT_ITINS_P>;
3800 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3801 VR128, v2i64, v2i64, bc_v2i64,
3802 SSE_INTSHIFT_ITINS_P>;
3804 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3805 VR128, v8i16, v8i16, bc_v8i16,
3806 SSE_INTSHIFT_ITINS_P>;
3807 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3808 VR128, v4i32, v4i32, bc_v4i32,
3809 SSE_INTSHIFT_ITINS_P>;
3811 let ExeDomain = SSEPackedInt in {
3812 // 128-bit logical shifts.
3813 def PSLLDQri : PDIi8<0x73, MRM7r,
3814 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3815 "pslldq\t{$src2, $dst|$dst, $src2}",
3817 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3818 def PSRLDQri : PDIi8<0x73, MRM3r,
3819 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3820 "psrldq\t{$src2, $dst|$dst, $src2}",
3822 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3823 // PSRADQri doesn't exist in SSE[1-3].
3825 } // Constraints = "$src1 = $dst"
3827 let Predicates = [HasAVX] in {
3828 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3829 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3830 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3831 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3832 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3833 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3835 // Shift up / down and insert zero's.
3836 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3837 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3838 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3839 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3842 let Predicates = [HasAVX2] in {
3843 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3844 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3845 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3846 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3849 let Predicates = [UseSSE2] in {
3850 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3851 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3852 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3853 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3854 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3855 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3857 // Shift up / down and insert zero's.
3858 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3859 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3860 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3861 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3864 //===---------------------------------------------------------------------===//
3865 // SSE2 - Packed Integer Comparison Instructions
3866 //===---------------------------------------------------------------------===//
3868 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
3869 SSE_INTALU_ITINS_P, 1>;
3870 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
3871 SSE_INTALU_ITINS_P, 1>;
3872 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
3873 SSE_INTALU_ITINS_P, 1>;
3874 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
3875 SSE_INTALU_ITINS_P, 0>;
3876 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
3877 SSE_INTALU_ITINS_P, 0>;
3878 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
3879 SSE_INTALU_ITINS_P, 0>;
3881 //===---------------------------------------------------------------------===//
3882 // SSE2 - Packed Integer Pack Instructions
3883 //===---------------------------------------------------------------------===//
3885 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3886 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
3887 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3888 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
3889 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3890 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
3892 //===---------------------------------------------------------------------===//
3893 // SSE2 - Packed Integer Shuffle Instructions
3894 //===---------------------------------------------------------------------===//
3896 let ExeDomain = SSEPackedInt in {
3897 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
3898 def ri : Ii8<0x70, MRMSrcReg,
3899 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3900 !strconcat(OpcodeStr,
3901 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3902 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
3904 def mi : Ii8<0x70, MRMSrcMem,
3905 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3906 !strconcat(OpcodeStr,
3907 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3909 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
3914 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3915 def Yri : Ii8<0x70, MRMSrcReg,
3916 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3917 !strconcat(OpcodeStr,
3918 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3919 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3920 def Ymi : Ii8<0x70, MRMSrcMem,
3921 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3922 !strconcat(OpcodeStr,
3923 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3925 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3926 (i8 imm:$src2))))]>;
3928 } // ExeDomain = SSEPackedInt
3930 let Predicates = [HasAVX] in {
3931 let AddedComplexity = 5 in
3932 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
3934 // SSE2 with ImmT == Imm8 and XS prefix.
3935 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
3937 // SSE2 with ImmT == Imm8 and XD prefix.
3938 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
3940 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3941 (VPSHUFDmi addr:$src1, imm:$imm)>;
3942 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3943 (VPSHUFDri VR128:$src1, imm:$imm)>;
3946 let Predicates = [HasAVX2] in {
3947 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>,
3948 TB, OpSize, VEX,VEX_L;
3949 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>,
3951 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>,
3955 let Predicates = [UseSSE2] in {
3956 let AddedComplexity = 5 in
3957 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
3959 // SSE2 with ImmT == Imm8 and XS prefix.
3960 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
3962 // SSE2 with ImmT == Imm8 and XD prefix.
3963 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
3965 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3966 (PSHUFDmi addr:$src1, imm:$imm)>;
3967 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3968 (PSHUFDri VR128:$src1, imm:$imm)>;
3971 //===---------------------------------------------------------------------===//
3972 // SSE2 - Packed Integer Unpack Instructions
3973 //===---------------------------------------------------------------------===//
3975 let ExeDomain = SSEPackedInt in {
3976 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3977 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3978 def rr : PDI<opc, MRMSrcReg,
3979 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3981 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3982 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3983 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
3985 def rm : PDI<opc, MRMSrcMem,
3986 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3988 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3989 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3990 [(set VR128:$dst, (OpNode VR128:$src1,
3991 (bc_frag (memopv2i64
3996 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
3997 SDNode OpNode, PatFrag bc_frag> {
3998 def Yrr : PDI<opc, MRMSrcReg,
3999 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4000 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4001 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4002 def Yrm : PDI<opc, MRMSrcMem,
4003 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4004 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4005 [(set VR256:$dst, (OpNode VR256:$src1,
4006 (bc_frag (memopv4i64 addr:$src2))))]>;
4009 let Predicates = [HasAVX] in {
4010 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4011 bc_v16i8, 0>, VEX_4V;
4012 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4013 bc_v8i16, 0>, VEX_4V;
4014 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4015 bc_v4i32, 0>, VEX_4V;
4016 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4017 bc_v2i64, 0>, VEX_4V;
4019 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4020 bc_v16i8, 0>, VEX_4V;
4021 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4022 bc_v8i16, 0>, VEX_4V;
4023 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4024 bc_v4i32, 0>, VEX_4V;
4025 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4026 bc_v2i64, 0>, VEX_4V;
4029 let Predicates = [HasAVX2] in {
4030 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4031 bc_v32i8>, VEX_4V, VEX_L;
4032 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4033 bc_v16i16>, VEX_4V, VEX_L;
4034 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4035 bc_v8i32>, VEX_4V, VEX_L;
4036 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4037 bc_v4i64>, VEX_4V, VEX_L;
4039 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4040 bc_v32i8>, VEX_4V, VEX_L;
4041 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4042 bc_v16i16>, VEX_4V, VEX_L;
4043 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4044 bc_v8i32>, VEX_4V, VEX_L;
4045 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4046 bc_v4i64>, VEX_4V, VEX_L;
4049 let Constraints = "$src1 = $dst" in {
4050 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4052 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4054 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4056 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4059 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4061 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4063 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4065 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4068 } // ExeDomain = SSEPackedInt
4070 //===---------------------------------------------------------------------===//
4071 // SSE2 - Packed Integer Extract and Insert
4072 //===---------------------------------------------------------------------===//
4074 let ExeDomain = SSEPackedInt in {
4075 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4076 def rri : Ii8<0xC4, MRMSrcReg,
4077 (outs VR128:$dst), (ins VR128:$src1,
4078 GR32:$src2, i32i8imm:$src3),
4080 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4081 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4083 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4084 def rmi : Ii8<0xC4, MRMSrcMem,
4085 (outs VR128:$dst), (ins VR128:$src1,
4086 i16mem:$src2, i32i8imm:$src3),
4088 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4089 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4091 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4092 imm:$src3))], IIC_SSE_PINSRW>;
4096 let Predicates = [HasAVX] in
4097 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4098 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4099 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4100 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4101 imm:$src2))]>, TB, OpSize, VEX;
4102 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4103 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4104 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4105 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4106 imm:$src2))], IIC_SSE_PEXTRW>;
4109 let Predicates = [HasAVX] in {
4110 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4111 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4112 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4113 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4114 []>, TB, OpSize, VEX_4V;
4117 let Constraints = "$src1 = $dst" in
4118 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4120 } // ExeDomain = SSEPackedInt
4122 //===---------------------------------------------------------------------===//
4123 // SSE2 - Packed Mask Creation
4124 //===---------------------------------------------------------------------===//
4126 let ExeDomain = SSEPackedInt in {
4128 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4129 "pmovmskb\t{$src, $dst|$dst, $src}",
4130 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4131 IIC_SSE_MOVMSK>, VEX;
4132 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4133 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4135 let Predicates = [HasAVX2] in {
4136 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4137 "pmovmskb\t{$src, $dst|$dst, $src}",
4138 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4139 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4140 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4143 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4144 "pmovmskb\t{$src, $dst|$dst, $src}",
4145 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4148 } // ExeDomain = SSEPackedInt
4150 //===---------------------------------------------------------------------===//
4151 // SSE2 - Conditional Store
4152 //===---------------------------------------------------------------------===//
4154 let ExeDomain = SSEPackedInt in {
4157 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4158 (ins VR128:$src, VR128:$mask),
4159 "maskmovdqu\t{$mask, $src|$src, $mask}",
4160 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4161 IIC_SSE_MASKMOV>, VEX;
4163 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4164 (ins VR128:$src, VR128:$mask),
4165 "maskmovdqu\t{$mask, $src|$src, $mask}",
4166 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4167 IIC_SSE_MASKMOV>, VEX;
4170 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4171 "maskmovdqu\t{$mask, $src|$src, $mask}",
4172 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4175 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4176 "maskmovdqu\t{$mask, $src|$src, $mask}",
4177 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4180 } // ExeDomain = SSEPackedInt
4182 //===---------------------------------------------------------------------===//
4183 // SSE2 - Move Doubleword
4184 //===---------------------------------------------------------------------===//
4186 //===---------------------------------------------------------------------===//
4187 // Move Int Doubleword to Packed Double Int
4189 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4190 "movd\t{$src, $dst|$dst, $src}",
4192 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4194 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4195 "movd\t{$src, $dst|$dst, $src}",
4197 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4200 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4201 "mov{d|q}\t{$src, $dst|$dst, $src}",
4203 (v2i64 (scalar_to_vector GR64:$src)))],
4204 IIC_SSE_MOVDQ>, VEX;
4205 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4206 "mov{d|q}\t{$src, $dst|$dst, $src}",
4207 [(set FR64:$dst, (bitconvert GR64:$src))],
4208 IIC_SSE_MOVDQ>, VEX;
4210 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4211 "movd\t{$src, $dst|$dst, $src}",
4213 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4214 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4215 "movd\t{$src, $dst|$dst, $src}",
4217 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4219 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4220 "mov{d|q}\t{$src, $dst|$dst, $src}",
4222 (v2i64 (scalar_to_vector GR64:$src)))],
4224 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4225 "mov{d|q}\t{$src, $dst|$dst, $src}",
4226 [(set FR64:$dst, (bitconvert GR64:$src))],
4229 //===---------------------------------------------------------------------===//
4230 // Move Int Doubleword to Single Scalar
4232 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4233 "movd\t{$src, $dst|$dst, $src}",
4234 [(set FR32:$dst, (bitconvert GR32:$src))],
4235 IIC_SSE_MOVDQ>, VEX;
4237 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4238 "movd\t{$src, $dst|$dst, $src}",
4239 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4242 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4243 "movd\t{$src, $dst|$dst, $src}",
4244 [(set FR32:$dst, (bitconvert GR32:$src))],
4247 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4248 "movd\t{$src, $dst|$dst, $src}",
4249 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4252 //===---------------------------------------------------------------------===//
4253 // Move Packed Doubleword Int to Packed Double Int
4255 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4256 "movd\t{$src, $dst|$dst, $src}",
4257 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4258 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4259 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4260 (ins i32mem:$dst, VR128:$src),
4261 "movd\t{$src, $dst|$dst, $src}",
4262 [(store (i32 (vector_extract (v4i32 VR128:$src),
4263 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4265 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4266 "movd\t{$src, $dst|$dst, $src}",
4267 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4268 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4269 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4270 "movd\t{$src, $dst|$dst, $src}",
4271 [(store (i32 (vector_extract (v4i32 VR128:$src),
4272 (iPTR 0))), addr:$dst)],
4275 //===---------------------------------------------------------------------===//
4276 // Move Packed Doubleword Int first element to Doubleword Int
4278 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4279 "vmov{d|q}\t{$src, $dst|$dst, $src}",
4280 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4283 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4285 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4286 "mov{d|q}\t{$src, $dst|$dst, $src}",
4287 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4291 //===---------------------------------------------------------------------===//
4292 // Bitcast FR64 <-> GR64
4294 let Predicates = [HasAVX] in
4295 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4296 "vmovq\t{$src, $dst|$dst, $src}",
4297 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4299 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4300 "mov{d|q}\t{$src, $dst|$dst, $src}",
4301 [(set GR64:$dst, (bitconvert FR64:$src))],
4302 IIC_SSE_MOVDQ>, VEX;
4303 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4304 "movq\t{$src, $dst|$dst, $src}",
4305 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4306 IIC_SSE_MOVDQ>, VEX;
4308 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4309 "movq\t{$src, $dst|$dst, $src}",
4310 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4312 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4313 "mov{d|q}\t{$src, $dst|$dst, $src}",
4314 [(set GR64:$dst, (bitconvert FR64:$src))],
4316 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4317 "movq\t{$src, $dst|$dst, $src}",
4318 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4321 //===---------------------------------------------------------------------===//
4322 // Move Scalar Single to Double Int
4324 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4325 "movd\t{$src, $dst|$dst, $src}",
4326 [(set GR32:$dst, (bitconvert FR32:$src))],
4327 IIC_SSE_MOVD_ToGP>, VEX;
4328 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4329 "movd\t{$src, $dst|$dst, $src}",
4330 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4331 IIC_SSE_MOVDQ>, VEX;
4332 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4333 "movd\t{$src, $dst|$dst, $src}",
4334 [(set GR32:$dst, (bitconvert FR32:$src))],
4336 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4337 "movd\t{$src, $dst|$dst, $src}",
4338 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4341 //===---------------------------------------------------------------------===//
4342 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4344 let AddedComplexity = 15 in {
4345 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4346 "movd\t{$src, $dst|$dst, $src}",
4347 [(set VR128:$dst, (v4i32 (X86vzmovl
4348 (v4i32 (scalar_to_vector GR32:$src)))))],
4349 IIC_SSE_MOVDQ>, VEX;
4350 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4351 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4352 [(set VR128:$dst, (v2i64 (X86vzmovl
4353 (v2i64 (scalar_to_vector GR64:$src)))))],
4357 let AddedComplexity = 15 in {
4358 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4359 "movd\t{$src, $dst|$dst, $src}",
4360 [(set VR128:$dst, (v4i32 (X86vzmovl
4361 (v4i32 (scalar_to_vector GR32:$src)))))],
4363 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4364 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4365 [(set VR128:$dst, (v2i64 (X86vzmovl
4366 (v2i64 (scalar_to_vector GR64:$src)))))],
4370 let AddedComplexity = 20 in {
4371 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4372 "movd\t{$src, $dst|$dst, $src}",
4374 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4375 (loadi32 addr:$src))))))],
4376 IIC_SSE_MOVDQ>, VEX;
4377 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4378 "movd\t{$src, $dst|$dst, $src}",
4380 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4381 (loadi32 addr:$src))))))],
4385 let Predicates = [HasAVX] in {
4386 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4387 let AddedComplexity = 20 in {
4388 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4389 (VMOVZDI2PDIrm addr:$src)>;
4390 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4391 (VMOVZDI2PDIrm addr:$src)>;
4393 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4394 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4395 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4396 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4397 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4398 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4399 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4402 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4403 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4404 (MOVZDI2PDIrm addr:$src)>;
4405 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4406 (MOVZDI2PDIrm addr:$src)>;
4409 // These are the correct encodings of the instructions so that we know how to
4410 // read correct assembly, even though we continue to emit the wrong ones for
4411 // compatibility with Darwin's buggy assembler.
4412 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4413 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4414 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4415 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4416 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4417 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4418 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4419 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4420 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4421 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4422 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4423 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4425 //===---------------------------------------------------------------------===//
4426 // SSE2 - Move Quadword
4427 //===---------------------------------------------------------------------===//
4429 //===---------------------------------------------------------------------===//
4430 // Move Quadword Int to Packed Quadword Int
4432 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4433 "vmovq\t{$src, $dst|$dst, $src}",
4435 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4436 VEX, Requires<[HasAVX]>;
4437 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4438 "movq\t{$src, $dst|$dst, $src}",
4440 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4442 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4444 //===---------------------------------------------------------------------===//
4445 // Move Packed Quadword Int to Quadword Int
4447 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4448 "movq\t{$src, $dst|$dst, $src}",
4449 [(store (i64 (vector_extract (v2i64 VR128:$src),
4450 (iPTR 0))), addr:$dst)],
4451 IIC_SSE_MOVDQ>, VEX;
4452 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4453 "movq\t{$src, $dst|$dst, $src}",
4454 [(store (i64 (vector_extract (v2i64 VR128:$src),
4455 (iPTR 0))), addr:$dst)],
4458 //===---------------------------------------------------------------------===//
4459 // Store / copy lower 64-bits of a XMM register.
4461 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4462 "movq\t{$src, $dst|$dst, $src}",
4463 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4464 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4465 "movq\t{$src, $dst|$dst, $src}",
4466 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4469 let AddedComplexity = 20 in
4470 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4471 "vmovq\t{$src, $dst|$dst, $src}",
4473 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4474 (loadi64 addr:$src))))))],
4476 XS, VEX, Requires<[HasAVX]>;
4478 let AddedComplexity = 20 in
4479 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4480 "movq\t{$src, $dst|$dst, $src}",
4482 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4483 (loadi64 addr:$src))))))],
4485 XS, Requires<[UseSSE2]>;
4487 let Predicates = [HasAVX], AddedComplexity = 20 in {
4488 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4489 (VMOVZQI2PQIrm addr:$src)>;
4490 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4491 (VMOVZQI2PQIrm addr:$src)>;
4492 def : Pat<(v2i64 (X86vzload addr:$src)),
4493 (VMOVZQI2PQIrm addr:$src)>;
4496 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4497 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4498 (MOVZQI2PQIrm addr:$src)>;
4499 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4500 (MOVZQI2PQIrm addr:$src)>;
4501 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4504 let Predicates = [HasAVX] in {
4505 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4506 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4507 def : Pat<(v4i64 (X86vzload addr:$src)),
4508 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4511 //===---------------------------------------------------------------------===//
4512 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4513 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4515 let AddedComplexity = 15 in
4516 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4517 "vmovq\t{$src, $dst|$dst, $src}",
4518 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4520 XS, VEX, Requires<[HasAVX]>;
4521 let AddedComplexity = 15 in
4522 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4523 "movq\t{$src, $dst|$dst, $src}",
4524 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4526 XS, Requires<[UseSSE2]>;
4528 let AddedComplexity = 20 in
4529 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4530 "vmovq\t{$src, $dst|$dst, $src}",
4531 [(set VR128:$dst, (v2i64 (X86vzmovl
4532 (loadv2i64 addr:$src))))],
4534 XS, VEX, Requires<[HasAVX]>;
4535 let AddedComplexity = 20 in {
4536 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4537 "movq\t{$src, $dst|$dst, $src}",
4538 [(set VR128:$dst, (v2i64 (X86vzmovl
4539 (loadv2i64 addr:$src))))],
4541 XS, Requires<[UseSSE2]>;
4544 let AddedComplexity = 20 in {
4545 let Predicates = [HasAVX] in {
4546 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4547 (VMOVZPQILo2PQIrm addr:$src)>;
4548 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4549 (VMOVZPQILo2PQIrr VR128:$src)>;
4551 let Predicates = [UseSSE2] in {
4552 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4553 (MOVZPQILo2PQIrm addr:$src)>;
4554 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4555 (MOVZPQILo2PQIrr VR128:$src)>;
4559 // Instructions to match in the assembler
4560 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4561 "movq\t{$src, $dst|$dst, $src}", [],
4562 IIC_SSE_MOVDQ>, VEX, VEX_W;
4563 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4564 "movq\t{$src, $dst|$dst, $src}", [],
4565 IIC_SSE_MOVDQ>, VEX, VEX_W;
4566 // Recognize "movd" with GR64 destination, but encode as a "movq"
4567 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4568 "movd\t{$src, $dst|$dst, $src}", [],
4569 IIC_SSE_MOVDQ>, VEX, VEX_W;
4571 // Instructions for the disassembler
4572 // xr = XMM register
4575 let Predicates = [HasAVX] in
4576 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4577 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4578 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4579 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4581 //===---------------------------------------------------------------------===//
4582 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4583 //===---------------------------------------------------------------------===//
4584 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4585 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4586 X86MemOperand x86memop> {
4587 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4588 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4589 [(set RC:$dst, (vt (OpNode RC:$src)))],
4591 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4592 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4593 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4597 let Predicates = [HasAVX] in {
4598 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4599 v4f32, VR128, memopv4f32, f128mem>, VEX;
4600 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4601 v4f32, VR128, memopv4f32, f128mem>, VEX;
4602 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4603 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4604 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4605 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4607 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4608 memopv4f32, f128mem>;
4609 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4610 memopv4f32, f128mem>;
4612 let Predicates = [HasAVX] in {
4613 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4614 (VMOVSHDUPrr VR128:$src)>;
4615 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4616 (VMOVSHDUPrm addr:$src)>;
4617 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4618 (VMOVSLDUPrr VR128:$src)>;
4619 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4620 (VMOVSLDUPrm addr:$src)>;
4621 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4622 (VMOVSHDUPYrr VR256:$src)>;
4623 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4624 (VMOVSHDUPYrm addr:$src)>;
4625 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4626 (VMOVSLDUPYrr VR256:$src)>;
4627 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4628 (VMOVSLDUPYrm addr:$src)>;
4631 let Predicates = [UseSSE3] in {
4632 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4633 (MOVSHDUPrr VR128:$src)>;
4634 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4635 (MOVSHDUPrm addr:$src)>;
4636 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4637 (MOVSLDUPrr VR128:$src)>;
4638 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4639 (MOVSLDUPrm addr:$src)>;
4642 //===---------------------------------------------------------------------===//
4643 // SSE3 - Replicate Double FP - MOVDDUP
4644 //===---------------------------------------------------------------------===//
4646 multiclass sse3_replicate_dfp<string OpcodeStr> {
4647 let neverHasSideEffects = 1 in
4648 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4649 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4650 [], IIC_SSE_MOV_LH>;
4651 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4652 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4655 (scalar_to_vector (loadf64 addr:$src)))))],
4659 // FIXME: Merge with above classe when there're patterns for the ymm version
4660 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4661 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4662 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4663 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4664 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4665 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4668 (scalar_to_vector (loadf64 addr:$src)))))]>;
4671 let Predicates = [HasAVX] in {
4672 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4673 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4676 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4678 let Predicates = [HasAVX] in {
4679 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4680 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4681 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4682 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4683 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4684 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4685 def : Pat<(X86Movddup (bc_v2f64
4686 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4687 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4690 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4691 (VMOVDDUPYrm addr:$src)>;
4692 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4693 (VMOVDDUPYrm addr:$src)>;
4694 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4695 (VMOVDDUPYrm addr:$src)>;
4696 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4697 (VMOVDDUPYrr VR256:$src)>;
4700 let Predicates = [UseSSE3] in {
4701 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4702 (MOVDDUPrm addr:$src)>;
4703 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4704 (MOVDDUPrm addr:$src)>;
4705 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4706 (MOVDDUPrm addr:$src)>;
4707 def : Pat<(X86Movddup (bc_v2f64
4708 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4709 (MOVDDUPrm addr:$src)>;
4712 //===---------------------------------------------------------------------===//
4713 // SSE3 - Move Unaligned Integer
4714 //===---------------------------------------------------------------------===//
4716 let Predicates = [HasAVX] in {
4717 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4718 "vlddqu\t{$src, $dst|$dst, $src}",
4719 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4720 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4721 "vlddqu\t{$src, $dst|$dst, $src}",
4722 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4725 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4726 "lddqu\t{$src, $dst|$dst, $src}",
4727 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4730 //===---------------------------------------------------------------------===//
4731 // SSE3 - Arithmetic
4732 //===---------------------------------------------------------------------===//
4734 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4735 X86MemOperand x86memop, OpndItins itins,
4737 def rr : I<0xD0, MRMSrcReg,
4738 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4740 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4741 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4742 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4743 def rm : I<0xD0, MRMSrcMem,
4744 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4746 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4747 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4748 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4751 let Predicates = [HasAVX] in {
4752 let ExeDomain = SSEPackedSingle in {
4753 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4754 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4755 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4756 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4758 let ExeDomain = SSEPackedDouble in {
4759 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4760 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4761 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4762 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4765 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4766 let ExeDomain = SSEPackedSingle in
4767 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4768 f128mem, SSE_ALU_F32P>, TB, XD;
4769 let ExeDomain = SSEPackedDouble in
4770 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4771 f128mem, SSE_ALU_F64P>, TB, OpSize;
4774 //===---------------------------------------------------------------------===//
4775 // SSE3 Instructions
4776 //===---------------------------------------------------------------------===//
4779 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4780 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4781 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4783 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4784 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4785 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4787 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4789 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4790 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4791 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4792 IIC_SSE_HADDSUB_RM>;
4794 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4795 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4796 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4798 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4799 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4800 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4802 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4804 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4805 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4806 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4807 IIC_SSE_HADDSUB_RM>;
4810 let Predicates = [HasAVX] in {
4811 let ExeDomain = SSEPackedSingle in {
4812 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4813 X86fhadd, 0>, VEX_4V;
4814 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4815 X86fhsub, 0>, VEX_4V;
4816 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4817 X86fhadd, 0>, VEX_4V, VEX_L;
4818 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4819 X86fhsub, 0>, VEX_4V, VEX_L;
4821 let ExeDomain = SSEPackedDouble in {
4822 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4823 X86fhadd, 0>, VEX_4V;
4824 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4825 X86fhsub, 0>, VEX_4V;
4826 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4827 X86fhadd, 0>, VEX_4V, VEX_L;
4828 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4829 X86fhsub, 0>, VEX_4V, VEX_L;
4833 let Constraints = "$src1 = $dst" in {
4834 let ExeDomain = SSEPackedSingle in {
4835 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4836 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4838 let ExeDomain = SSEPackedDouble in {
4839 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4840 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4844 //===---------------------------------------------------------------------===//
4845 // SSSE3 - Packed Absolute Instructions
4846 //===---------------------------------------------------------------------===//
4849 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4850 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4851 Intrinsic IntId128> {
4852 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4854 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4855 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
4858 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4860 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4863 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
4867 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4868 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4869 Intrinsic IntId256> {
4870 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4872 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4873 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4876 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4878 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4881 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4884 let Predicates = [HasAVX] in {
4885 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4886 int_x86_ssse3_pabs_b_128>, VEX;
4887 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4888 int_x86_ssse3_pabs_w_128>, VEX;
4889 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4890 int_x86_ssse3_pabs_d_128>, VEX;
4893 let Predicates = [HasAVX2] in {
4894 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4895 int_x86_avx2_pabs_b>, VEX, VEX_L;
4896 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4897 int_x86_avx2_pabs_w>, VEX, VEX_L;
4898 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4899 int_x86_avx2_pabs_d>, VEX, VEX_L;
4902 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4903 int_x86_ssse3_pabs_b_128>;
4904 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4905 int_x86_ssse3_pabs_w_128>;
4906 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4907 int_x86_ssse3_pabs_d_128>;
4909 //===---------------------------------------------------------------------===//
4910 // SSSE3 - Packed Binary Operator Instructions
4911 //===---------------------------------------------------------------------===//
4913 def SSE_PHADDSUBD : OpndItins<
4914 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
4916 def SSE_PHADDSUBSW : OpndItins<
4917 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
4919 def SSE_PHADDSUBW : OpndItins<
4920 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
4922 def SSE_PSHUFB : OpndItins<
4923 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
4925 def SSE_PSIGN : OpndItins<
4926 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
4928 def SSE_PMULHRSW : OpndItins<
4929 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
4932 /// SS3I_binop_rm - Simple SSSE3 bin op
4933 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4934 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
4935 X86MemOperand x86memop, OpndItins itins,
4937 let isCommutable = 1 in
4938 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
4939 (ins RC:$src1, RC:$src2),
4941 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4942 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4943 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
4945 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
4946 (ins RC:$src1, x86memop:$src2),
4948 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4949 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4951 (OpVT (OpNode RC:$src1,
4952 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
4955 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4956 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4957 Intrinsic IntId128, OpndItins itins,
4959 let isCommutable = 1 in
4960 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4961 (ins VR128:$src1, VR128:$src2),
4963 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4964 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4965 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4967 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4968 (ins VR128:$src1, i128mem:$src2),
4970 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4971 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4973 (IntId128 VR128:$src1,
4974 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
4977 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
4978 Intrinsic IntId256> {
4979 let isCommutable = 1 in
4980 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4981 (ins VR256:$src1, VR256:$src2),
4982 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4983 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
4985 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4986 (ins VR256:$src1, i256mem:$src2),
4987 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4989 (IntId256 VR256:$src1,
4990 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
4993 let ImmT = NoImm, Predicates = [HasAVX] in {
4994 let isCommutable = 0 in {
4995 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
4996 memopv2i64, i128mem,
4997 SSE_PHADDSUBW, 0>, VEX_4V;
4998 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
4999 memopv2i64, i128mem,
5000 SSE_PHADDSUBD, 0>, VEX_4V;
5001 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5002 memopv2i64, i128mem,
5003 SSE_PHADDSUBW, 0>, VEX_4V;
5004 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5005 memopv2i64, i128mem,
5006 SSE_PHADDSUBD, 0>, VEX_4V;
5007 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5008 memopv2i64, i128mem,
5009 SSE_PSIGN, 0>, VEX_4V;
5010 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5011 memopv2i64, i128mem,
5012 SSE_PSIGN, 0>, VEX_4V;
5013 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5014 memopv2i64, i128mem,
5015 SSE_PSIGN, 0>, VEX_4V;
5016 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5017 memopv2i64, i128mem,
5018 SSE_PSHUFB, 0>, VEX_4V;
5019 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5020 int_x86_ssse3_phadd_sw_128,
5021 SSE_PHADDSUBSW, 0>, VEX_4V;
5022 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5023 int_x86_ssse3_phsub_sw_128,
5024 SSE_PHADDSUBSW, 0>, VEX_4V;
5025 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5026 int_x86_ssse3_pmadd_ub_sw_128,
5027 SSE_PMADD, 0>, VEX_4V;
5029 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5030 int_x86_ssse3_pmul_hr_sw_128,
5031 SSE_PMULHRSW, 0>, VEX_4V;
5034 let ImmT = NoImm, Predicates = [HasAVX2] in {
5035 let isCommutable = 0 in {
5036 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5037 memopv4i64, i256mem,
5038 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5039 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5040 memopv4i64, i256mem,
5041 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5042 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5043 memopv4i64, i256mem,
5044 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5045 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5046 memopv4i64, i256mem,
5047 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5048 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5049 memopv4i64, i256mem,
5050 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5051 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5052 memopv4i64, i256mem,
5053 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5054 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5055 memopv4i64, i256mem,
5056 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5057 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5058 memopv4i64, i256mem,
5059 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5060 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5061 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5062 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5063 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5064 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5065 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5067 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5068 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5071 // None of these have i8 immediate fields.
5072 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5073 let isCommutable = 0 in {
5074 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5075 memopv2i64, i128mem, SSE_PHADDSUBW>;
5076 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5077 memopv2i64, i128mem, SSE_PHADDSUBD>;
5078 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5079 memopv2i64, i128mem, SSE_PHADDSUBW>;
5080 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5081 memopv2i64, i128mem, SSE_PHADDSUBD>;
5082 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5083 memopv2i64, i128mem, SSE_PSIGN>;
5084 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5085 memopv2i64, i128mem, SSE_PSIGN>;
5086 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5087 memopv2i64, i128mem, SSE_PSIGN>;
5088 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5089 memopv2i64, i128mem, SSE_PSHUFB>;
5090 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5091 int_x86_ssse3_phadd_sw_128,
5093 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5094 int_x86_ssse3_phsub_sw_128,
5096 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5097 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5099 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5100 int_x86_ssse3_pmul_hr_sw_128,
5104 //===---------------------------------------------------------------------===//
5105 // SSSE3 - Packed Align Instruction Patterns
5106 //===---------------------------------------------------------------------===//
5108 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5109 let neverHasSideEffects = 1 in {
5110 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5111 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5113 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5115 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5116 [], IIC_SSE_PALIGNR>, OpSize;
5118 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5119 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5121 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5123 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5124 [], IIC_SSE_PALIGNR>, OpSize;
5128 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5129 let neverHasSideEffects = 1 in {
5130 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5131 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5133 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5136 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5137 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5139 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5144 let Predicates = [HasAVX] in
5145 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5146 let Predicates = [HasAVX2] in
5147 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
5148 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5149 defm PALIGN : ssse3_palign<"palignr">;
5151 let Predicates = [HasAVX2] in {
5152 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5153 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5154 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5155 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5156 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5157 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5158 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5159 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5162 let Predicates = [HasAVX] in {
5163 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5164 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5165 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5166 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5167 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5168 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5169 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5170 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5173 let Predicates = [UseSSSE3] in {
5174 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5175 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5176 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5177 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5178 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5179 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5180 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5181 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5184 //===---------------------------------------------------------------------===//
5185 // SSSE3 - Thread synchronization
5186 //===---------------------------------------------------------------------===//
5188 let usesCustomInserter = 1 in {
5189 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5190 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5191 Requires<[HasSSE3]>;
5194 let Uses = [EAX, ECX, EDX] in
5195 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5196 TB, Requires<[HasSSE3]>;
5197 let Uses = [ECX, EAX] in
5198 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5199 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5200 TB, Requires<[HasSSE3]>;
5202 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5203 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5205 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5206 Requires<[In32BitMode]>;
5207 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5208 Requires<[In64BitMode]>;
5210 //===----------------------------------------------------------------------===//
5211 // SSE4.1 - Packed Move with Sign/Zero Extend
5212 //===----------------------------------------------------------------------===//
5214 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5215 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5216 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5217 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5219 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5220 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5222 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5226 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5228 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5229 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5230 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5232 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5233 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5234 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5237 let Predicates = [HasAVX] in {
5238 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5240 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5242 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5244 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5246 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5248 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5252 let Predicates = [HasAVX2] in {
5253 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5254 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5255 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5256 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5257 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5258 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5259 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5260 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5261 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5262 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5263 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5264 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5267 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5268 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5269 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5270 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5271 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5272 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5274 let Predicates = [HasAVX] in {
5275 // Common patterns involving scalar load.
5276 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5277 (VPMOVSXBWrm addr:$src)>;
5278 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5279 (VPMOVSXBWrm addr:$src)>;
5280 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5281 (VPMOVSXBWrm addr:$src)>;
5283 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5284 (VPMOVSXWDrm addr:$src)>;
5285 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5286 (VPMOVSXWDrm addr:$src)>;
5287 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5288 (VPMOVSXWDrm addr:$src)>;
5290 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5291 (VPMOVSXDQrm addr:$src)>;
5292 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5293 (VPMOVSXDQrm addr:$src)>;
5294 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5295 (VPMOVSXDQrm addr:$src)>;
5297 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5298 (VPMOVZXBWrm addr:$src)>;
5299 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5300 (VPMOVZXBWrm addr:$src)>;
5301 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5302 (VPMOVZXBWrm addr:$src)>;
5304 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5305 (VPMOVZXWDrm addr:$src)>;
5306 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5307 (VPMOVZXWDrm addr:$src)>;
5308 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5309 (VPMOVZXWDrm addr:$src)>;
5311 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5312 (VPMOVZXDQrm addr:$src)>;
5313 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5314 (VPMOVZXDQrm addr:$src)>;
5315 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5316 (VPMOVZXDQrm addr:$src)>;
5319 let Predicates = [UseSSE41] in {
5320 // Common patterns involving scalar load.
5321 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5322 (PMOVSXBWrm addr:$src)>;
5323 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5324 (PMOVSXBWrm addr:$src)>;
5325 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5326 (PMOVSXBWrm addr:$src)>;
5328 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5329 (PMOVSXWDrm addr:$src)>;
5330 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5331 (PMOVSXWDrm addr:$src)>;
5332 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5333 (PMOVSXWDrm addr:$src)>;
5335 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5336 (PMOVSXDQrm addr:$src)>;
5337 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5338 (PMOVSXDQrm addr:$src)>;
5339 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5340 (PMOVSXDQrm addr:$src)>;
5342 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5343 (PMOVZXBWrm addr:$src)>;
5344 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5345 (PMOVZXBWrm addr:$src)>;
5346 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5347 (PMOVZXBWrm addr:$src)>;
5349 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5350 (PMOVZXWDrm addr:$src)>;
5351 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5352 (PMOVZXWDrm addr:$src)>;
5353 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5354 (PMOVZXWDrm addr:$src)>;
5356 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5357 (PMOVZXDQrm addr:$src)>;
5358 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5359 (PMOVZXDQrm addr:$src)>;
5360 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5361 (PMOVZXDQrm addr:$src)>;
5364 let Predicates = [HasAVX2] in {
5365 let AddedComplexity = 15 in {
5366 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5367 (VPMOVZXDQYrr VR128:$src)>;
5368 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5369 (VPMOVZXWDYrr VR128:$src)>;
5372 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5373 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5376 let Predicates = [HasAVX] in {
5377 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5378 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5381 let Predicates = [UseSSE41] in {
5382 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5383 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5387 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5388 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5389 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5390 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5392 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5393 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5395 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5399 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5401 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5402 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5403 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5405 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5406 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5408 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5412 let Predicates = [HasAVX] in {
5413 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5415 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5417 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5419 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5423 let Predicates = [HasAVX2] in {
5424 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5425 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5426 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5427 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5428 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5429 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5430 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5431 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5434 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5435 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5436 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5437 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5439 let Predicates = [HasAVX] in {
5440 // Common patterns involving scalar load
5441 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5442 (VPMOVSXBDrm addr:$src)>;
5443 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5444 (VPMOVSXWQrm addr:$src)>;
5446 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5447 (VPMOVZXBDrm addr:$src)>;
5448 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5449 (VPMOVZXWQrm addr:$src)>;
5452 let Predicates = [UseSSE41] in {
5453 // Common patterns involving scalar load
5454 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5455 (PMOVSXBDrm addr:$src)>;
5456 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5457 (PMOVSXWQrm addr:$src)>;
5459 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5460 (PMOVZXBDrm addr:$src)>;
5461 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5462 (PMOVZXWQrm addr:$src)>;
5465 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5466 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5467 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5468 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5470 // Expecting a i16 load any extended to i32 value.
5471 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5472 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5473 [(set VR128:$dst, (IntId (bitconvert
5474 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5478 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5480 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5481 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5482 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5484 // Expecting a i16 load any extended to i32 value.
5485 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5486 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5487 [(set VR256:$dst, (IntId (bitconvert
5488 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5492 let Predicates = [HasAVX] in {
5493 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5495 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5498 let Predicates = [HasAVX2] in {
5499 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5500 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5501 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5502 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5504 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5505 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5507 let Predicates = [HasAVX2] in {
5508 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5509 (VPMOVSXWDYrm addr:$src)>;
5510 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5511 (VPMOVSXDQYrm addr:$src)>;
5513 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5514 (scalar_to_vector (loadi64 addr:$src))))))),
5515 (VPMOVSXBDYrm addr:$src)>;
5516 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5517 (scalar_to_vector (loadf64 addr:$src))))))),
5518 (VPMOVSXBDYrm addr:$src)>;
5520 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5521 (scalar_to_vector (loadi64 addr:$src))))))),
5522 (VPMOVSXWQYrm addr:$src)>;
5523 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5524 (scalar_to_vector (loadf64 addr:$src))))))),
5525 (VPMOVSXWQYrm addr:$src)>;
5527 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5528 (scalar_to_vector (loadi32 addr:$src))))))),
5529 (VPMOVSXBQYrm addr:$src)>;
5532 let Predicates = [HasAVX] in {
5533 // Common patterns involving scalar load
5534 def : Pat<(int_x86_sse41_pmovsxbq
5535 (bitconvert (v4i32 (X86vzmovl
5536 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5537 (VPMOVSXBQrm addr:$src)>;
5539 def : Pat<(int_x86_sse41_pmovzxbq
5540 (bitconvert (v4i32 (X86vzmovl
5541 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5542 (VPMOVZXBQrm addr:$src)>;
5545 let Predicates = [UseSSE41] in {
5546 // Common patterns involving scalar load
5547 def : Pat<(int_x86_sse41_pmovsxbq
5548 (bitconvert (v4i32 (X86vzmovl
5549 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5550 (PMOVSXBQrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovzxbq
5553 (bitconvert (v4i32 (X86vzmovl
5554 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5555 (PMOVZXBQrm addr:$src)>;
5557 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5558 (scalar_to_vector (loadi64 addr:$src))))))),
5559 (PMOVSXWDrm addr:$src)>;
5560 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5561 (scalar_to_vector (loadf64 addr:$src))))))),
5562 (PMOVSXWDrm addr:$src)>;
5563 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5564 (scalar_to_vector (loadi32 addr:$src))))))),
5565 (PMOVSXBDrm addr:$src)>;
5566 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5567 (scalar_to_vector (loadi32 addr:$src))))))),
5568 (PMOVSXWQrm addr:$src)>;
5569 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5570 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5571 (PMOVSXBQrm addr:$src)>;
5572 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5573 (scalar_to_vector (loadi64 addr:$src))))))),
5574 (PMOVSXDQrm addr:$src)>;
5575 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5576 (scalar_to_vector (loadf64 addr:$src))))))),
5577 (PMOVSXDQrm addr:$src)>;
5578 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5579 (scalar_to_vector (loadi64 addr:$src))))))),
5580 (PMOVSXBWrm addr:$src)>;
5581 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5582 (scalar_to_vector (loadf64 addr:$src))))))),
5583 (PMOVSXBWrm addr:$src)>;
5586 let Predicates = [HasAVX2] in {
5587 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5588 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5589 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5591 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5592 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5594 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5596 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5597 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5598 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5599 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5600 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5601 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5603 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5604 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5605 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5606 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5608 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5609 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5612 let Predicates = [HasAVX] in {
5613 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5614 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5615 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5617 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5618 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5620 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5622 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5623 (VPMOVZXBWrm addr:$src)>;
5624 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5625 (VPMOVZXBWrm addr:$src)>;
5626 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5627 (VPMOVZXBDrm addr:$src)>;
5628 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5629 (VPMOVZXBQrm addr:$src)>;
5631 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5632 (VPMOVZXWDrm addr:$src)>;
5633 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5634 (VPMOVZXWDrm addr:$src)>;
5635 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5636 (VPMOVZXWQrm addr:$src)>;
5638 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5639 (VPMOVZXDQrm addr:$src)>;
5640 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5641 (VPMOVZXDQrm addr:$src)>;
5642 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5643 (VPMOVZXDQrm addr:$src)>;
5645 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5646 (scalar_to_vector (loadi64 addr:$src))))))),
5647 (VPMOVSXWDrm addr:$src)>;
5648 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5649 (scalar_to_vector (loadi64 addr:$src))))))),
5650 (VPMOVSXDQrm addr:$src)>;
5651 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5652 (scalar_to_vector (loadf64 addr:$src))))))),
5653 (VPMOVSXWDrm addr:$src)>;
5654 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5655 (scalar_to_vector (loadf64 addr:$src))))))),
5656 (VPMOVSXDQrm addr:$src)>;
5657 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5658 (scalar_to_vector (loadi64 addr:$src))))))),
5659 (VPMOVSXBWrm addr:$src)>;
5660 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5661 (scalar_to_vector (loadf64 addr:$src))))))),
5662 (VPMOVSXBWrm addr:$src)>;
5664 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5665 (scalar_to_vector (loadi32 addr:$src))))))),
5666 (VPMOVSXBDrm addr:$src)>;
5667 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5668 (scalar_to_vector (loadi32 addr:$src))))))),
5669 (VPMOVSXWQrm addr:$src)>;
5670 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5671 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5672 (VPMOVSXBQrm addr:$src)>;
5675 let Predicates = [UseSSE41] in {
5676 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5677 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5678 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5680 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5681 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5683 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5685 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5686 (PMOVZXBWrm addr:$src)>;
5687 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5688 (PMOVZXBWrm addr:$src)>;
5689 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5690 (PMOVZXBDrm addr:$src)>;
5691 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5692 (PMOVZXBQrm addr:$src)>;
5694 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5695 (PMOVZXWDrm addr:$src)>;
5696 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5697 (PMOVZXWDrm addr:$src)>;
5698 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5699 (PMOVZXWQrm addr:$src)>;
5701 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5702 (PMOVZXDQrm addr:$src)>;
5703 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5704 (PMOVZXDQrm addr:$src)>;
5705 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5706 (PMOVZXDQrm addr:$src)>;
5709 //===----------------------------------------------------------------------===//
5710 // SSE4.1 - Extract Instructions
5711 //===----------------------------------------------------------------------===//
5713 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5714 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5715 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5716 (ins VR128:$src1, i32i8imm:$src2),
5717 !strconcat(OpcodeStr,
5718 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5719 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5721 let neverHasSideEffects = 1, mayStore = 1 in
5722 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5723 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5724 !strconcat(OpcodeStr,
5725 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5728 // There's an AssertZext in the way of writing the store pattern
5729 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5732 let Predicates = [HasAVX] in {
5733 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5734 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5735 (ins VR128:$src1, i32i8imm:$src2),
5736 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5739 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5742 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5743 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5744 let neverHasSideEffects = 1, mayStore = 1 in
5745 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5746 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5747 !strconcat(OpcodeStr,
5748 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5751 // There's an AssertZext in the way of writing the store pattern
5752 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5755 let Predicates = [HasAVX] in
5756 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5758 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5761 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5762 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5763 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5764 (ins VR128:$src1, i32i8imm:$src2),
5765 !strconcat(OpcodeStr,
5766 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5768 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5769 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5770 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5771 !strconcat(OpcodeStr,
5772 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5773 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5774 addr:$dst)]>, OpSize;
5777 let Predicates = [HasAVX] in
5778 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5780 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5782 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5783 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5784 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5785 (ins VR128:$src1, i32i8imm:$src2),
5786 !strconcat(OpcodeStr,
5787 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5789 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5790 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5791 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5792 !strconcat(OpcodeStr,
5793 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5794 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5795 addr:$dst)]>, OpSize, REX_W;
5798 let Predicates = [HasAVX] in
5799 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5801 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5803 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5805 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5806 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5807 (ins VR128:$src1, i32i8imm:$src2),
5808 !strconcat(OpcodeStr,
5809 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5811 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5813 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5814 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5815 !strconcat(OpcodeStr,
5816 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5817 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5818 addr:$dst)]>, OpSize;
5821 let ExeDomain = SSEPackedSingle in {
5822 let Predicates = [HasAVX] in {
5823 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5824 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5825 (ins VR128:$src1, i32i8imm:$src2),
5826 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5829 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5832 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5833 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5836 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5838 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5841 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5842 Requires<[UseSSE41]>;
5844 //===----------------------------------------------------------------------===//
5845 // SSE4.1 - Insert Instructions
5846 //===----------------------------------------------------------------------===//
5848 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5849 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5850 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5852 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5854 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5856 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5857 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5858 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5860 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5862 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5864 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5865 imm:$src3))]>, OpSize;
5868 let Predicates = [HasAVX] in
5869 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5870 let Constraints = "$src1 = $dst" in
5871 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5873 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5874 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5875 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5877 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5879 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5881 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5883 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5884 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5886 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5888 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5890 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5891 imm:$src3)))]>, OpSize;
5894 let Predicates = [HasAVX] in
5895 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5896 let Constraints = "$src1 = $dst" in
5897 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5899 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5900 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5901 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5903 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5905 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5907 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5909 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5910 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5912 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5914 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5916 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5917 imm:$src3)))]>, OpSize;
5920 let Predicates = [HasAVX] in
5921 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5922 let Constraints = "$src1 = $dst" in
5923 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5925 // insertps has a few different modes, there's the first two here below which
5926 // are optimized inserts that won't zero arbitrary elements in the destination
5927 // vector. The next one matches the intrinsic and could zero arbitrary elements
5928 // in the target vector.
5929 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5930 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5931 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5933 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5935 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5937 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5939 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5940 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5942 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5944 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5946 (X86insrtps VR128:$src1,
5947 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5948 imm:$src3))]>, OpSize;
5951 let ExeDomain = SSEPackedSingle in {
5952 let Predicates = [HasAVX] in
5953 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5954 let Constraints = "$src1 = $dst" in
5955 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5958 //===----------------------------------------------------------------------===//
5959 // SSE4.1 - Round Instructions
5960 //===----------------------------------------------------------------------===//
5962 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5963 X86MemOperand x86memop, RegisterClass RC,
5964 PatFrag mem_frag32, PatFrag mem_frag64,
5965 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5966 let ExeDomain = SSEPackedSingle in {
5967 // Intrinsic operation, reg.
5968 // Vector intrinsic operation, reg
5969 def PSr : SS4AIi8<opcps, MRMSrcReg,
5970 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5971 !strconcat(OpcodeStr,
5972 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5973 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5976 // Vector intrinsic operation, mem
5977 def PSm : SS4AIi8<opcps, MRMSrcMem,
5978 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5979 !strconcat(OpcodeStr,
5980 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5982 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5984 } // ExeDomain = SSEPackedSingle
5986 let ExeDomain = SSEPackedDouble in {
5987 // Vector intrinsic operation, reg
5988 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5989 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5990 !strconcat(OpcodeStr,
5991 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5992 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5995 // Vector intrinsic operation, mem
5996 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5997 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5998 !strconcat(OpcodeStr,
5999 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6001 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6003 } // ExeDomain = SSEPackedDouble
6006 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6009 Intrinsic F64Int, bit Is2Addr = 1> {
6010 let ExeDomain = GenericDomain in {
6012 let hasSideEffects = 0 in
6013 def SSr : SS4AIi8<opcss, MRMSrcReg,
6014 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6016 !strconcat(OpcodeStr,
6017 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6018 !strconcat(OpcodeStr,
6019 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6022 // Intrinsic operation, reg.
6023 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6024 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6026 !strconcat(OpcodeStr,
6027 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6028 !strconcat(OpcodeStr,
6029 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6030 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6033 // Intrinsic operation, mem.
6034 def SSm : SS4AIi8<opcss, MRMSrcMem,
6035 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6037 !strconcat(OpcodeStr,
6038 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6039 !strconcat(OpcodeStr,
6040 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6042 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6046 let hasSideEffects = 0 in
6047 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6048 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6050 !strconcat(OpcodeStr,
6051 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6052 !strconcat(OpcodeStr,
6053 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6056 // Intrinsic operation, reg.
6057 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6058 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6060 !strconcat(OpcodeStr,
6061 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6062 !strconcat(OpcodeStr,
6063 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6064 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6067 // Intrinsic operation, mem.
6068 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6069 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6071 !strconcat(OpcodeStr,
6072 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6073 !strconcat(OpcodeStr,
6074 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6076 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6078 } // ExeDomain = GenericDomain
6081 // FP round - roundss, roundps, roundsd, roundpd
6082 let Predicates = [HasAVX] in {
6084 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6085 memopv4f32, memopv2f64,
6086 int_x86_sse41_round_ps,
6087 int_x86_sse41_round_pd>, VEX;
6088 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6089 memopv8f32, memopv4f64,
6090 int_x86_avx_round_ps_256,
6091 int_x86_avx_round_pd_256>, VEX, VEX_L;
6092 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6093 int_x86_sse41_round_ss,
6094 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6096 def : Pat<(ffloor FR32:$src),
6097 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6098 def : Pat<(f64 (ffloor FR64:$src)),
6099 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6100 def : Pat<(f32 (fnearbyint FR32:$src)),
6101 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6102 def : Pat<(f64 (fnearbyint FR64:$src)),
6103 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6104 def : Pat<(f32 (fceil FR32:$src)),
6105 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6106 def : Pat<(f64 (fceil FR64:$src)),
6107 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6108 def : Pat<(f32 (frint FR32:$src)),
6109 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6110 def : Pat<(f64 (frint FR64:$src)),
6111 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6112 def : Pat<(f32 (ftrunc FR32:$src)),
6113 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6114 def : Pat<(f64 (ftrunc FR64:$src)),
6115 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6117 def : Pat<(v4f32 (ffloor VR128:$src)),
6118 (VROUNDPSr VR128:$src, (i32 0x1))>;
6119 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6120 (VROUNDPSr VR128:$src, (i32 0xC))>;
6121 def : Pat<(v4f32 (fceil VR128:$src)),
6122 (VROUNDPSr VR128:$src, (i32 0x2))>;
6123 def : Pat<(v4f32 (frint VR128:$src)),
6124 (VROUNDPSr VR128:$src, (i32 0x4))>;
6125 def : Pat<(v4f32 (ftrunc VR128:$src)),
6126 (VROUNDPSr VR128:$src, (i32 0x3))>;
6128 def : Pat<(v2f64 (ffloor VR128:$src)),
6129 (VROUNDPDr VR128:$src, (i32 0x1))>;
6130 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6131 (VROUNDPDr VR128:$src, (i32 0xC))>;
6132 def : Pat<(v2f64 (fceil VR128:$src)),
6133 (VROUNDPDr VR128:$src, (i32 0x2))>;
6134 def : Pat<(v2f64 (frint VR128:$src)),
6135 (VROUNDPDr VR128:$src, (i32 0x4))>;
6136 def : Pat<(v2f64 (ftrunc VR128:$src)),
6137 (VROUNDPDr VR128:$src, (i32 0x3))>;
6139 def : Pat<(v8f32 (ffloor VR256:$src)),
6140 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6141 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6142 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6143 def : Pat<(v8f32 (fceil VR256:$src)),
6144 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6145 def : Pat<(v8f32 (frint VR256:$src)),
6146 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6147 def : Pat<(v8f32 (ftrunc VR256:$src)),
6148 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6150 def : Pat<(v4f64 (ffloor VR256:$src)),
6151 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6152 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6153 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6154 def : Pat<(v4f64 (fceil VR256:$src)),
6155 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6156 def : Pat<(v4f64 (frint VR256:$src)),
6157 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6158 def : Pat<(v4f64 (ftrunc VR256:$src)),
6159 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6162 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6163 memopv4f32, memopv2f64,
6164 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6165 let Constraints = "$src1 = $dst" in
6166 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6167 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6169 let Predicates = [UseSSE41] in {
6170 def : Pat<(ffloor FR32:$src),
6171 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6172 def : Pat<(f64 (ffloor FR64:$src)),
6173 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6174 def : Pat<(f32 (fnearbyint FR32:$src)),
6175 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6176 def : Pat<(f64 (fnearbyint FR64:$src)),
6177 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6178 def : Pat<(f32 (fceil FR32:$src)),
6179 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6180 def : Pat<(f64 (fceil FR64:$src)),
6181 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6182 def : Pat<(f32 (frint FR32:$src)),
6183 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6184 def : Pat<(f64 (frint FR64:$src)),
6185 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6186 def : Pat<(f32 (ftrunc FR32:$src)),
6187 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6188 def : Pat<(f64 (ftrunc FR64:$src)),
6189 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6191 def : Pat<(v4f32 (ffloor VR128:$src)),
6192 (ROUNDPSr VR128:$src, (i32 0x1))>;
6193 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6194 (ROUNDPSr VR128:$src, (i32 0xC))>;
6195 def : Pat<(v4f32 (fceil VR128:$src)),
6196 (ROUNDPSr VR128:$src, (i32 0x2))>;
6197 def : Pat<(v4f32 (frint VR128:$src)),
6198 (ROUNDPSr VR128:$src, (i32 0x4))>;
6199 def : Pat<(v4f32 (ftrunc VR128:$src)),
6200 (ROUNDPSr VR128:$src, (i32 0x3))>;
6202 def : Pat<(v2f64 (ffloor VR128:$src)),
6203 (ROUNDPDr VR128:$src, (i32 0x1))>;
6204 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6205 (ROUNDPDr VR128:$src, (i32 0xC))>;
6206 def : Pat<(v2f64 (fceil VR128:$src)),
6207 (ROUNDPDr VR128:$src, (i32 0x2))>;
6208 def : Pat<(v2f64 (frint VR128:$src)),
6209 (ROUNDPDr VR128:$src, (i32 0x4))>;
6210 def : Pat<(v2f64 (ftrunc VR128:$src)),
6211 (ROUNDPDr VR128:$src, (i32 0x3))>;
6214 //===----------------------------------------------------------------------===//
6215 // SSE4.1 - Packed Bit Test
6216 //===----------------------------------------------------------------------===//
6218 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6219 // the intel intrinsic that corresponds to this.
6220 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6221 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6222 "vptest\t{$src2, $src1|$src1, $src2}",
6223 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6225 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6226 "vptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6230 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6231 "vptest\t{$src2, $src1|$src1, $src2}",
6232 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6234 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6235 "vptest\t{$src2, $src1|$src1, $src2}",
6236 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6240 let Defs = [EFLAGS] in {
6241 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6242 "ptest\t{$src2, $src1|$src1, $src2}",
6243 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6245 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6246 "ptest\t{$src2, $src1|$src1, $src2}",
6247 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6251 // The bit test instructions below are AVX only
6252 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6253 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6254 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6255 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6256 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6257 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6258 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6259 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6263 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6264 let ExeDomain = SSEPackedSingle in {
6265 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6266 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6269 let ExeDomain = SSEPackedDouble in {
6270 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6271 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6276 //===----------------------------------------------------------------------===//
6277 // SSE4.1 - Misc Instructions
6278 //===----------------------------------------------------------------------===//
6280 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6281 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6282 "popcnt{w}\t{$src, $dst|$dst, $src}",
6283 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6285 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6286 "popcnt{w}\t{$src, $dst|$dst, $src}",
6287 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6288 (implicit EFLAGS)]>, OpSize, XS;
6290 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6291 "popcnt{l}\t{$src, $dst|$dst, $src}",
6292 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6294 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6295 "popcnt{l}\t{$src, $dst|$dst, $src}",
6296 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6297 (implicit EFLAGS)]>, XS;
6299 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6300 "popcnt{q}\t{$src, $dst|$dst, $src}",
6301 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6303 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6304 "popcnt{q}\t{$src, $dst|$dst, $src}",
6305 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6306 (implicit EFLAGS)]>, XS;
6311 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6312 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6313 Intrinsic IntId128> {
6314 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6316 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6317 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6318 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6320 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6323 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6326 let Predicates = [HasAVX] in
6327 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6328 int_x86_sse41_phminposuw>, VEX;
6329 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6330 int_x86_sse41_phminposuw>;
6332 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6333 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6334 Intrinsic IntId128, bit Is2Addr = 1> {
6335 let isCommutable = 1 in
6336 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6337 (ins VR128:$src1, VR128:$src2),
6339 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6341 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6342 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6343 (ins VR128:$src1, i128mem:$src2),
6345 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6346 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6348 (IntId128 VR128:$src1,
6349 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6352 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6353 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6354 Intrinsic IntId256> {
6355 let isCommutable = 1 in
6356 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6357 (ins VR256:$src1, VR256:$src2),
6358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6359 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6360 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6361 (ins VR256:$src1, i256mem:$src2),
6362 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6364 (IntId256 VR256:$src1,
6365 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6369 /// SS48I_binop_rm - Simple SSE41 binary operator.
6370 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6371 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6372 X86MemOperand x86memop, bit Is2Addr = 1> {
6373 let isCommutable = 1 in
6374 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6375 (ins RC:$src1, RC:$src2),
6377 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6379 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6380 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6381 (ins RC:$src1, x86memop:$src2),
6383 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6384 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6386 (OpVT (OpNode RC:$src1,
6387 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6390 let Predicates = [HasAVX] in {
6391 let isCommutable = 0 in
6392 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6394 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6395 memopv2i64, i128mem, 0>, VEX_4V;
6396 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6397 memopv2i64, i128mem, 0>, VEX_4V;
6398 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6399 memopv2i64, i128mem, 0>, VEX_4V;
6400 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6401 memopv2i64, i128mem, 0>, VEX_4V;
6402 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6403 memopv2i64, i128mem, 0>, VEX_4V;
6404 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6405 memopv2i64, i128mem, 0>, VEX_4V;
6406 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6407 memopv2i64, i128mem, 0>, VEX_4V;
6408 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6409 memopv2i64, i128mem, 0>, VEX_4V;
6410 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6414 let Predicates = [HasAVX2] in {
6415 let isCommutable = 0 in
6416 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6417 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6418 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6419 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6420 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6421 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6422 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6423 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6424 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6425 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6426 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6427 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6428 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6429 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6430 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6431 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6432 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6433 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6434 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6435 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6438 let Constraints = "$src1 = $dst" in {
6439 let isCommutable = 0 in
6440 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6441 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6442 memopv2i64, i128mem>;
6443 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6444 memopv2i64, i128mem>;
6445 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6446 memopv2i64, i128mem>;
6447 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6448 memopv2i64, i128mem>;
6449 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6450 memopv2i64, i128mem>;
6451 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6452 memopv2i64, i128mem>;
6453 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6454 memopv2i64, i128mem>;
6455 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6456 memopv2i64, i128mem>;
6457 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6460 let Predicates = [HasAVX] in {
6461 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6462 memopv2i64, i128mem, 0>, VEX_4V;
6463 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6464 memopv2i64, i128mem, 0>, VEX_4V;
6466 let Predicates = [HasAVX2] in {
6467 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6468 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6469 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6470 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6473 let Constraints = "$src1 = $dst" in {
6474 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6475 memopv2i64, i128mem>;
6476 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6477 memopv2i64, i128mem>;
6480 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6481 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6482 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6483 X86MemOperand x86memop, bit Is2Addr = 1> {
6484 let isCommutable = 1 in
6485 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6486 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6488 !strconcat(OpcodeStr,
6489 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6490 !strconcat(OpcodeStr,
6491 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6492 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6494 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6495 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6497 !strconcat(OpcodeStr,
6498 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6499 !strconcat(OpcodeStr,
6500 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6503 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6507 let Predicates = [HasAVX] in {
6508 let isCommutable = 0 in {
6509 let ExeDomain = SSEPackedSingle in {
6510 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6511 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6512 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6513 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6514 f256mem, 0>, VEX_4V, VEX_L;
6516 let ExeDomain = SSEPackedDouble in {
6517 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6518 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6519 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6520 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6521 f256mem, 0>, VEX_4V, VEX_L;
6523 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6524 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6525 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6526 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6528 let ExeDomain = SSEPackedSingle in
6529 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6530 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6531 let ExeDomain = SSEPackedDouble in
6532 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6533 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6534 let ExeDomain = SSEPackedSingle in
6535 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6536 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6539 let Predicates = [HasAVX2] in {
6540 let isCommutable = 0 in {
6541 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6542 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6543 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6544 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6548 let Constraints = "$src1 = $dst" in {
6549 let isCommutable = 0 in {
6550 let ExeDomain = SSEPackedSingle in
6551 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6552 VR128, memopv4f32, f128mem>;
6553 let ExeDomain = SSEPackedDouble in
6554 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6555 VR128, memopv2f64, f128mem>;
6556 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6557 VR128, memopv2i64, i128mem>;
6558 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6559 VR128, memopv2i64, i128mem>;
6561 let ExeDomain = SSEPackedSingle in
6562 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6563 VR128, memopv4f32, f128mem>;
6564 let ExeDomain = SSEPackedDouble in
6565 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6566 VR128, memopv2f64, f128mem>;
6569 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6570 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6571 RegisterClass RC, X86MemOperand x86memop,
6572 PatFrag mem_frag, Intrinsic IntId> {
6573 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6574 (ins RC:$src1, RC:$src2, RC:$src3),
6575 !strconcat(OpcodeStr,
6576 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6577 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6578 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6580 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6581 (ins RC:$src1, x86memop:$src2, RC:$src3),
6582 !strconcat(OpcodeStr,
6583 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6585 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6587 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6590 let Predicates = [HasAVX] in {
6591 let ExeDomain = SSEPackedDouble in {
6592 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6593 memopv2f64, int_x86_sse41_blendvpd>;
6594 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6595 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6596 } // ExeDomain = SSEPackedDouble
6597 let ExeDomain = SSEPackedSingle in {
6598 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6599 memopv4f32, int_x86_sse41_blendvps>;
6600 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6601 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6602 } // ExeDomain = SSEPackedSingle
6603 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6604 memopv2i64, int_x86_sse41_pblendvb>;
6607 let Predicates = [HasAVX2] in {
6608 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6609 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6612 let Predicates = [HasAVX] in {
6613 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6614 (v16i8 VR128:$src2))),
6615 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6616 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6617 (v4i32 VR128:$src2))),
6618 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6619 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6620 (v4f32 VR128:$src2))),
6621 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6622 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6623 (v2i64 VR128:$src2))),
6624 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6625 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6626 (v2f64 VR128:$src2))),
6627 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6628 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6629 (v8i32 VR256:$src2))),
6630 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6631 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6632 (v8f32 VR256:$src2))),
6633 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6634 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6635 (v4i64 VR256:$src2))),
6636 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6637 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6638 (v4f64 VR256:$src2))),
6639 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6641 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6643 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6644 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6646 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6648 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6650 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6651 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6653 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6654 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6656 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6659 let Predicates = [HasAVX2] in {
6660 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6661 (v32i8 VR256:$src2))),
6662 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6663 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6665 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6668 /// SS41I_ternary_int - SSE 4.1 ternary operator
6669 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6670 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6671 X86MemOperand x86memop, Intrinsic IntId> {
6672 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6673 (ins VR128:$src1, VR128:$src2),
6674 !strconcat(OpcodeStr,
6675 "\t{$src2, $dst|$dst, $src2}"),
6676 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6679 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6680 (ins VR128:$src1, x86memop:$src2),
6681 !strconcat(OpcodeStr,
6682 "\t{$src2, $dst|$dst, $src2}"),
6685 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6689 let ExeDomain = SSEPackedDouble in
6690 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6691 int_x86_sse41_blendvpd>;
6692 let ExeDomain = SSEPackedSingle in
6693 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6694 int_x86_sse41_blendvps>;
6695 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6696 int_x86_sse41_pblendvb>;
6698 // Aliases with the implicit xmm0 argument
6699 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6700 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6701 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6702 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6703 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6704 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6705 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6706 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6707 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6708 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6709 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6710 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6712 let Predicates = [UseSSE41] in {
6713 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6714 (v16i8 VR128:$src2))),
6715 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6716 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6717 (v4i32 VR128:$src2))),
6718 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6719 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6720 (v4f32 VR128:$src2))),
6721 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6722 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6723 (v2i64 VR128:$src2))),
6724 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6725 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6726 (v2f64 VR128:$src2))),
6727 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6729 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6731 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6732 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6734 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6735 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6737 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6741 let Predicates = [HasAVX] in
6742 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6743 "vmovntdqa\t{$src, $dst|$dst, $src}",
6744 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6746 let Predicates = [HasAVX2] in
6747 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6748 "vmovntdqa\t{$src, $dst|$dst, $src}",
6749 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6751 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6752 "movntdqa\t{$src, $dst|$dst, $src}",
6753 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6756 //===----------------------------------------------------------------------===//
6757 // SSE4.2 - Compare Instructions
6758 //===----------------------------------------------------------------------===//
6760 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6761 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6762 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6763 X86MemOperand x86memop, bit Is2Addr = 1> {
6764 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6765 (ins RC:$src1, RC:$src2),
6767 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6768 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6769 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6771 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6772 (ins RC:$src1, x86memop:$src2),
6774 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6775 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6777 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6780 let Predicates = [HasAVX] in
6781 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6782 memopv2i64, i128mem, 0>, VEX_4V;
6784 let Predicates = [HasAVX2] in
6785 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6786 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6788 let Constraints = "$src1 = $dst" in
6789 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6790 memopv2i64, i128mem>;
6792 //===----------------------------------------------------------------------===//
6793 // SSE4.2 - String/text Processing Instructions
6794 //===----------------------------------------------------------------------===//
6796 // Packed Compare Implicit Length Strings, Return Mask
6797 multiclass pseudo_pcmpistrm<string asm> {
6798 def REG : PseudoI<(outs VR128:$dst),
6799 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6800 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6802 def MEM : PseudoI<(outs VR128:$dst),
6803 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6804 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
6805 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6808 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6809 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6810 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
6813 multiclass pcmpistrm_SS42AI<string asm> {
6814 def rr : SS42AI<0x62, MRMSrcReg, (outs),
6815 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6816 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6819 def rm :SS42AI<0x62, MRMSrcMem, (outs),
6820 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6821 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6825 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6826 let Predicates = [HasAVX] in
6827 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
6828 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
6831 // Packed Compare Explicit Length Strings, Return Mask
6832 multiclass pseudo_pcmpestrm<string asm> {
6833 def REG : PseudoI<(outs VR128:$dst),
6834 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6835 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6836 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6837 def MEM : PseudoI<(outs VR128:$dst),
6838 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6839 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
6840 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
6843 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6844 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6845 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
6848 multiclass SS42AI_pcmpestrm<string asm> {
6849 def rr : SS42AI<0x60, MRMSrcReg, (outs),
6850 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6851 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6854 def rm : SS42AI<0x60, MRMSrcMem, (outs),
6855 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6856 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6860 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6861 let Predicates = [HasAVX] in
6862 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
6863 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
6866 // Packed Compare Implicit Length Strings, Return Index
6867 multiclass pseudo_pcmpistri<string asm> {
6868 def REG : PseudoI<(outs GR32:$dst),
6869 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6870 [(set GR32:$dst, EFLAGS,
6871 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
6872 def MEM : PseudoI<(outs GR32:$dst),
6873 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6874 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
6875 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6878 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6879 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
6880 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
6883 multiclass SS42AI_pcmpistri<string asm> {
6884 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6885 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6886 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6889 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6890 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6891 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6895 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
6896 let Predicates = [HasAVX] in
6897 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
6898 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
6901 // Packed Compare Explicit Length Strings, Return Index
6902 multiclass pseudo_pcmpestri<string asm> {
6903 def REG : PseudoI<(outs GR32:$dst),
6904 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6905 [(set GR32:$dst, EFLAGS,
6906 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6907 def MEM : PseudoI<(outs GR32:$dst),
6908 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6909 [(set GR32:$dst, EFLAGS,
6910 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
6914 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6915 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
6916 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
6919 multiclass SS42AI_pcmpestri<string asm> {
6920 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6921 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6922 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6925 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6926 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6927 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6931 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6932 let Predicates = [HasAVX] in
6933 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
6934 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
6937 //===----------------------------------------------------------------------===//
6938 // SSE4.2 - CRC Instructions
6939 //===----------------------------------------------------------------------===//
6941 // No CRC instructions have AVX equivalents
6943 // crc intrinsic instruction
6944 // This set of instructions are only rm, the only difference is the size
6946 let Constraints = "$src1 = $dst" in {
6947 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6948 (ins GR32:$src1, i8mem:$src2),
6949 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6951 (int_x86_sse42_crc32_32_8 GR32:$src1,
6952 (load addr:$src2)))]>;
6953 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6954 (ins GR32:$src1, GR8:$src2),
6955 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6957 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6958 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6959 (ins GR32:$src1, i16mem:$src2),
6960 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6962 (int_x86_sse42_crc32_32_16 GR32:$src1,
6963 (load addr:$src2)))]>,
6965 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6966 (ins GR32:$src1, GR16:$src2),
6967 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6969 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6971 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6972 (ins GR32:$src1, i32mem:$src2),
6973 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6975 (int_x86_sse42_crc32_32_32 GR32:$src1,
6976 (load addr:$src2)))]>;
6977 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6978 (ins GR32:$src1, GR32:$src2),
6979 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6981 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6982 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6983 (ins GR64:$src1, i8mem:$src2),
6984 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6986 (int_x86_sse42_crc32_64_8 GR64:$src1,
6987 (load addr:$src2)))]>,
6989 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6990 (ins GR64:$src1, GR8:$src2),
6991 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6993 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6995 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6996 (ins GR64:$src1, i64mem:$src2),
6997 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6999 (int_x86_sse42_crc32_64_64 GR64:$src1,
7000 (load addr:$src2)))]>,
7002 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7003 (ins GR64:$src1, GR64:$src2),
7004 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7006 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7010 //===----------------------------------------------------------------------===//
7011 // AES-NI Instructions
7012 //===----------------------------------------------------------------------===//
7014 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7015 Intrinsic IntId128, bit Is2Addr = 1> {
7016 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7017 (ins VR128:$src1, VR128:$src2),
7019 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7020 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7021 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7023 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7024 (ins VR128:$src1, i128mem:$src2),
7026 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7027 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7029 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7032 // Perform One Round of an AES Encryption/Decryption Flow
7033 let Predicates = [HasAVX, HasAES] in {
7034 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7035 int_x86_aesni_aesenc, 0>, VEX_4V;
7036 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7037 int_x86_aesni_aesenclast, 0>, VEX_4V;
7038 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7039 int_x86_aesni_aesdec, 0>, VEX_4V;
7040 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7041 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7044 let Constraints = "$src1 = $dst" in {
7045 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7046 int_x86_aesni_aesenc>;
7047 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7048 int_x86_aesni_aesenclast>;
7049 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7050 int_x86_aesni_aesdec>;
7051 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7052 int_x86_aesni_aesdeclast>;
7055 // Perform the AES InvMixColumn Transformation
7056 let Predicates = [HasAVX, HasAES] in {
7057 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7059 "vaesimc\t{$src1, $dst|$dst, $src1}",
7061 (int_x86_aesni_aesimc VR128:$src1))]>,
7063 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7064 (ins i128mem:$src1),
7065 "vaesimc\t{$src1, $dst|$dst, $src1}",
7066 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7069 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7071 "aesimc\t{$src1, $dst|$dst, $src1}",
7073 (int_x86_aesni_aesimc VR128:$src1))]>,
7075 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7076 (ins i128mem:$src1),
7077 "aesimc\t{$src1, $dst|$dst, $src1}",
7078 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7081 // AES Round Key Generation Assist
7082 let Predicates = [HasAVX, HasAES] in {
7083 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7084 (ins VR128:$src1, i8imm:$src2),
7085 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7087 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7089 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7090 (ins i128mem:$src1, i8imm:$src2),
7091 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7093 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7096 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7097 (ins VR128:$src1, i8imm:$src2),
7098 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7100 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7102 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7103 (ins i128mem:$src1, i8imm:$src2),
7104 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7106 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7109 //===----------------------------------------------------------------------===//
7110 // PCLMUL Instructions
7111 //===----------------------------------------------------------------------===//
7113 // AVX carry-less Multiplication instructions
7114 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7115 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7116 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7118 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7120 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7121 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7122 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7123 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7124 (memopv2i64 addr:$src2), imm:$src3))]>;
7126 // Carry-less Multiplication instructions
7127 let Constraints = "$src1 = $dst" in {
7128 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7129 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7130 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7132 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7134 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7135 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7136 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7137 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7138 (memopv2i64 addr:$src2), imm:$src3))]>;
7139 } // Constraints = "$src1 = $dst"
7142 multiclass pclmul_alias<string asm, int immop> {
7143 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7144 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7146 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7147 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7149 def : InstAlias<!strconcat("vpclmul", asm,
7150 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7151 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7153 def : InstAlias<!strconcat("vpclmul", asm,
7154 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7155 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7157 defm : pclmul_alias<"hqhq", 0x11>;
7158 defm : pclmul_alias<"hqlq", 0x01>;
7159 defm : pclmul_alias<"lqhq", 0x10>;
7160 defm : pclmul_alias<"lqlq", 0x00>;
7162 //===----------------------------------------------------------------------===//
7163 // SSE4A Instructions
7164 //===----------------------------------------------------------------------===//
7166 let Predicates = [HasSSE4A] in {
7168 let Constraints = "$src = $dst" in {
7169 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7170 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7171 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7172 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7173 imm:$idx))]>, TB, OpSize;
7174 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7175 (ins VR128:$src, VR128:$mask),
7176 "extrq\t{$mask, $src|$src, $mask}",
7177 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7178 VR128:$mask))]>, TB, OpSize;
7180 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7181 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7182 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7183 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7184 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7185 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7186 (ins VR128:$src, VR128:$mask),
7187 "insertq\t{$mask, $src|$src, $mask}",
7188 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7189 VR128:$mask))]>, XD;
7192 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7193 "movntss\t{$src, $dst|$dst, $src}",
7194 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7196 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7197 "movntsd\t{$src, $dst|$dst, $src}",
7198 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7201 //===----------------------------------------------------------------------===//
7203 //===----------------------------------------------------------------------===//
7205 //===----------------------------------------------------------------------===//
7206 // VBROADCAST - Load from memory and broadcast to all elements of the
7207 // destination operand
7209 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7210 X86MemOperand x86memop, Intrinsic Int> :
7211 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7212 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7213 [(set RC:$dst, (Int addr:$src))]>, VEX;
7215 // AVX2 adds register forms
7216 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7218 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7219 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7220 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7222 let ExeDomain = SSEPackedSingle in {
7223 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7224 int_x86_avx_vbroadcast_ss>;
7225 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7226 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7228 let ExeDomain = SSEPackedDouble in
7229 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7230 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7231 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7232 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7234 let ExeDomain = SSEPackedSingle in {
7235 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7236 int_x86_avx2_vbroadcast_ss_ps>;
7237 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7238 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7240 let ExeDomain = SSEPackedDouble in
7241 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7242 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7244 let Predicates = [HasAVX2] in
7245 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7246 int_x86_avx2_vbroadcasti128>, VEX_L;
7248 let Predicates = [HasAVX] in
7249 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7250 (VBROADCASTF128 addr:$src)>;
7253 //===----------------------------------------------------------------------===//
7254 // VINSERTF128 - Insert packed floating-point values
7256 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7257 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7258 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7259 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7262 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7263 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7264 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7268 let Predicates = [HasAVX] in {
7269 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7271 (VINSERTF128rr VR256:$src1, VR128:$src2,
7272 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7273 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7275 (VINSERTF128rr VR256:$src1, VR128:$src2,
7276 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7278 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7280 (VINSERTF128rm VR256:$src1, addr:$src2,
7281 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7282 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7284 (VINSERTF128rm VR256:$src1, addr:$src2,
7285 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7288 let Predicates = [HasAVX1Only] in {
7289 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7291 (VINSERTF128rr VR256:$src1, VR128:$src2,
7292 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7293 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7295 (VINSERTF128rr VR256:$src1, VR128:$src2,
7296 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7297 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7299 (VINSERTF128rr VR256:$src1, VR128:$src2,
7300 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7301 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7303 (VINSERTF128rr VR256:$src1, VR128:$src2,
7304 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7306 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7308 (VINSERTF128rm VR256:$src1, addr:$src2,
7309 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7310 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7311 (bc_v4i32 (memopv2i64 addr:$src2)),
7313 (VINSERTF128rm VR256:$src1, addr:$src2,
7314 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7315 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7316 (bc_v16i8 (memopv2i64 addr:$src2)),
7318 (VINSERTF128rm VR256:$src1, addr:$src2,
7319 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7320 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7321 (bc_v8i16 (memopv2i64 addr:$src2)),
7323 (VINSERTF128rm VR256:$src1, addr:$src2,
7324 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7327 //===----------------------------------------------------------------------===//
7328 // VEXTRACTF128 - Extract packed floating-point values
7330 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7331 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7332 (ins VR256:$src1, i8imm:$src2),
7333 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7336 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7337 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7338 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7343 let Predicates = [HasAVX] in {
7344 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7345 (v4f32 (VEXTRACTF128rr
7346 (v8f32 VR256:$src1),
7347 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7348 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7349 (v2f64 (VEXTRACTF128rr
7350 (v4f64 VR256:$src1),
7351 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7353 def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
7354 (iPTR imm))), addr:$dst),
7355 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7356 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7357 def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
7358 (iPTR imm))), addr:$dst),
7359 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7360 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7363 let Predicates = [HasAVX1Only] in {
7364 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7365 (v2i64 (VEXTRACTF128rr
7366 (v4i64 VR256:$src1),
7367 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7368 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7369 (v4i32 (VEXTRACTF128rr
7370 (v8i32 VR256:$src1),
7371 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7372 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7373 (v8i16 (VEXTRACTF128rr
7374 (v16i16 VR256:$src1),
7375 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7376 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7377 (v16i8 (VEXTRACTF128rr
7378 (v32i8 VR256:$src1),
7379 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7381 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7382 (iPTR imm))), addr:$dst),
7383 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7384 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7385 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7386 (iPTR imm))), addr:$dst),
7387 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7388 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7389 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7390 (iPTR imm))), addr:$dst),
7391 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7392 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7393 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7394 (iPTR imm))), addr:$dst),
7395 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7396 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7399 //===----------------------------------------------------------------------===//
7400 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7402 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7403 Intrinsic IntLd, Intrinsic IntLd256,
7404 Intrinsic IntSt, Intrinsic IntSt256> {
7405 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7406 (ins VR128:$src1, f128mem:$src2),
7407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7408 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7410 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7411 (ins VR256:$src1, f256mem:$src2),
7412 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7413 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7415 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7416 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7417 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7418 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7419 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7420 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7421 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7422 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7425 let ExeDomain = SSEPackedSingle in
7426 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7427 int_x86_avx_maskload_ps,
7428 int_x86_avx_maskload_ps_256,
7429 int_x86_avx_maskstore_ps,
7430 int_x86_avx_maskstore_ps_256>;
7431 let ExeDomain = SSEPackedDouble in
7432 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7433 int_x86_avx_maskload_pd,
7434 int_x86_avx_maskload_pd_256,
7435 int_x86_avx_maskstore_pd,
7436 int_x86_avx_maskstore_pd_256>;
7438 //===----------------------------------------------------------------------===//
7439 // VPERMIL - Permute Single and Double Floating-Point Values
7441 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7442 RegisterClass RC, X86MemOperand x86memop_f,
7443 X86MemOperand x86memop_i, PatFrag i_frag,
7444 Intrinsic IntVar, ValueType vt> {
7445 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7446 (ins RC:$src1, RC:$src2),
7447 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7448 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7449 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7450 (ins RC:$src1, x86memop_i:$src2),
7451 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7452 [(set RC:$dst, (IntVar RC:$src1,
7453 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7455 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7456 (ins RC:$src1, i8imm:$src2),
7457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7458 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7459 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7460 (ins x86memop_f:$src1, i8imm:$src2),
7461 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7463 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7466 let ExeDomain = SSEPackedSingle in {
7467 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7468 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7469 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7470 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7472 let ExeDomain = SSEPackedDouble in {
7473 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7474 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7475 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7476 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7479 let Predicates = [HasAVX] in {
7480 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7481 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7482 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7483 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7484 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7486 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7487 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7488 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7490 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7491 (VPERMILPDri VR128:$src1, imm:$imm)>;
7492 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7493 (VPERMILPDmi addr:$src1, imm:$imm)>;
7496 //===----------------------------------------------------------------------===//
7497 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7499 let ExeDomain = SSEPackedSingle in {
7500 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7501 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7502 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7503 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7504 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7505 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7506 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7507 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7508 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7509 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7512 let Predicates = [HasAVX] in {
7513 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7514 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7515 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7516 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7517 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7520 let Predicates = [HasAVX1Only] in {
7521 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7522 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7523 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7524 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7525 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7526 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7527 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7528 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7530 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7531 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7532 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7533 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7534 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7535 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7536 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7537 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7538 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7539 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7540 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7541 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7544 //===----------------------------------------------------------------------===//
7545 // VZERO - Zero YMM registers
7547 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7548 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7549 // Zero All YMM registers
7550 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7551 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7553 // Zero Upper bits of YMM registers
7554 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7555 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7558 //===----------------------------------------------------------------------===//
7559 // Half precision conversion instructions
7560 //===----------------------------------------------------------------------===//
7561 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7562 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7563 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7564 [(set RC:$dst, (Int VR128:$src))]>,
7566 let neverHasSideEffects = 1, mayLoad = 1 in
7567 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7568 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7571 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7572 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7573 (ins RC:$src1, i32i8imm:$src2),
7574 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7575 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7577 let neverHasSideEffects = 1, mayStore = 1 in
7578 def mr : Ii8<0x1D, MRMDestMem, (outs),
7579 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7580 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7584 let Predicates = [HasAVX, HasF16C] in {
7585 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7586 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7587 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7588 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7591 //===----------------------------------------------------------------------===//
7592 // AVX2 Instructions
7593 //===----------------------------------------------------------------------===//
7595 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7596 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7597 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7598 X86MemOperand x86memop> {
7599 let isCommutable = 1 in
7600 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7601 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7602 !strconcat(OpcodeStr,
7603 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7604 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7606 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7607 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7608 !strconcat(OpcodeStr,
7609 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7612 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7616 let isCommutable = 0 in {
7617 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7618 VR128, memopv2i64, i128mem>;
7619 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7620 VR256, memopv4i64, i256mem>, VEX_L;
7623 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7625 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7626 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7628 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7630 //===----------------------------------------------------------------------===//
7631 // VPBROADCAST - Load from memory and broadcast to all elements of the
7632 // destination operand
7634 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7635 X86MemOperand x86memop, PatFrag ld_frag,
7636 Intrinsic Int128, Intrinsic Int256> {
7637 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7638 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7639 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7640 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7641 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7643 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7644 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7645 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7646 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7647 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7648 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7650 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7654 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7655 int_x86_avx2_pbroadcastb_128,
7656 int_x86_avx2_pbroadcastb_256>;
7657 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7658 int_x86_avx2_pbroadcastw_128,
7659 int_x86_avx2_pbroadcastw_256>;
7660 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7661 int_x86_avx2_pbroadcastd_128,
7662 int_x86_avx2_pbroadcastd_256>;
7663 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7664 int_x86_avx2_pbroadcastq_128,
7665 int_x86_avx2_pbroadcastq_256>;
7667 let Predicates = [HasAVX2] in {
7668 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7669 (VPBROADCASTBrm addr:$src)>;
7670 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7671 (VPBROADCASTBYrm addr:$src)>;
7672 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7673 (VPBROADCASTWrm addr:$src)>;
7674 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7675 (VPBROADCASTWYrm addr:$src)>;
7676 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7677 (VPBROADCASTDrm addr:$src)>;
7678 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7679 (VPBROADCASTDYrm addr:$src)>;
7680 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7681 (VPBROADCASTQrm addr:$src)>;
7682 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7683 (VPBROADCASTQYrm addr:$src)>;
7685 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7686 (VPBROADCASTBrr VR128:$src)>;
7687 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7688 (VPBROADCASTBYrr VR128:$src)>;
7689 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7690 (VPBROADCASTWrr VR128:$src)>;
7691 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7692 (VPBROADCASTWYrr VR128:$src)>;
7693 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7694 (VPBROADCASTDrr VR128:$src)>;
7695 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7696 (VPBROADCASTDYrr VR128:$src)>;
7697 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7698 (VPBROADCASTQrr VR128:$src)>;
7699 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7700 (VPBROADCASTQYrr VR128:$src)>;
7701 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7702 (VBROADCASTSSrr VR128:$src)>;
7703 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7704 (VBROADCASTSSYrr VR128:$src)>;
7705 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7706 (VPBROADCASTQrr VR128:$src)>;
7707 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7708 (VBROADCASTSDYrr VR128:$src)>;
7710 // Provide fallback in case the load node that is used in the patterns above
7711 // is used by additional users, which prevents the pattern selection.
7712 let AddedComplexity = 20 in {
7713 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7714 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7715 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7716 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7717 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7718 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7720 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7721 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7722 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7723 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7724 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7725 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7729 // AVX1 broadcast patterns
7730 let Predicates = [HasAVX1Only] in {
7731 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7732 (VBROADCASTSSYrm addr:$src)>;
7733 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7734 (VBROADCASTSDYrm addr:$src)>;
7735 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7736 (VBROADCASTSSrm addr:$src)>;
7739 let Predicates = [HasAVX] in {
7740 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7741 (VBROADCASTSSYrm addr:$src)>;
7742 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7743 (VBROADCASTSDYrm addr:$src)>;
7744 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7745 (VBROADCASTSSrm addr:$src)>;
7747 // Provide fallback in case the load node that is used in the patterns above
7748 // is used by additional users, which prevents the pattern selection.
7749 let AddedComplexity = 20 in {
7750 // 128bit broadcasts:
7751 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7752 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7753 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7754 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7755 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7756 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7757 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7758 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7759 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7760 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7762 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7763 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7764 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7765 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7766 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7767 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7768 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7769 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7770 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7771 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7775 //===----------------------------------------------------------------------===//
7776 // VPERM - Permute instructions
7779 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7781 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7782 (ins VR256:$src1, VR256:$src2),
7783 !strconcat(OpcodeStr,
7784 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7786 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
7788 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7789 (ins VR256:$src1, i256mem:$src2),
7790 !strconcat(OpcodeStr,
7791 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7793 (OpVT (X86VPermv VR256:$src1,
7794 (bitconvert (mem_frag addr:$src2)))))]>,
7798 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7799 let ExeDomain = SSEPackedSingle in
7800 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7802 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7804 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7805 (ins VR256:$src1, i8imm:$src2),
7806 !strconcat(OpcodeStr,
7807 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7809 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
7811 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7812 (ins i256mem:$src1, i8imm:$src2),
7813 !strconcat(OpcodeStr,
7814 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7816 (OpVT (X86VPermi (mem_frag addr:$src1),
7817 (i8 imm:$src2))))]>, VEX, VEX_L;
7820 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7821 let ExeDomain = SSEPackedDouble in
7822 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7824 //===----------------------------------------------------------------------===//
7825 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7827 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7828 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7829 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7830 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7831 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7832 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7833 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7834 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7835 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7836 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7838 let Predicates = [HasAVX2] in {
7839 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7840 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7841 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7842 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7843 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7844 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7846 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7848 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7849 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7850 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7851 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7852 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7854 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7858 //===----------------------------------------------------------------------===//
7859 // VINSERTI128 - Insert packed integer values
7861 let neverHasSideEffects = 1 in {
7862 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7863 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7864 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7867 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7868 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7869 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7873 let Predicates = [HasAVX2] in {
7874 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7876 (VINSERTI128rr VR256:$src1, VR128:$src2,
7877 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7878 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7880 (VINSERTI128rr VR256:$src1, VR128:$src2,
7881 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7882 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7884 (VINSERTI128rr VR256:$src1, VR128:$src2,
7885 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7886 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7888 (VINSERTI128rr VR256:$src1, VR128:$src2,
7889 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7891 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7893 (VINSERTI128rm VR256:$src1, addr:$src2,
7894 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7895 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7896 (bc_v4i32 (memopv2i64 addr:$src2)),
7898 (VINSERTI128rm VR256:$src1, addr:$src2,
7899 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7900 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7901 (bc_v16i8 (memopv2i64 addr:$src2)),
7903 (VINSERTI128rm VR256:$src1, addr:$src2,
7904 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7905 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7906 (bc_v8i16 (memopv2i64 addr:$src2)),
7908 (VINSERTI128rm VR256:$src1, addr:$src2,
7909 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7912 //===----------------------------------------------------------------------===//
7913 // VEXTRACTI128 - Extract packed integer values
7915 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7916 (ins VR256:$src1, i8imm:$src2),
7917 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7919 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7921 let neverHasSideEffects = 1, mayStore = 1 in
7922 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7923 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7924 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7927 let Predicates = [HasAVX2] in {
7928 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7929 (v2i64 (VEXTRACTI128rr
7930 (v4i64 VR256:$src1),
7931 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7932 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7933 (v4i32 (VEXTRACTI128rr
7934 (v8i32 VR256:$src1),
7935 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7936 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7937 (v8i16 (VEXTRACTI128rr
7938 (v16i16 VR256:$src1),
7939 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7940 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7941 (v16i8 (VEXTRACTI128rr
7942 (v32i8 VR256:$src1),
7943 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7945 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7946 (iPTR imm))), addr:$dst),
7947 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7948 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7949 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7950 (iPTR imm))), addr:$dst),
7951 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7952 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7953 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7954 (iPTR imm))), addr:$dst),
7955 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7956 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7957 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7958 (iPTR imm))), addr:$dst),
7959 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7960 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7963 //===----------------------------------------------------------------------===//
7964 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7966 multiclass avx2_pmovmask<string OpcodeStr,
7967 Intrinsic IntLd128, Intrinsic IntLd256,
7968 Intrinsic IntSt128, Intrinsic IntSt256> {
7969 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7970 (ins VR128:$src1, i128mem:$src2),
7971 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7972 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7973 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7974 (ins VR256:$src1, i256mem:$src2),
7975 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7976 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7978 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7979 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7980 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7981 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7982 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7983 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7984 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7985 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7988 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7989 int_x86_avx2_maskload_d,
7990 int_x86_avx2_maskload_d_256,
7991 int_x86_avx2_maskstore_d,
7992 int_x86_avx2_maskstore_d_256>;
7993 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7994 int_x86_avx2_maskload_q,
7995 int_x86_avx2_maskload_q_256,
7996 int_x86_avx2_maskstore_q,
7997 int_x86_avx2_maskstore_q_256>, VEX_W;
8000 //===----------------------------------------------------------------------===//
8001 // Variable Bit Shifts
8003 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8004 ValueType vt128, ValueType vt256> {
8005 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8006 (ins VR128:$src1, VR128:$src2),
8007 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8009 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8011 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8012 (ins VR128:$src1, i128mem:$src2),
8013 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8015 (vt128 (OpNode VR128:$src1,
8016 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8018 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8019 (ins VR256:$src1, VR256:$src2),
8020 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8022 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8024 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8025 (ins VR256:$src1, i256mem:$src2),
8026 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8028 (vt256 (OpNode VR256:$src1,
8029 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8033 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8034 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8035 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8036 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8037 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8039 //===----------------------------------------------------------------------===//
8040 // VGATHER - GATHER Operations
8041 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8042 X86MemOperand memop128, X86MemOperand memop256> {
8043 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8044 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8045 !strconcat(OpcodeStr,
8046 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8048 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8049 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8050 !strconcat(OpcodeStr,
8051 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8052 []>, VEX_4VOp3, VEX_L;
8055 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8056 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8057 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8058 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8059 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8060 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8061 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8062 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8063 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;