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, $src2}"), []>;
2974 /// sse1_fp_unop_p - SSE1 unops in packed form.
2975 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2977 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2978 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
2980 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2981 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2982 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
2985 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2986 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
2988 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2989 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2990 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
2992 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2993 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2994 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
2998 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2999 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3000 Intrinsic V4F32Int, OpndItins itins> {
3001 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3002 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3003 [(set VR128:$dst, (V4F32Int VR128:$src))],
3005 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3006 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3007 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3011 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3012 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3013 Intrinsic V4F32Int, OpndItins itins> {
3014 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3015 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3016 [(set VR256:$dst, (V4F32Int VR256:$src))],
3018 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3019 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3020 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))],
3024 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3025 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3026 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3027 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3028 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3029 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3030 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3031 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3032 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3033 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3034 Requires<[UseSSE2, OptForSize]>;
3035 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3036 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3037 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3038 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3039 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3040 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3043 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3044 let hasSideEffects = 0 in
3045 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3046 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3047 !strconcat(OpcodeStr,
3048 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3049 let mayLoad = 1 in {
3050 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3051 !strconcat(OpcodeStr,
3052 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3053 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3054 (ins VR128:$src1, sdmem:$src2),
3055 !strconcat(OpcodeStr,
3056 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3060 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3061 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3062 SDNode OpNode, OpndItins itins> {
3063 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3064 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3065 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3066 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3067 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3068 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3071 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3072 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
3074 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3075 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3076 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3078 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3079 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3080 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3084 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3085 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3086 Intrinsic V2F64Int, OpndItins itins> {
3087 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3088 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3089 [(set VR128:$dst, (V2F64Int VR128:$src))],
3091 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3092 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3093 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))],
3097 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3098 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3099 Intrinsic V2F64Int, OpndItins itins> {
3100 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3101 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3102 [(set VR256:$dst, (V2F64Int VR256:$src))],
3104 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3105 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3106 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))],
3110 let Predicates = [HasAVX] in {
3112 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3113 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3115 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3116 sse2_fp_unop_p<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3117 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3118 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt, SSE_SQRTP>,
3119 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps,
3121 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd,
3123 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256,
3125 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256,
3129 // Reciprocal approximations. Note that these typically require refinement
3130 // in order to obtain suitable precision.
3131 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3132 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3133 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt, SSE_SQRTP>,
3134 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256,
3136 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps,
3139 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3140 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, SSE_RCPP>,
3141 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp, SSE_RCPP>,
3142 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256,
3144 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps,
3148 def : Pat<(f32 (fsqrt FR32:$src)),
3149 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3150 def : Pat<(f32 (fsqrt (load addr:$src))),
3151 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3152 Requires<[HasAVX, OptForSize]>;
3153 def : Pat<(f64 (fsqrt FR64:$src)),
3154 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3155 def : Pat<(f64 (fsqrt (load addr:$src))),
3156 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3157 Requires<[HasAVX, OptForSize]>;
3159 def : Pat<(f32 (X86frsqrt FR32:$src)),
3160 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3161 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3162 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3163 Requires<[HasAVX, OptForSize]>;
3165 def : Pat<(f32 (X86frcp FR32:$src)),
3166 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3167 def : Pat<(f32 (X86frcp (load addr:$src))),
3168 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3169 Requires<[HasAVX, OptForSize]>;
3171 let Predicates = [HasAVX] in {
3172 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3173 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3174 (COPY_TO_REGCLASS VR128:$src, FR32)),
3176 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3177 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3179 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3180 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3181 (COPY_TO_REGCLASS VR128:$src, FR64)),
3183 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3184 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3186 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3187 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3188 (COPY_TO_REGCLASS VR128:$src, FR32)),
3190 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3191 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3193 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3194 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3195 (COPY_TO_REGCLASS VR128:$src, FR32)),
3197 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3198 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3202 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3204 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3205 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps, SSE_SQRTS>,
3206 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3208 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTS>,
3209 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd, SSE_SQRTS>;
3211 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
3212 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
3213 Intrinsic F32Int, OpndItins itins> {
3214 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3215 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3216 [(set FR32:$dst, (OpNode FR32:$src))]>;
3217 // For scalar unary operations, fold a load into the operation
3218 // only in OptForSize mode. It eliminates an instruction, but it also
3219 // eliminates a whole-register clobber (the load), so it introduces a
3220 // partial register update condition.
3221 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3222 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3223 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3224 Requires<[UseSSE1, OptForSize]>;
3225 let Constraints = "$src1 = $dst" in {
3226 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3227 (ins VR128:$src1, VR128:$src2),
3228 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3230 let mayLoad = 1, hasSideEffects = 0 in
3231 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3232 (ins VR128:$src1, ssmem:$src2),
3233 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3238 // Reciprocal approximations. Note that these typically require refinement
3239 // in order to obtain suitable precision.
3240 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss,
3242 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3243 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3245 let Predicates = [UseSSE1] in {
3246 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3247 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3250 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss,
3252 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPS>,
3253 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps, SSE_RCPS>;
3254 let Predicates = [UseSSE1] in {
3255 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3256 (RCPSSr_Int VR128:$src, VR128:$src)>;
3259 // There is no f64 version of the reciprocal approximation instructions.
3261 //===----------------------------------------------------------------------===//
3262 // SSE 1 & 2 - Non-temporal stores
3263 //===----------------------------------------------------------------------===//
3265 let AddedComplexity = 400 in { // Prefer non-temporal versions
3266 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3267 (ins f128mem:$dst, VR128:$src),
3268 "movntps\t{$src, $dst|$dst, $src}",
3269 [(alignednontemporalstore (v4f32 VR128:$src),
3271 IIC_SSE_MOVNT>, VEX;
3272 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3273 (ins f128mem:$dst, VR128:$src),
3274 "movntpd\t{$src, $dst|$dst, $src}",
3275 [(alignednontemporalstore (v2f64 VR128:$src),
3277 IIC_SSE_MOVNT>, VEX;
3279 let ExeDomain = SSEPackedInt in
3280 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3281 (ins f128mem:$dst, VR128:$src),
3282 "movntdq\t{$src, $dst|$dst, $src}",
3283 [(alignednontemporalstore (v2i64 VR128:$src),
3285 IIC_SSE_MOVNT>, VEX;
3287 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3288 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3290 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3291 (ins f256mem:$dst, VR256:$src),
3292 "movntps\t{$src, $dst|$dst, $src}",
3293 [(alignednontemporalstore (v8f32 VR256:$src),
3295 IIC_SSE_MOVNT>, VEX, VEX_L;
3296 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3297 (ins f256mem:$dst, VR256:$src),
3298 "movntpd\t{$src, $dst|$dst, $src}",
3299 [(alignednontemporalstore (v4f64 VR256:$src),
3301 IIC_SSE_MOVNT>, VEX, VEX_L;
3302 let ExeDomain = SSEPackedInt in
3303 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3304 (ins f256mem:$dst, VR256:$src),
3305 "movntdq\t{$src, $dst|$dst, $src}",
3306 [(alignednontemporalstore (v4i64 VR256:$src),
3308 IIC_SSE_MOVNT>, VEX, VEX_L;
3311 let AddedComplexity = 400 in { // Prefer non-temporal versions
3312 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3313 "movntps\t{$src, $dst|$dst, $src}",
3314 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3316 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3317 "movntpd\t{$src, $dst|$dst, $src}",
3318 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3321 let ExeDomain = SSEPackedInt in
3322 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3323 "movntdq\t{$src, $dst|$dst, $src}",
3324 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3327 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3328 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3330 // There is no AVX form for instructions below this point
3331 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3332 "movnti{l}\t{$src, $dst|$dst, $src}",
3333 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3335 TB, Requires<[HasSSE2]>;
3336 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3337 "movnti{q}\t{$src, $dst|$dst, $src}",
3338 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3340 TB, Requires<[HasSSE2]>;
3343 //===----------------------------------------------------------------------===//
3344 // SSE 1 & 2 - Prefetch and memory fence
3345 //===----------------------------------------------------------------------===//
3347 // Prefetch intrinsic.
3348 let Predicates = [HasSSE1] in {
3349 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3350 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3351 IIC_SSE_PREFETCH>, TB;
3352 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3353 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3354 IIC_SSE_PREFETCH>, TB;
3355 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3356 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3357 IIC_SSE_PREFETCH>, TB;
3358 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3359 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3360 IIC_SSE_PREFETCH>, TB;
3364 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3365 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3366 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3368 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3369 // was introduced with SSE2, it's backward compatible.
3370 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3372 // Load, store, and memory fence
3373 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3374 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3375 TB, Requires<[HasSSE1]>;
3376 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3377 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3378 TB, Requires<[HasSSE2]>;
3379 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3380 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3381 TB, Requires<[HasSSE2]>;
3383 def : Pat<(X86SFence), (SFENCE)>;
3384 def : Pat<(X86LFence), (LFENCE)>;
3385 def : Pat<(X86MFence), (MFENCE)>;
3387 //===----------------------------------------------------------------------===//
3388 // SSE 1 & 2 - Load/Store XCSR register
3389 //===----------------------------------------------------------------------===//
3391 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3392 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3393 IIC_SSE_LDMXCSR>, VEX;
3394 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3395 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3396 IIC_SSE_STMXCSR>, VEX;
3398 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3399 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3401 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3402 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3405 //===---------------------------------------------------------------------===//
3406 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3407 //===---------------------------------------------------------------------===//
3409 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3411 let neverHasSideEffects = 1 in {
3412 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3413 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3415 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3416 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3418 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3419 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3421 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3422 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3427 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3428 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3429 "movdqa\t{$src, $dst|$dst, $src}", [],
3432 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3433 "movdqa\t{$src, $dst|$dst, $src}", [],
3434 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3435 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3436 "movdqu\t{$src, $dst|$dst, $src}", [],
3439 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3440 "movdqu\t{$src, $dst|$dst, $src}", [],
3441 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3444 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3445 neverHasSideEffects = 1 in {
3446 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3447 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3449 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3450 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3452 let Predicates = [HasAVX] in {
3453 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3454 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3456 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3457 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3462 let mayStore = 1, neverHasSideEffects = 1 in {
3463 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3464 (ins i128mem:$dst, VR128:$src),
3465 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3467 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3468 (ins i256mem:$dst, VR256:$src),
3469 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3471 let Predicates = [HasAVX] in {
3472 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3473 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3475 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3476 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3481 let neverHasSideEffects = 1 in
3482 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3483 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3485 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3486 "movdqu\t{$src, $dst|$dst, $src}",
3487 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3490 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3491 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3492 "movdqa\t{$src, $dst|$dst, $src}", [],
3495 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3496 "movdqu\t{$src, $dst|$dst, $src}",
3497 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3500 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3501 neverHasSideEffects = 1 in {
3502 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3503 "movdqa\t{$src, $dst|$dst, $src}",
3504 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3506 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3507 "movdqu\t{$src, $dst|$dst, $src}",
3508 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3510 XS, Requires<[UseSSE2]>;
3513 let mayStore = 1 in {
3514 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3515 "movdqa\t{$src, $dst|$dst, $src}",
3516 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3518 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3519 "movdqu\t{$src, $dst|$dst, $src}",
3520 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3522 XS, Requires<[UseSSE2]>;
3525 } // ExeDomain = SSEPackedInt
3527 let Predicates = [HasAVX] in {
3528 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3529 (VMOVDQUmr addr:$dst, VR128:$src)>;
3530 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3531 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3533 let Predicates = [UseSSE2] in
3534 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3535 (MOVDQUmr addr:$dst, VR128:$src)>;
3537 //===---------------------------------------------------------------------===//
3538 // SSE2 - Packed Integer Arithmetic Instructions
3539 //===---------------------------------------------------------------------===//
3541 def SSE_PMADD : OpndItins<
3542 IIC_SSE_PMADD, IIC_SSE_PMADD
3545 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3547 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3548 RegisterClass RC, PatFrag memop_frag,
3549 X86MemOperand x86memop,
3551 bit IsCommutable = 0,
3553 let isCommutable = IsCommutable in
3554 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3555 (ins RC:$src1, RC:$src2),
3557 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3558 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3559 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3560 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3561 (ins RC:$src1, x86memop:$src2),
3563 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3564 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3565 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3569 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3570 Intrinsic IntId256, OpndItins itins,
3571 bit IsCommutable = 0> {
3572 let Predicates = [HasAVX] in
3573 defm V#NAME# : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3574 VR128, memopv2i64, i128mem, itins,
3575 IsCommutable, 0>, VEX_4V;
3577 let Constraints = "$src1 = $dst" in
3578 defm #NAME# : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3579 i128mem, itins, IsCommutable, 1>;
3581 let Predicates = [HasAVX2] in
3582 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3583 VR256, memopv4i64, i256mem, itins,
3584 IsCommutable, 0>, VEX_4V, VEX_L;
3587 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3588 string OpcodeStr, SDNode OpNode,
3589 SDNode OpNode2, RegisterClass RC,
3590 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3591 ShiftOpndItins itins,
3593 // src2 is always 128-bit
3594 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3595 (ins RC:$src1, VR128:$src2),
3597 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3598 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3599 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3601 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3602 (ins RC:$src1, i128mem:$src2),
3604 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3605 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3606 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3607 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3608 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3609 (ins RC:$src1, i32i8imm:$src2),
3611 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3612 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3613 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3616 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3617 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3618 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3619 PatFrag memop_frag, X86MemOperand x86memop,
3621 bit IsCommutable = 0, bit Is2Addr = 1> {
3622 let isCommutable = IsCommutable in
3623 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3624 (ins RC:$src1, RC:$src2),
3626 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3627 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3628 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3629 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3630 (ins RC:$src1, x86memop:$src2),
3632 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3634 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3635 (bitconvert (memop_frag addr:$src2)))))]>;
3637 } // ExeDomain = SSEPackedInt
3639 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3640 SSE_INTALU_ITINS_P, 1>;
3641 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3642 SSE_INTALU_ITINS_P, 1>;
3643 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3644 SSE_INTALU_ITINS_P, 1>;
3645 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3646 SSE_INTALUQ_ITINS_P, 1>;
3647 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3648 SSE_INTMUL_ITINS_P, 1>;
3649 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3650 SSE_INTALU_ITINS_P, 0>;
3651 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3652 SSE_INTALU_ITINS_P, 0>;
3653 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3654 SSE_INTALU_ITINS_P, 0>;
3655 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3656 SSE_INTALUQ_ITINS_P, 0>;
3657 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3658 SSE_INTALU_ITINS_P, 0>;
3659 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3660 SSE_INTALU_ITINS_P, 0>;
3661 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3662 SSE_INTALU_ITINS_P, 1>;
3663 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3664 SSE_INTALU_ITINS_P, 1>;
3665 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3666 SSE_INTALU_ITINS_P, 1>;
3667 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3668 SSE_INTALU_ITINS_P, 1>;
3671 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3672 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3673 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3674 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3675 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3676 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3677 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3678 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3679 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3680 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3681 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3682 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3683 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3684 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3685 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3686 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3687 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3688 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3689 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3690 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3691 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3692 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3693 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3694 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3696 let Predicates = [HasAVX] in
3697 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3698 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3700 let Predicates = [HasAVX2] in
3701 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3702 VR256, memopv4i64, i256mem,
3703 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3704 let Constraints = "$src1 = $dst" in
3705 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3706 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3708 //===---------------------------------------------------------------------===//
3709 // SSE2 - Packed Integer Logical Instructions
3710 //===---------------------------------------------------------------------===//
3712 let Predicates = [HasAVX] in {
3713 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3714 VR128, v8i16, v8i16, bc_v8i16,
3715 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3716 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3717 VR128, v4i32, v4i32, bc_v4i32,
3718 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3719 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3720 VR128, v2i64, v2i64, bc_v2i64,
3721 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3723 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3724 VR128, v8i16, v8i16, bc_v8i16,
3725 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3726 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3727 VR128, v4i32, v4i32, bc_v4i32,
3728 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3729 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3730 VR128, v2i64, v2i64, bc_v2i64,
3731 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3733 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3734 VR128, v8i16, v8i16, bc_v8i16,
3735 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3736 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3737 VR128, v4i32, v4i32, bc_v4i32,
3738 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3740 let ExeDomain = SSEPackedInt in {
3741 // 128-bit logical shifts.
3742 def VPSLLDQri : PDIi8<0x73, MRM7r,
3743 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3744 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3746 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3748 def VPSRLDQri : PDIi8<0x73, MRM3r,
3749 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3750 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3752 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3754 // PSRADQri doesn't exist in SSE[1-3].
3756 } // Predicates = [HasAVX]
3758 let Predicates = [HasAVX2] in {
3759 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3760 VR256, v16i16, v8i16, bc_v8i16,
3761 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3762 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3763 VR256, v8i32, v4i32, bc_v4i32,
3764 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3765 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3766 VR256, v4i64, v2i64, bc_v2i64,
3767 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3769 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3770 VR256, v16i16, v8i16, bc_v8i16,
3771 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3772 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3773 VR256, v8i32, v4i32, bc_v4i32,
3774 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3775 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3776 VR256, v4i64, v2i64, bc_v2i64,
3777 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3779 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3780 VR256, v16i16, v8i16, bc_v8i16,
3781 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3782 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3783 VR256, v8i32, v4i32, bc_v4i32,
3784 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3786 let ExeDomain = SSEPackedInt in {
3787 // 256-bit logical shifts.
3788 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3789 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3790 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3792 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3794 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3795 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3796 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3798 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3800 // PSRADQYri doesn't exist in SSE[1-3].
3802 } // Predicates = [HasAVX2]
3804 let Constraints = "$src1 = $dst" in {
3805 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3806 VR128, v8i16, v8i16, bc_v8i16,
3807 SSE_INTSHIFT_ITINS_P>;
3808 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3809 VR128, v4i32, v4i32, bc_v4i32,
3810 SSE_INTSHIFT_ITINS_P>;
3811 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3812 VR128, v2i64, v2i64, bc_v2i64,
3813 SSE_INTSHIFT_ITINS_P>;
3815 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3816 VR128, v8i16, v8i16, bc_v8i16,
3817 SSE_INTSHIFT_ITINS_P>;
3818 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3819 VR128, v4i32, v4i32, bc_v4i32,
3820 SSE_INTSHIFT_ITINS_P>;
3821 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3822 VR128, v2i64, v2i64, bc_v2i64,
3823 SSE_INTSHIFT_ITINS_P>;
3825 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3826 VR128, v8i16, v8i16, bc_v8i16,
3827 SSE_INTSHIFT_ITINS_P>;
3828 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3829 VR128, v4i32, v4i32, bc_v4i32,
3830 SSE_INTSHIFT_ITINS_P>;
3832 let ExeDomain = SSEPackedInt in {
3833 // 128-bit logical shifts.
3834 def PSLLDQri : PDIi8<0x73, MRM7r,
3835 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3836 "pslldq\t{$src2, $dst|$dst, $src2}",
3838 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3839 def PSRLDQri : PDIi8<0x73, MRM3r,
3840 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3841 "psrldq\t{$src2, $dst|$dst, $src2}",
3843 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3844 // PSRADQri doesn't exist in SSE[1-3].
3846 } // Constraints = "$src1 = $dst"
3848 let Predicates = [HasAVX] in {
3849 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3850 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3851 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3852 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3853 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3854 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3856 // Shift up / down and insert zero's.
3857 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3858 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3859 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3860 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3863 let Predicates = [HasAVX2] in {
3864 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3865 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3866 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3867 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3870 let Predicates = [UseSSE2] in {
3871 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3872 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3873 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3874 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3875 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3876 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3878 // Shift up / down and insert zero's.
3879 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3880 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3881 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3882 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3885 //===---------------------------------------------------------------------===//
3886 // SSE2 - Packed Integer Comparison Instructions
3887 //===---------------------------------------------------------------------===//
3889 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
3890 SSE_INTALU_ITINS_P, 1>;
3891 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
3892 SSE_INTALU_ITINS_P, 1>;
3893 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
3894 SSE_INTALU_ITINS_P, 1>;
3895 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
3896 SSE_INTALU_ITINS_P, 0>;
3897 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
3898 SSE_INTALU_ITINS_P, 0>;
3899 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
3900 SSE_INTALU_ITINS_P, 0>;
3902 //===---------------------------------------------------------------------===//
3903 // SSE2 - Packed Integer Pack Instructions
3904 //===---------------------------------------------------------------------===//
3906 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3907 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
3908 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3909 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
3910 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3911 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
3913 //===---------------------------------------------------------------------===//
3914 // SSE2 - Packed Integer Shuffle Instructions
3915 //===---------------------------------------------------------------------===//
3917 let ExeDomain = SSEPackedInt in {
3918 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
3919 def ri : Ii8<0x70, MRMSrcReg,
3920 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3921 !strconcat(OpcodeStr,
3922 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3923 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
3925 def mi : Ii8<0x70, MRMSrcMem,
3926 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3927 !strconcat(OpcodeStr,
3928 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3930 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
3935 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3936 def Yri : Ii8<0x70, MRMSrcReg,
3937 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3938 !strconcat(OpcodeStr,
3939 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3940 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3941 def Ymi : Ii8<0x70, MRMSrcMem,
3942 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3943 !strconcat(OpcodeStr,
3944 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3946 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3947 (i8 imm:$src2))))]>;
3949 } // ExeDomain = SSEPackedInt
3951 let Predicates = [HasAVX] in {
3952 let AddedComplexity = 5 in
3953 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
3955 // SSE2 with ImmT == Imm8 and XS prefix.
3956 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
3958 // SSE2 with ImmT == Imm8 and XD prefix.
3959 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
3961 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3962 (VPSHUFDmi addr:$src1, imm:$imm)>;
3963 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3964 (VPSHUFDri VR128:$src1, imm:$imm)>;
3967 let Predicates = [HasAVX2] in {
3968 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>,
3969 TB, OpSize, VEX,VEX_L;
3970 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>,
3972 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>,
3976 let Predicates = [UseSSE2] in {
3977 let AddedComplexity = 5 in
3978 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
3980 // SSE2 with ImmT == Imm8 and XS prefix.
3981 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
3983 // SSE2 with ImmT == Imm8 and XD prefix.
3984 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
3986 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3987 (PSHUFDmi addr:$src1, imm:$imm)>;
3988 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3989 (PSHUFDri VR128:$src1, imm:$imm)>;
3992 //===---------------------------------------------------------------------===//
3993 // SSE2 - Packed Integer Unpack Instructions
3994 //===---------------------------------------------------------------------===//
3996 let ExeDomain = SSEPackedInt in {
3997 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3998 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3999 def rr : PDI<opc, MRMSrcReg,
4000 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4002 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4003 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4004 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
4006 def rm : PDI<opc, MRMSrcMem,
4007 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4009 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4010 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4011 [(set VR128:$dst, (OpNode VR128:$src1,
4012 (bc_frag (memopv2i64
4017 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4018 SDNode OpNode, PatFrag bc_frag> {
4019 def Yrr : PDI<opc, MRMSrcReg,
4020 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4021 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4022 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4023 def Yrm : PDI<opc, MRMSrcMem,
4024 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4025 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4026 [(set VR256:$dst, (OpNode VR256:$src1,
4027 (bc_frag (memopv4i64 addr:$src2))))]>;
4030 let Predicates = [HasAVX] in {
4031 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4032 bc_v16i8, 0>, VEX_4V;
4033 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4034 bc_v8i16, 0>, VEX_4V;
4035 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4036 bc_v4i32, 0>, VEX_4V;
4037 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4038 bc_v2i64, 0>, VEX_4V;
4040 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4041 bc_v16i8, 0>, VEX_4V;
4042 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4043 bc_v8i16, 0>, VEX_4V;
4044 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4045 bc_v4i32, 0>, VEX_4V;
4046 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4047 bc_v2i64, 0>, VEX_4V;
4050 let Predicates = [HasAVX2] in {
4051 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4052 bc_v32i8>, VEX_4V, VEX_L;
4053 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4054 bc_v16i16>, VEX_4V, VEX_L;
4055 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4056 bc_v8i32>, VEX_4V, VEX_L;
4057 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4058 bc_v4i64>, VEX_4V, VEX_L;
4060 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4061 bc_v32i8>, VEX_4V, VEX_L;
4062 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4063 bc_v16i16>, VEX_4V, VEX_L;
4064 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4065 bc_v8i32>, VEX_4V, VEX_L;
4066 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4067 bc_v4i64>, VEX_4V, VEX_L;
4070 let Constraints = "$src1 = $dst" in {
4071 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4073 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4075 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4077 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4080 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4082 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4084 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4086 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4089 } // ExeDomain = SSEPackedInt
4091 //===---------------------------------------------------------------------===//
4092 // SSE2 - Packed Integer Extract and Insert
4093 //===---------------------------------------------------------------------===//
4095 let ExeDomain = SSEPackedInt in {
4096 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4097 def rri : Ii8<0xC4, MRMSrcReg,
4098 (outs VR128:$dst), (ins VR128:$src1,
4099 GR32:$src2, i32i8imm:$src3),
4101 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4102 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4104 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4105 def rmi : Ii8<0xC4, MRMSrcMem,
4106 (outs VR128:$dst), (ins VR128:$src1,
4107 i16mem:$src2, i32i8imm:$src3),
4109 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4110 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4112 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4113 imm:$src3))], IIC_SSE_PINSRW>;
4117 let Predicates = [HasAVX] in
4118 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4119 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4120 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4121 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4122 imm:$src2))]>, TB, OpSize, VEX;
4123 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4124 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4125 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4126 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4127 imm:$src2))], IIC_SSE_PEXTRW>;
4130 let Predicates = [HasAVX] in {
4131 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4132 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4133 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4134 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4135 []>, TB, OpSize, VEX_4V;
4138 let Constraints = "$src1 = $dst" in
4139 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4141 } // ExeDomain = SSEPackedInt
4143 //===---------------------------------------------------------------------===//
4144 // SSE2 - Packed Mask Creation
4145 //===---------------------------------------------------------------------===//
4147 let ExeDomain = SSEPackedInt in {
4149 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4150 "pmovmskb\t{$src, $dst|$dst, $src}",
4151 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4152 IIC_SSE_MOVMSK>, VEX;
4153 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4154 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4156 let Predicates = [HasAVX2] in {
4157 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4158 "pmovmskb\t{$src, $dst|$dst, $src}",
4159 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4160 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4161 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4164 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4165 "pmovmskb\t{$src, $dst|$dst, $src}",
4166 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4169 } // ExeDomain = SSEPackedInt
4171 //===---------------------------------------------------------------------===//
4172 // SSE2 - Conditional Store
4173 //===---------------------------------------------------------------------===//
4175 let ExeDomain = SSEPackedInt in {
4178 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4179 (ins VR128:$src, VR128:$mask),
4180 "maskmovdqu\t{$mask, $src|$src, $mask}",
4181 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4182 IIC_SSE_MASKMOV>, VEX;
4184 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4185 (ins VR128:$src, VR128:$mask),
4186 "maskmovdqu\t{$mask, $src|$src, $mask}",
4187 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4188 IIC_SSE_MASKMOV>, VEX;
4191 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4192 "maskmovdqu\t{$mask, $src|$src, $mask}",
4193 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4196 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4197 "maskmovdqu\t{$mask, $src|$src, $mask}",
4198 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4201 } // ExeDomain = SSEPackedInt
4203 //===---------------------------------------------------------------------===//
4204 // SSE2 - Move Doubleword
4205 //===---------------------------------------------------------------------===//
4207 //===---------------------------------------------------------------------===//
4208 // Move Int Doubleword to Packed Double Int
4210 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4211 "movd\t{$src, $dst|$dst, $src}",
4213 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4215 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4216 "movd\t{$src, $dst|$dst, $src}",
4218 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4221 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4222 "mov{d|q}\t{$src, $dst|$dst, $src}",
4224 (v2i64 (scalar_to_vector GR64:$src)))],
4225 IIC_SSE_MOVDQ>, VEX;
4226 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4227 "mov{d|q}\t{$src, $dst|$dst, $src}",
4228 [(set FR64:$dst, (bitconvert GR64:$src))],
4229 IIC_SSE_MOVDQ>, VEX;
4231 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4232 "movd\t{$src, $dst|$dst, $src}",
4234 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4235 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4236 "movd\t{$src, $dst|$dst, $src}",
4238 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4240 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4241 "mov{d|q}\t{$src, $dst|$dst, $src}",
4243 (v2i64 (scalar_to_vector GR64:$src)))],
4245 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4246 "mov{d|q}\t{$src, $dst|$dst, $src}",
4247 [(set FR64:$dst, (bitconvert GR64:$src))],
4250 //===---------------------------------------------------------------------===//
4251 // Move Int Doubleword to Single Scalar
4253 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4254 "movd\t{$src, $dst|$dst, $src}",
4255 [(set FR32:$dst, (bitconvert GR32:$src))],
4256 IIC_SSE_MOVDQ>, VEX;
4258 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4259 "movd\t{$src, $dst|$dst, $src}",
4260 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4263 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4264 "movd\t{$src, $dst|$dst, $src}",
4265 [(set FR32:$dst, (bitconvert GR32:$src))],
4268 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4269 "movd\t{$src, $dst|$dst, $src}",
4270 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4273 //===---------------------------------------------------------------------===//
4274 // Move Packed Doubleword Int to Packed Double Int
4276 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4277 "movd\t{$src, $dst|$dst, $src}",
4278 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4279 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4280 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4281 (ins i32mem:$dst, VR128:$src),
4282 "movd\t{$src, $dst|$dst, $src}",
4283 [(store (i32 (vector_extract (v4i32 VR128:$src),
4284 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4286 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4287 "movd\t{$src, $dst|$dst, $src}",
4288 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4289 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4290 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4291 "movd\t{$src, $dst|$dst, $src}",
4292 [(store (i32 (vector_extract (v4i32 VR128:$src),
4293 (iPTR 0))), addr:$dst)],
4296 //===---------------------------------------------------------------------===//
4297 // Move Packed Doubleword Int first element to Doubleword Int
4299 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4300 "vmov{d|q}\t{$src, $dst|$dst, $src}",
4301 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4304 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4306 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4307 "mov{d|q}\t{$src, $dst|$dst, $src}",
4308 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4312 //===---------------------------------------------------------------------===//
4313 // Bitcast FR64 <-> GR64
4315 let Predicates = [HasAVX] in
4316 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4317 "vmovq\t{$src, $dst|$dst, $src}",
4318 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4320 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4321 "mov{d|q}\t{$src, $dst|$dst, $src}",
4322 [(set GR64:$dst, (bitconvert FR64:$src))],
4323 IIC_SSE_MOVDQ>, VEX;
4324 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4325 "movq\t{$src, $dst|$dst, $src}",
4326 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4327 IIC_SSE_MOVDQ>, VEX;
4329 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4330 "movq\t{$src, $dst|$dst, $src}",
4331 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4333 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4334 "mov{d|q}\t{$src, $dst|$dst, $src}",
4335 [(set GR64:$dst, (bitconvert FR64:$src))],
4337 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4338 "movq\t{$src, $dst|$dst, $src}",
4339 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4342 //===---------------------------------------------------------------------===//
4343 // Move Scalar Single to Double Int
4345 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4346 "movd\t{$src, $dst|$dst, $src}",
4347 [(set GR32:$dst, (bitconvert FR32:$src))],
4348 IIC_SSE_MOVD_ToGP>, VEX;
4349 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4350 "movd\t{$src, $dst|$dst, $src}",
4351 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4352 IIC_SSE_MOVDQ>, VEX;
4353 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4354 "movd\t{$src, $dst|$dst, $src}",
4355 [(set GR32:$dst, (bitconvert FR32:$src))],
4357 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4358 "movd\t{$src, $dst|$dst, $src}",
4359 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4362 //===---------------------------------------------------------------------===//
4363 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4365 let AddedComplexity = 15 in {
4366 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4367 "movd\t{$src, $dst|$dst, $src}",
4368 [(set VR128:$dst, (v4i32 (X86vzmovl
4369 (v4i32 (scalar_to_vector GR32:$src)))))],
4370 IIC_SSE_MOVDQ>, VEX;
4371 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4372 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4373 [(set VR128:$dst, (v2i64 (X86vzmovl
4374 (v2i64 (scalar_to_vector GR64:$src)))))],
4378 let AddedComplexity = 15 in {
4379 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4380 "movd\t{$src, $dst|$dst, $src}",
4381 [(set VR128:$dst, (v4i32 (X86vzmovl
4382 (v4i32 (scalar_to_vector GR32:$src)))))],
4384 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4385 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4386 [(set VR128:$dst, (v2i64 (X86vzmovl
4387 (v2i64 (scalar_to_vector GR64:$src)))))],
4391 let AddedComplexity = 20 in {
4392 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4393 "movd\t{$src, $dst|$dst, $src}",
4395 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4396 (loadi32 addr:$src))))))],
4397 IIC_SSE_MOVDQ>, VEX;
4398 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4399 "movd\t{$src, $dst|$dst, $src}",
4401 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4402 (loadi32 addr:$src))))))],
4406 let Predicates = [HasAVX] in {
4407 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4408 let AddedComplexity = 20 in {
4409 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4410 (VMOVZDI2PDIrm addr:$src)>;
4411 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4412 (VMOVZDI2PDIrm addr:$src)>;
4414 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4415 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4416 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4417 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4418 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4419 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4420 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4423 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4424 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4425 (MOVZDI2PDIrm addr:$src)>;
4426 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4427 (MOVZDI2PDIrm addr:$src)>;
4430 // These are the correct encodings of the instructions so that we know how to
4431 // read correct assembly, even though we continue to emit the wrong ones for
4432 // compatibility with Darwin's buggy assembler.
4433 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4434 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4435 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4436 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4437 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4438 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4439 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4440 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4441 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4442 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4443 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4444 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4446 //===---------------------------------------------------------------------===//
4447 // SSE2 - Move Quadword
4448 //===---------------------------------------------------------------------===//
4450 //===---------------------------------------------------------------------===//
4451 // Move Quadword Int to Packed Quadword Int
4453 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4454 "vmovq\t{$src, $dst|$dst, $src}",
4456 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4457 VEX, Requires<[HasAVX]>;
4458 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4459 "movq\t{$src, $dst|$dst, $src}",
4461 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4463 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4465 //===---------------------------------------------------------------------===//
4466 // Move Packed Quadword Int to Quadword Int
4468 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4469 "movq\t{$src, $dst|$dst, $src}",
4470 [(store (i64 (vector_extract (v2i64 VR128:$src),
4471 (iPTR 0))), addr:$dst)],
4472 IIC_SSE_MOVDQ>, VEX;
4473 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4474 "movq\t{$src, $dst|$dst, $src}",
4475 [(store (i64 (vector_extract (v2i64 VR128:$src),
4476 (iPTR 0))), addr:$dst)],
4479 //===---------------------------------------------------------------------===//
4480 // Store / copy lower 64-bits of a XMM register.
4482 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4483 "movq\t{$src, $dst|$dst, $src}",
4484 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4485 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4486 "movq\t{$src, $dst|$dst, $src}",
4487 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4490 let AddedComplexity = 20 in
4491 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4492 "vmovq\t{$src, $dst|$dst, $src}",
4494 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4495 (loadi64 addr:$src))))))],
4497 XS, VEX, Requires<[HasAVX]>;
4499 let AddedComplexity = 20 in
4500 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4501 "movq\t{$src, $dst|$dst, $src}",
4503 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4504 (loadi64 addr:$src))))))],
4506 XS, Requires<[UseSSE2]>;
4508 let Predicates = [HasAVX], AddedComplexity = 20 in {
4509 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4510 (VMOVZQI2PQIrm addr:$src)>;
4511 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4512 (VMOVZQI2PQIrm addr:$src)>;
4513 def : Pat<(v2i64 (X86vzload addr:$src)),
4514 (VMOVZQI2PQIrm addr:$src)>;
4517 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4518 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4519 (MOVZQI2PQIrm addr:$src)>;
4520 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4521 (MOVZQI2PQIrm addr:$src)>;
4522 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4525 let Predicates = [HasAVX] in {
4526 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4527 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4528 def : Pat<(v4i64 (X86vzload addr:$src)),
4529 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4532 //===---------------------------------------------------------------------===//
4533 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4534 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4536 let AddedComplexity = 15 in
4537 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4538 "vmovq\t{$src, $dst|$dst, $src}",
4539 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4541 XS, VEX, Requires<[HasAVX]>;
4542 let AddedComplexity = 15 in
4543 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4544 "movq\t{$src, $dst|$dst, $src}",
4545 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4547 XS, Requires<[UseSSE2]>;
4549 let AddedComplexity = 20 in
4550 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4551 "vmovq\t{$src, $dst|$dst, $src}",
4552 [(set VR128:$dst, (v2i64 (X86vzmovl
4553 (loadv2i64 addr:$src))))],
4555 XS, VEX, Requires<[HasAVX]>;
4556 let AddedComplexity = 20 in {
4557 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4558 "movq\t{$src, $dst|$dst, $src}",
4559 [(set VR128:$dst, (v2i64 (X86vzmovl
4560 (loadv2i64 addr:$src))))],
4562 XS, Requires<[UseSSE2]>;
4565 let AddedComplexity = 20 in {
4566 let Predicates = [HasAVX] in {
4567 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4568 (VMOVZPQILo2PQIrm addr:$src)>;
4569 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4570 (VMOVZPQILo2PQIrr VR128:$src)>;
4572 let Predicates = [UseSSE2] in {
4573 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4574 (MOVZPQILo2PQIrm addr:$src)>;
4575 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4576 (MOVZPQILo2PQIrr VR128:$src)>;
4580 // Instructions to match in the assembler
4581 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4582 "movq\t{$src, $dst|$dst, $src}", [],
4583 IIC_SSE_MOVDQ>, VEX, VEX_W;
4584 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4585 "movq\t{$src, $dst|$dst, $src}", [],
4586 IIC_SSE_MOVDQ>, VEX, VEX_W;
4587 // Recognize "movd" with GR64 destination, but encode as a "movq"
4588 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4589 "movd\t{$src, $dst|$dst, $src}", [],
4590 IIC_SSE_MOVDQ>, VEX, VEX_W;
4592 // Instructions for the disassembler
4593 // xr = XMM register
4596 let Predicates = [HasAVX] in
4597 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4598 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4599 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4600 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4602 //===---------------------------------------------------------------------===//
4603 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4604 //===---------------------------------------------------------------------===//
4605 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4606 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4607 X86MemOperand x86memop> {
4608 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4610 [(set RC:$dst, (vt (OpNode RC:$src)))],
4612 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4614 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4618 let Predicates = [HasAVX] in {
4619 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4620 v4f32, VR128, memopv4f32, f128mem>, VEX;
4621 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4622 v4f32, VR128, memopv4f32, f128mem>, VEX;
4623 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4624 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4625 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4626 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4628 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4629 memopv4f32, f128mem>;
4630 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4631 memopv4f32, f128mem>;
4633 let Predicates = [HasAVX] in {
4634 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4635 (VMOVSHDUPrr VR128:$src)>;
4636 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4637 (VMOVSHDUPrm addr:$src)>;
4638 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4639 (VMOVSLDUPrr VR128:$src)>;
4640 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4641 (VMOVSLDUPrm addr:$src)>;
4642 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4643 (VMOVSHDUPYrr VR256:$src)>;
4644 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4645 (VMOVSHDUPYrm addr:$src)>;
4646 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4647 (VMOVSLDUPYrr VR256:$src)>;
4648 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4649 (VMOVSLDUPYrm addr:$src)>;
4652 let Predicates = [UseSSE3] in {
4653 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4654 (MOVSHDUPrr VR128:$src)>;
4655 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4656 (MOVSHDUPrm addr:$src)>;
4657 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4658 (MOVSLDUPrr VR128:$src)>;
4659 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4660 (MOVSLDUPrm addr:$src)>;
4663 //===---------------------------------------------------------------------===//
4664 // SSE3 - Replicate Double FP - MOVDDUP
4665 //===---------------------------------------------------------------------===//
4667 multiclass sse3_replicate_dfp<string OpcodeStr> {
4668 let neverHasSideEffects = 1 in
4669 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4670 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4671 [], IIC_SSE_MOV_LH>;
4672 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4673 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4676 (scalar_to_vector (loadf64 addr:$src)))))],
4680 // FIXME: Merge with above classe when there're patterns for the ymm version
4681 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4682 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4683 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4684 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4685 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4686 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4689 (scalar_to_vector (loadf64 addr:$src)))))]>;
4692 let Predicates = [HasAVX] in {
4693 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4694 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4697 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4699 let Predicates = [HasAVX] in {
4700 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4701 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4702 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4703 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4704 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4705 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4706 def : Pat<(X86Movddup (bc_v2f64
4707 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4708 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4711 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4712 (VMOVDDUPYrm addr:$src)>;
4713 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4714 (VMOVDDUPYrm addr:$src)>;
4715 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4716 (VMOVDDUPYrm addr:$src)>;
4717 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4718 (VMOVDDUPYrr VR256:$src)>;
4721 let Predicates = [UseSSE3] in {
4722 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4723 (MOVDDUPrm addr:$src)>;
4724 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4725 (MOVDDUPrm addr:$src)>;
4726 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4727 (MOVDDUPrm addr:$src)>;
4728 def : Pat<(X86Movddup (bc_v2f64
4729 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4730 (MOVDDUPrm addr:$src)>;
4733 //===---------------------------------------------------------------------===//
4734 // SSE3 - Move Unaligned Integer
4735 //===---------------------------------------------------------------------===//
4737 let Predicates = [HasAVX] in {
4738 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4739 "vlddqu\t{$src, $dst|$dst, $src}",
4740 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4741 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4742 "vlddqu\t{$src, $dst|$dst, $src}",
4743 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4746 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4747 "lddqu\t{$src, $dst|$dst, $src}",
4748 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4751 //===---------------------------------------------------------------------===//
4752 // SSE3 - Arithmetic
4753 //===---------------------------------------------------------------------===//
4755 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4756 X86MemOperand x86memop, OpndItins itins,
4758 def rr : I<0xD0, MRMSrcReg,
4759 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4761 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4762 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4763 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4764 def rm : I<0xD0, MRMSrcMem,
4765 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4767 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4768 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4769 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4772 let Predicates = [HasAVX] in {
4773 let ExeDomain = SSEPackedSingle in {
4774 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4775 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4776 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4777 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4779 let ExeDomain = SSEPackedDouble in {
4780 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4781 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4782 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4783 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4786 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4787 let ExeDomain = SSEPackedSingle in
4788 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4789 f128mem, SSE_ALU_F32P>, TB, XD;
4790 let ExeDomain = SSEPackedDouble in
4791 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4792 f128mem, SSE_ALU_F64P>, TB, OpSize;
4795 //===---------------------------------------------------------------------===//
4796 // SSE3 Instructions
4797 //===---------------------------------------------------------------------===//
4800 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4801 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4802 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$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:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4808 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4810 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4811 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4812 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4813 IIC_SSE_HADDSUB_RM>;
4815 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4816 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4817 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4819 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4820 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4821 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4823 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4825 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4826 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4827 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4828 IIC_SSE_HADDSUB_RM>;
4831 let Predicates = [HasAVX] in {
4832 let ExeDomain = SSEPackedSingle in {
4833 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4834 X86fhadd, 0>, VEX_4V;
4835 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4836 X86fhsub, 0>, VEX_4V;
4837 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4838 X86fhadd, 0>, VEX_4V, VEX_L;
4839 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4840 X86fhsub, 0>, VEX_4V, VEX_L;
4842 let ExeDomain = SSEPackedDouble in {
4843 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4844 X86fhadd, 0>, VEX_4V;
4845 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4846 X86fhsub, 0>, VEX_4V;
4847 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4848 X86fhadd, 0>, VEX_4V, VEX_L;
4849 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4850 X86fhsub, 0>, VEX_4V, VEX_L;
4854 let Constraints = "$src1 = $dst" in {
4855 let ExeDomain = SSEPackedSingle in {
4856 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4857 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4859 let ExeDomain = SSEPackedDouble in {
4860 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4861 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4865 //===---------------------------------------------------------------------===//
4866 // SSSE3 - Packed Absolute Instructions
4867 //===---------------------------------------------------------------------===//
4870 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4871 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4872 Intrinsic IntId128> {
4873 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4875 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4876 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
4879 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4881 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4884 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
4888 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4889 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4890 Intrinsic IntId256> {
4891 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4893 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4894 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4897 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4899 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4902 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4905 let Predicates = [HasAVX] in {
4906 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4907 int_x86_ssse3_pabs_b_128>, VEX;
4908 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4909 int_x86_ssse3_pabs_w_128>, VEX;
4910 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4911 int_x86_ssse3_pabs_d_128>, VEX;
4914 let Predicates = [HasAVX2] in {
4915 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4916 int_x86_avx2_pabs_b>, VEX, VEX_L;
4917 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4918 int_x86_avx2_pabs_w>, VEX, VEX_L;
4919 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4920 int_x86_avx2_pabs_d>, VEX, VEX_L;
4923 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4924 int_x86_ssse3_pabs_b_128>;
4925 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4926 int_x86_ssse3_pabs_w_128>;
4927 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4928 int_x86_ssse3_pabs_d_128>;
4930 //===---------------------------------------------------------------------===//
4931 // SSSE3 - Packed Binary Operator Instructions
4932 //===---------------------------------------------------------------------===//
4934 def SSE_PHADDSUBD : OpndItins<
4935 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
4937 def SSE_PHADDSUBSW : OpndItins<
4938 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
4940 def SSE_PHADDSUBW : OpndItins<
4941 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
4943 def SSE_PSHUFB : OpndItins<
4944 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
4946 def SSE_PSIGN : OpndItins<
4947 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
4949 def SSE_PMULHRSW : OpndItins<
4950 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
4953 /// SS3I_binop_rm - Simple SSSE3 bin op
4954 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4955 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
4956 X86MemOperand x86memop, OpndItins itins,
4958 let isCommutable = 1 in
4959 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
4960 (ins RC:$src1, RC:$src2),
4962 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4963 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4964 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
4966 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
4967 (ins RC:$src1, x86memop:$src2),
4969 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4970 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4972 (OpVT (OpNode RC:$src1,
4973 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
4976 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4977 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4978 Intrinsic IntId128, OpndItins itins,
4980 let isCommutable = 1 in
4981 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4982 (ins VR128:$src1, VR128:$src2),
4984 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4985 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4986 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4988 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4989 (ins VR128:$src1, i128mem:$src2),
4991 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4992 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4994 (IntId128 VR128:$src1,
4995 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
4998 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
4999 Intrinsic IntId256> {
5000 let isCommutable = 1 in
5001 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5002 (ins VR256:$src1, VR256:$src2),
5003 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5004 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5006 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5007 (ins VR256:$src1, i256mem:$src2),
5008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5010 (IntId256 VR256:$src1,
5011 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5014 let ImmT = NoImm, Predicates = [HasAVX] in {
5015 let isCommutable = 0 in {
5016 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5017 memopv2i64, i128mem,
5018 SSE_PHADDSUBW, 0>, VEX_4V;
5019 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5020 memopv2i64, i128mem,
5021 SSE_PHADDSUBD, 0>, VEX_4V;
5022 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5023 memopv2i64, i128mem,
5024 SSE_PHADDSUBW, 0>, VEX_4V;
5025 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5026 memopv2i64, i128mem,
5027 SSE_PHADDSUBD, 0>, VEX_4V;
5028 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5029 memopv2i64, i128mem,
5030 SSE_PSIGN, 0>, VEX_4V;
5031 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5032 memopv2i64, i128mem,
5033 SSE_PSIGN, 0>, VEX_4V;
5034 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5035 memopv2i64, i128mem,
5036 SSE_PSIGN, 0>, VEX_4V;
5037 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5038 memopv2i64, i128mem,
5039 SSE_PSHUFB, 0>, VEX_4V;
5040 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5041 int_x86_ssse3_phadd_sw_128,
5042 SSE_PHADDSUBSW, 0>, VEX_4V;
5043 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5044 int_x86_ssse3_phsub_sw_128,
5045 SSE_PHADDSUBSW, 0>, VEX_4V;
5046 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5047 int_x86_ssse3_pmadd_ub_sw_128,
5048 SSE_PMADD, 0>, VEX_4V;
5050 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5051 int_x86_ssse3_pmul_hr_sw_128,
5052 SSE_PMULHRSW, 0>, VEX_4V;
5055 let ImmT = NoImm, Predicates = [HasAVX2] in {
5056 let isCommutable = 0 in {
5057 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5058 memopv4i64, i256mem,
5059 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5060 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5061 memopv4i64, i256mem,
5062 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5063 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5064 memopv4i64, i256mem,
5065 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5066 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5067 memopv4i64, i256mem,
5068 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5069 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5070 memopv4i64, i256mem,
5071 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5072 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5073 memopv4i64, i256mem,
5074 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5075 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5076 memopv4i64, i256mem,
5077 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5078 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5079 memopv4i64, i256mem,
5080 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5081 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5082 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5083 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5084 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5085 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5086 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5088 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5089 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5092 // None of these have i8 immediate fields.
5093 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5094 let isCommutable = 0 in {
5095 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5096 memopv2i64, i128mem, SSE_PHADDSUBW>;
5097 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5098 memopv2i64, i128mem, SSE_PHADDSUBD>;
5099 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5100 memopv2i64, i128mem, SSE_PHADDSUBW>;
5101 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5102 memopv2i64, i128mem, SSE_PHADDSUBD>;
5103 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5104 memopv2i64, i128mem, SSE_PSIGN>;
5105 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5106 memopv2i64, i128mem, SSE_PSIGN>;
5107 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5108 memopv2i64, i128mem, SSE_PSIGN>;
5109 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5110 memopv2i64, i128mem, SSE_PSHUFB>;
5111 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5112 int_x86_ssse3_phadd_sw_128,
5114 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5115 int_x86_ssse3_phsub_sw_128,
5117 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5118 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5120 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5121 int_x86_ssse3_pmul_hr_sw_128,
5125 //===---------------------------------------------------------------------===//
5126 // SSSE3 - Packed Align Instruction Patterns
5127 //===---------------------------------------------------------------------===//
5129 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5130 let neverHasSideEffects = 1 in {
5131 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5132 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5134 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5136 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5137 [], IIC_SSE_PALIGNR>, OpSize;
5139 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5140 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5142 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5144 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5145 [], IIC_SSE_PALIGNR>, OpSize;
5149 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5150 let neverHasSideEffects = 1 in {
5151 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5152 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5154 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5157 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5158 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5160 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5165 let Predicates = [HasAVX] in
5166 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5167 let Predicates = [HasAVX2] in
5168 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
5169 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5170 defm PALIGN : ssse3_palign<"palignr">;
5172 let Predicates = [HasAVX2] in {
5173 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5174 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5175 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5176 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5177 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5178 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5179 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5180 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5183 let Predicates = [HasAVX] in {
5184 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5185 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5186 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5187 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5188 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5189 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5190 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5191 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5194 let Predicates = [UseSSSE3] in {
5195 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5196 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5197 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5198 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5199 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5200 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5201 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5202 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5205 //===---------------------------------------------------------------------===//
5206 // SSSE3 - Thread synchronization
5207 //===---------------------------------------------------------------------===//
5209 let usesCustomInserter = 1 in {
5210 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5211 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5212 Requires<[HasSSE3]>;
5215 let Uses = [EAX, ECX, EDX] in
5216 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5217 TB, Requires<[HasSSE3]>;
5218 let Uses = [ECX, EAX] in
5219 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5220 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5221 TB, Requires<[HasSSE3]>;
5223 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5224 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5226 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5227 Requires<[In32BitMode]>;
5228 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5229 Requires<[In64BitMode]>;
5231 //===----------------------------------------------------------------------===//
5232 // SSE4.1 - Packed Move with Sign/Zero Extend
5233 //===----------------------------------------------------------------------===//
5235 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5236 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5237 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5238 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5240 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5241 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5243 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5247 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5249 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5250 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5251 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5253 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5254 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5255 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5258 let Predicates = [HasAVX] in {
5259 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5261 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5263 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5265 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5267 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5269 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5273 let Predicates = [HasAVX2] in {
5274 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5275 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5276 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5277 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5278 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5279 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5280 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5281 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5282 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5283 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5284 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5285 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5288 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5289 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5290 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5291 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5292 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5293 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5295 let Predicates = [HasAVX] in {
5296 // Common patterns involving scalar load.
5297 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5298 (VPMOVSXBWrm addr:$src)>;
5299 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5300 (VPMOVSXBWrm addr:$src)>;
5301 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5302 (VPMOVSXBWrm addr:$src)>;
5304 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5305 (VPMOVSXWDrm addr:$src)>;
5306 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5307 (VPMOVSXWDrm addr:$src)>;
5308 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5309 (VPMOVSXWDrm addr:$src)>;
5311 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5312 (VPMOVSXDQrm addr:$src)>;
5313 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5314 (VPMOVSXDQrm addr:$src)>;
5315 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5316 (VPMOVSXDQrm addr:$src)>;
5318 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5319 (VPMOVZXBWrm addr:$src)>;
5320 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5321 (VPMOVZXBWrm addr:$src)>;
5322 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5323 (VPMOVZXBWrm addr:$src)>;
5325 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5326 (VPMOVZXWDrm addr:$src)>;
5327 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5328 (VPMOVZXWDrm addr:$src)>;
5329 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5330 (VPMOVZXWDrm addr:$src)>;
5332 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5333 (VPMOVZXDQrm addr:$src)>;
5334 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5335 (VPMOVZXDQrm addr:$src)>;
5336 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5337 (VPMOVZXDQrm addr:$src)>;
5340 let Predicates = [UseSSE41] in {
5341 // Common patterns involving scalar load.
5342 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5343 (PMOVSXBWrm addr:$src)>;
5344 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5345 (PMOVSXBWrm addr:$src)>;
5346 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5347 (PMOVSXBWrm addr:$src)>;
5349 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5350 (PMOVSXWDrm addr:$src)>;
5351 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5352 (PMOVSXWDrm addr:$src)>;
5353 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5354 (PMOVSXWDrm addr:$src)>;
5356 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5357 (PMOVSXDQrm addr:$src)>;
5358 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5359 (PMOVSXDQrm addr:$src)>;
5360 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5361 (PMOVSXDQrm addr:$src)>;
5363 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5364 (PMOVZXBWrm addr:$src)>;
5365 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5366 (PMOVZXBWrm addr:$src)>;
5367 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5368 (PMOVZXBWrm addr:$src)>;
5370 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5371 (PMOVZXWDrm addr:$src)>;
5372 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5373 (PMOVZXWDrm addr:$src)>;
5374 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5375 (PMOVZXWDrm addr:$src)>;
5377 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5378 (PMOVZXDQrm addr:$src)>;
5379 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5380 (PMOVZXDQrm addr:$src)>;
5381 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5382 (PMOVZXDQrm addr:$src)>;
5385 let Predicates = [HasAVX2] in {
5386 let AddedComplexity = 15 in {
5387 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5388 (VPMOVZXDQYrr VR128:$src)>;
5389 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5390 (VPMOVZXWDYrr VR128:$src)>;
5393 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5394 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5397 let Predicates = [HasAVX] in {
5398 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5399 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5402 let Predicates = [UseSSE41] in {
5403 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5404 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5408 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5409 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5410 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5411 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5413 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5414 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5416 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5420 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5422 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5423 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5424 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5426 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5427 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5429 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5433 let Predicates = [HasAVX] in {
5434 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5436 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5438 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5440 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5444 let Predicates = [HasAVX2] in {
5445 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5446 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5447 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5448 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5449 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5450 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5451 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5452 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5455 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5456 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5457 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5458 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5460 let Predicates = [HasAVX] in {
5461 // Common patterns involving scalar load
5462 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5463 (VPMOVSXBDrm addr:$src)>;
5464 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5465 (VPMOVSXWQrm addr:$src)>;
5467 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5468 (VPMOVZXBDrm addr:$src)>;
5469 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5470 (VPMOVZXWQrm addr:$src)>;
5473 let Predicates = [UseSSE41] in {
5474 // Common patterns involving scalar load
5475 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5476 (PMOVSXBDrm addr:$src)>;
5477 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5478 (PMOVSXWQrm addr:$src)>;
5480 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5481 (PMOVZXBDrm addr:$src)>;
5482 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5483 (PMOVZXWQrm addr:$src)>;
5486 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5487 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5488 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5489 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5491 // Expecting a i16 load any extended to i32 value.
5492 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5493 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5494 [(set VR128:$dst, (IntId (bitconvert
5495 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5499 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5501 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5502 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5503 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5505 // Expecting a i16 load any extended to i32 value.
5506 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5507 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5508 [(set VR256:$dst, (IntId (bitconvert
5509 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5513 let Predicates = [HasAVX] in {
5514 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5516 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5519 let Predicates = [HasAVX2] in {
5520 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5521 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5522 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5523 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5525 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5526 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5528 let Predicates = [HasAVX2] in {
5529 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5530 (VPMOVSXWDYrm addr:$src)>;
5531 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5532 (VPMOVSXDQYrm addr:$src)>;
5534 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5535 (scalar_to_vector (loadi64 addr:$src))))))),
5536 (VPMOVSXBDYrm addr:$src)>;
5537 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5538 (scalar_to_vector (loadf64 addr:$src))))))),
5539 (VPMOVSXBDYrm addr:$src)>;
5541 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5542 (scalar_to_vector (loadi64 addr:$src))))))),
5543 (VPMOVSXWQYrm addr:$src)>;
5544 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5545 (scalar_to_vector (loadf64 addr:$src))))))),
5546 (VPMOVSXWQYrm addr:$src)>;
5548 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5549 (scalar_to_vector (loadi32 addr:$src))))))),
5550 (VPMOVSXBQYrm addr:$src)>;
5553 let Predicates = [HasAVX] in {
5554 // Common patterns involving scalar load
5555 def : Pat<(int_x86_sse41_pmovsxbq
5556 (bitconvert (v4i32 (X86vzmovl
5557 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5558 (VPMOVSXBQrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovzxbq
5561 (bitconvert (v4i32 (X86vzmovl
5562 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5563 (VPMOVZXBQrm addr:$src)>;
5566 let Predicates = [UseSSE41] in {
5567 // Common patterns involving scalar load
5568 def : Pat<(int_x86_sse41_pmovsxbq
5569 (bitconvert (v4i32 (X86vzmovl
5570 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5571 (PMOVSXBQrm addr:$src)>;
5573 def : Pat<(int_x86_sse41_pmovzxbq
5574 (bitconvert (v4i32 (X86vzmovl
5575 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5576 (PMOVZXBQrm addr:$src)>;
5578 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5579 (scalar_to_vector (loadi64 addr:$src))))))),
5580 (PMOVSXWDrm addr:$src)>;
5581 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5582 (scalar_to_vector (loadf64 addr:$src))))))),
5583 (PMOVSXWDrm addr:$src)>;
5584 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5585 (scalar_to_vector (loadi32 addr:$src))))))),
5586 (PMOVSXBDrm addr:$src)>;
5587 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5588 (scalar_to_vector (loadi32 addr:$src))))))),
5589 (PMOVSXWQrm addr:$src)>;
5590 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5591 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5592 (PMOVSXBQrm addr:$src)>;
5593 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5594 (scalar_to_vector (loadi64 addr:$src))))))),
5595 (PMOVSXDQrm addr:$src)>;
5596 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5597 (scalar_to_vector (loadf64 addr:$src))))))),
5598 (PMOVSXDQrm addr:$src)>;
5599 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5600 (scalar_to_vector (loadi64 addr:$src))))))),
5601 (PMOVSXBWrm addr:$src)>;
5602 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5603 (scalar_to_vector (loadf64 addr:$src))))))),
5604 (PMOVSXBWrm addr:$src)>;
5607 let Predicates = [HasAVX2] in {
5608 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5609 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5610 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5612 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5613 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5615 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5617 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5618 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5619 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5620 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5621 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5622 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5624 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5625 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5626 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5627 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5629 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5630 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5633 let Predicates = [HasAVX] in {
5634 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5635 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5636 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5638 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5639 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5641 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5643 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5644 (VPMOVZXBWrm addr:$src)>;
5645 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5646 (VPMOVZXBWrm addr:$src)>;
5647 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5648 (VPMOVZXBDrm addr:$src)>;
5649 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5650 (VPMOVZXBQrm addr:$src)>;
5652 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5653 (VPMOVZXWDrm addr:$src)>;
5654 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5655 (VPMOVZXWDrm addr:$src)>;
5656 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5657 (VPMOVZXWQrm addr:$src)>;
5659 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5660 (VPMOVZXDQrm addr:$src)>;
5661 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5662 (VPMOVZXDQrm addr:$src)>;
5663 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5664 (VPMOVZXDQrm addr:$src)>;
5666 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5667 (scalar_to_vector (loadi64 addr:$src))))))),
5668 (VPMOVSXWDrm addr:$src)>;
5669 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5670 (scalar_to_vector (loadi64 addr:$src))))))),
5671 (VPMOVSXDQrm addr:$src)>;
5672 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5673 (scalar_to_vector (loadf64 addr:$src))))))),
5674 (VPMOVSXWDrm addr:$src)>;
5675 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5676 (scalar_to_vector (loadf64 addr:$src))))))),
5677 (VPMOVSXDQrm addr:$src)>;
5678 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5679 (scalar_to_vector (loadi64 addr:$src))))))),
5680 (VPMOVSXBWrm addr:$src)>;
5681 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5682 (scalar_to_vector (loadf64 addr:$src))))))),
5683 (VPMOVSXBWrm addr:$src)>;
5685 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5686 (scalar_to_vector (loadi32 addr:$src))))))),
5687 (VPMOVSXBDrm addr:$src)>;
5688 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5689 (scalar_to_vector (loadi32 addr:$src))))))),
5690 (VPMOVSXWQrm addr:$src)>;
5691 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5692 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5693 (VPMOVSXBQrm addr:$src)>;
5696 let Predicates = [UseSSE41] in {
5697 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5698 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5699 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5701 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5702 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5704 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5706 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5707 (PMOVZXBWrm addr:$src)>;
5708 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5709 (PMOVZXBWrm addr:$src)>;
5710 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5711 (PMOVZXBDrm addr:$src)>;
5712 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5713 (PMOVZXBQrm addr:$src)>;
5715 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5716 (PMOVZXWDrm addr:$src)>;
5717 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5718 (PMOVZXWDrm addr:$src)>;
5719 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5720 (PMOVZXWQrm addr:$src)>;
5722 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5723 (PMOVZXDQrm addr:$src)>;
5724 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5725 (PMOVZXDQrm addr:$src)>;
5726 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5727 (PMOVZXDQrm addr:$src)>;
5730 //===----------------------------------------------------------------------===//
5731 // SSE4.1 - Extract Instructions
5732 //===----------------------------------------------------------------------===//
5734 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5735 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5736 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5737 (ins VR128:$src1, i32i8imm:$src2),
5738 !strconcat(OpcodeStr,
5739 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5740 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5742 let neverHasSideEffects = 1, mayStore = 1 in
5743 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5744 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5745 !strconcat(OpcodeStr,
5746 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5749 // There's an AssertZext in the way of writing the store pattern
5750 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5753 let Predicates = [HasAVX] in {
5754 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5755 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5756 (ins VR128:$src1, i32i8imm:$src2),
5757 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5760 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5763 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5764 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5765 let neverHasSideEffects = 1, mayStore = 1 in
5766 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5767 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5768 !strconcat(OpcodeStr,
5769 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5772 // There's an AssertZext in the way of writing the store pattern
5773 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5776 let Predicates = [HasAVX] in
5777 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5779 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5782 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5783 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5784 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5785 (ins VR128:$src1, i32i8imm:$src2),
5786 !strconcat(OpcodeStr,
5787 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5789 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5790 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5791 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5792 !strconcat(OpcodeStr,
5793 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5794 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5795 addr:$dst)]>, OpSize;
5798 let Predicates = [HasAVX] in
5799 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5801 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5803 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5804 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5805 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5806 (ins VR128:$src1, i32i8imm:$src2),
5807 !strconcat(OpcodeStr,
5808 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5810 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5811 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5812 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5813 !strconcat(OpcodeStr,
5814 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5815 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5816 addr:$dst)]>, OpSize, REX_W;
5819 let Predicates = [HasAVX] in
5820 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5822 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5824 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5826 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5827 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5828 (ins VR128:$src1, i32i8imm:$src2),
5829 !strconcat(OpcodeStr,
5830 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5832 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5834 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5835 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5836 !strconcat(OpcodeStr,
5837 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5838 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5839 addr:$dst)]>, OpSize;
5842 let ExeDomain = SSEPackedSingle in {
5843 let Predicates = [HasAVX] in {
5844 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5845 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5846 (ins VR128:$src1, i32i8imm:$src2),
5847 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5850 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5853 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5854 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5857 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5859 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5862 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5863 Requires<[UseSSE41]>;
5865 //===----------------------------------------------------------------------===//
5866 // SSE4.1 - Insert Instructions
5867 //===----------------------------------------------------------------------===//
5869 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5870 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5871 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5873 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5875 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5877 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5878 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5879 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5881 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5883 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5885 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5886 imm:$src3))]>, OpSize;
5889 let Predicates = [HasAVX] in
5890 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5891 let Constraints = "$src1 = $dst" in
5892 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5894 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5895 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5896 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5898 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5900 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5902 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5904 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5905 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5907 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5909 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5911 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5912 imm:$src3)))]>, OpSize;
5915 let Predicates = [HasAVX] in
5916 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5917 let Constraints = "$src1 = $dst" in
5918 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5920 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5921 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5922 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5924 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5926 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5928 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5930 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5931 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5933 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5935 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5937 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5938 imm:$src3)))]>, OpSize;
5941 let Predicates = [HasAVX] in
5942 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5943 let Constraints = "$src1 = $dst" in
5944 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5946 // insertps has a few different modes, there's the first two here below which
5947 // are optimized inserts that won't zero arbitrary elements in the destination
5948 // vector. The next one matches the intrinsic and could zero arbitrary elements
5949 // in the target vector.
5950 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5951 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5952 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5954 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5956 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5958 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5960 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5961 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5963 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5965 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5967 (X86insrtps VR128:$src1,
5968 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5969 imm:$src3))]>, OpSize;
5972 let ExeDomain = SSEPackedSingle in {
5973 let Predicates = [HasAVX] in
5974 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5975 let Constraints = "$src1 = $dst" in
5976 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5979 //===----------------------------------------------------------------------===//
5980 // SSE4.1 - Round Instructions
5981 //===----------------------------------------------------------------------===//
5983 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5984 X86MemOperand x86memop, RegisterClass RC,
5985 PatFrag mem_frag32, PatFrag mem_frag64,
5986 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5987 let ExeDomain = SSEPackedSingle in {
5988 // Intrinsic operation, reg.
5989 // Vector intrinsic operation, reg
5990 def PSr : SS4AIi8<opcps, MRMSrcReg,
5991 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5992 !strconcat(OpcodeStr,
5993 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5994 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5997 // Vector intrinsic operation, mem
5998 def PSm : SS4AIi8<opcps, MRMSrcMem,
5999 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6000 !strconcat(OpcodeStr,
6001 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6003 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6005 } // ExeDomain = SSEPackedSingle
6007 let ExeDomain = SSEPackedDouble in {
6008 // Vector intrinsic operation, reg
6009 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6010 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6011 !strconcat(OpcodeStr,
6012 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6013 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6016 // Vector intrinsic operation, mem
6017 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6018 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6019 !strconcat(OpcodeStr,
6020 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6022 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6024 } // ExeDomain = SSEPackedDouble
6027 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6030 Intrinsic F64Int, bit Is2Addr = 1> {
6031 let ExeDomain = GenericDomain in {
6033 let hasSideEffects = 0 in
6034 def SSr : SS4AIi8<opcss, MRMSrcReg,
6035 (outs FR32:$dst), (ins FR32:$src1, FR32:$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, $src3}")),
6043 // Intrinsic operation, reg.
6044 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6045 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6047 !strconcat(OpcodeStr,
6048 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6049 !strconcat(OpcodeStr,
6050 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6051 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6054 // Intrinsic operation, mem.
6055 def SSm : SS4AIi8<opcss, MRMSrcMem,
6056 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6058 !strconcat(OpcodeStr,
6059 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6060 !strconcat(OpcodeStr,
6061 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6063 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6067 let hasSideEffects = 0 in
6068 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6069 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$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, $src3}")),
6077 // Intrinsic operation, reg.
6078 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6079 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6081 !strconcat(OpcodeStr,
6082 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6083 !strconcat(OpcodeStr,
6084 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6085 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6088 // Intrinsic operation, mem.
6089 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6090 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6092 !strconcat(OpcodeStr,
6093 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6094 !strconcat(OpcodeStr,
6095 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6097 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6099 } // ExeDomain = GenericDomain
6102 // FP round - roundss, roundps, roundsd, roundpd
6103 let Predicates = [HasAVX] in {
6105 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6106 memopv4f32, memopv2f64,
6107 int_x86_sse41_round_ps,
6108 int_x86_sse41_round_pd>, VEX;
6109 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6110 memopv8f32, memopv4f64,
6111 int_x86_avx_round_ps_256,
6112 int_x86_avx_round_pd_256>, VEX, VEX_L;
6113 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6114 int_x86_sse41_round_ss,
6115 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6117 def : Pat<(ffloor FR32:$src),
6118 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6119 def : Pat<(f64 (ffloor FR64:$src)),
6120 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6121 def : Pat<(f32 (fnearbyint FR32:$src)),
6122 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6123 def : Pat<(f64 (fnearbyint FR64:$src)),
6124 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6125 def : Pat<(f32 (fceil FR32:$src)),
6126 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6127 def : Pat<(f64 (fceil FR64:$src)),
6128 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6129 def : Pat<(f32 (frint FR32:$src)),
6130 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6131 def : Pat<(f64 (frint FR64:$src)),
6132 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6133 def : Pat<(f32 (ftrunc FR32:$src)),
6134 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6135 def : Pat<(f64 (ftrunc FR64:$src)),
6136 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6138 def : Pat<(v4f32 (ffloor VR128:$src)),
6139 (VROUNDPSr VR128:$src, (i32 0x1))>;
6140 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6141 (VROUNDPSr VR128:$src, (i32 0xC))>;
6142 def : Pat<(v4f32 (fceil VR128:$src)),
6143 (VROUNDPSr VR128:$src, (i32 0x2))>;
6144 def : Pat<(v4f32 (frint VR128:$src)),
6145 (VROUNDPSr VR128:$src, (i32 0x4))>;
6146 def : Pat<(v4f32 (ftrunc VR128:$src)),
6147 (VROUNDPSr VR128:$src, (i32 0x3))>;
6149 def : Pat<(v2f64 (ffloor VR128:$src)),
6150 (VROUNDPDr VR128:$src, (i32 0x1))>;
6151 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6152 (VROUNDPDr VR128:$src, (i32 0xC))>;
6153 def : Pat<(v2f64 (fceil VR128:$src)),
6154 (VROUNDPDr VR128:$src, (i32 0x2))>;
6155 def : Pat<(v2f64 (frint VR128:$src)),
6156 (VROUNDPDr VR128:$src, (i32 0x4))>;
6157 def : Pat<(v2f64 (ftrunc VR128:$src)),
6158 (VROUNDPDr VR128:$src, (i32 0x3))>;
6160 def : Pat<(v8f32 (ffloor VR256:$src)),
6161 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6162 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6163 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6164 def : Pat<(v8f32 (fceil VR256:$src)),
6165 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6166 def : Pat<(v8f32 (frint VR256:$src)),
6167 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6168 def : Pat<(v8f32 (ftrunc VR256:$src)),
6169 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6171 def : Pat<(v4f64 (ffloor VR256:$src)),
6172 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6173 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6174 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6175 def : Pat<(v4f64 (fceil VR256:$src)),
6176 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6177 def : Pat<(v4f64 (frint VR256:$src)),
6178 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6179 def : Pat<(v4f64 (ftrunc VR256:$src)),
6180 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6183 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6184 memopv4f32, memopv2f64,
6185 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6186 let Constraints = "$src1 = $dst" in
6187 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6188 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6190 let Predicates = [UseSSE41] in {
6191 def : Pat<(ffloor FR32:$src),
6192 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6193 def : Pat<(f64 (ffloor FR64:$src)),
6194 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6195 def : Pat<(f32 (fnearbyint FR32:$src)),
6196 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6197 def : Pat<(f64 (fnearbyint FR64:$src)),
6198 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6199 def : Pat<(f32 (fceil FR32:$src)),
6200 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6201 def : Pat<(f64 (fceil FR64:$src)),
6202 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6203 def : Pat<(f32 (frint FR32:$src)),
6204 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6205 def : Pat<(f64 (frint FR64:$src)),
6206 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6207 def : Pat<(f32 (ftrunc FR32:$src)),
6208 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6209 def : Pat<(f64 (ftrunc FR64:$src)),
6210 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6212 def : Pat<(v4f32 (ffloor VR128:$src)),
6213 (ROUNDPSr VR128:$src, (i32 0x1))>;
6214 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6215 (ROUNDPSr VR128:$src, (i32 0xC))>;
6216 def : Pat<(v4f32 (fceil VR128:$src)),
6217 (ROUNDPSr VR128:$src, (i32 0x2))>;
6218 def : Pat<(v4f32 (frint VR128:$src)),
6219 (ROUNDPSr VR128:$src, (i32 0x4))>;
6220 def : Pat<(v4f32 (ftrunc VR128:$src)),
6221 (ROUNDPSr VR128:$src, (i32 0x3))>;
6223 def : Pat<(v2f64 (ffloor VR128:$src)),
6224 (ROUNDPDr VR128:$src, (i32 0x1))>;
6225 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6226 (ROUNDPDr VR128:$src, (i32 0xC))>;
6227 def : Pat<(v2f64 (fceil VR128:$src)),
6228 (ROUNDPDr VR128:$src, (i32 0x2))>;
6229 def : Pat<(v2f64 (frint VR128:$src)),
6230 (ROUNDPDr VR128:$src, (i32 0x4))>;
6231 def : Pat<(v2f64 (ftrunc VR128:$src)),
6232 (ROUNDPDr VR128:$src, (i32 0x3))>;
6235 //===----------------------------------------------------------------------===//
6236 // SSE4.1 - Packed Bit Test
6237 //===----------------------------------------------------------------------===//
6239 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6240 // the intel intrinsic that corresponds to this.
6241 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6242 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6243 "vptest\t{$src2, $src1|$src1, $src2}",
6244 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6246 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6247 "vptest\t{$src2, $src1|$src1, $src2}",
6248 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6251 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6252 "vptest\t{$src2, $src1|$src1, $src2}",
6253 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6255 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6256 "vptest\t{$src2, $src1|$src1, $src2}",
6257 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6261 let Defs = [EFLAGS] in {
6262 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6263 "ptest\t{$src2, $src1|$src1, $src2}",
6264 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6266 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6267 "ptest\t{$src2, $src1|$src1, $src2}",
6268 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6272 // The bit test instructions below are AVX only
6273 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6274 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6275 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6276 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6277 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6278 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6279 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6280 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6284 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6285 let ExeDomain = SSEPackedSingle in {
6286 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6287 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6290 let ExeDomain = SSEPackedDouble in {
6291 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6292 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6297 //===----------------------------------------------------------------------===//
6298 // SSE4.1 - Misc Instructions
6299 //===----------------------------------------------------------------------===//
6301 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6302 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6303 "popcnt{w}\t{$src, $dst|$dst, $src}",
6304 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6306 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6307 "popcnt{w}\t{$src, $dst|$dst, $src}",
6308 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6309 (implicit EFLAGS)]>, OpSize, XS;
6311 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6312 "popcnt{l}\t{$src, $dst|$dst, $src}",
6313 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6315 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6316 "popcnt{l}\t{$src, $dst|$dst, $src}",
6317 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6318 (implicit EFLAGS)]>, XS;
6320 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6321 "popcnt{q}\t{$src, $dst|$dst, $src}",
6322 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6324 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6325 "popcnt{q}\t{$src, $dst|$dst, $src}",
6326 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6327 (implicit EFLAGS)]>, XS;
6332 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6333 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6334 Intrinsic IntId128> {
6335 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6337 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6338 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6339 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6341 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6344 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6347 let Predicates = [HasAVX] in
6348 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6349 int_x86_sse41_phminposuw>, VEX;
6350 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6351 int_x86_sse41_phminposuw>;
6353 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6354 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6355 Intrinsic IntId128, bit Is2Addr = 1> {
6356 let isCommutable = 1 in
6357 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6358 (ins VR128:$src1, VR128:$src2),
6360 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6362 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6363 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6364 (ins VR128:$src1, i128mem:$src2),
6366 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6367 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6369 (IntId128 VR128:$src1,
6370 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6373 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6374 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6375 Intrinsic IntId256> {
6376 let isCommutable = 1 in
6377 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6378 (ins VR256:$src1, VR256:$src2),
6379 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6380 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6381 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6382 (ins VR256:$src1, i256mem:$src2),
6383 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6385 (IntId256 VR256:$src1,
6386 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6390 /// SS48I_binop_rm - Simple SSE41 binary operator.
6391 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6392 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6393 X86MemOperand x86memop, bit Is2Addr = 1> {
6394 let isCommutable = 1 in
6395 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6396 (ins RC:$src1, RC:$src2),
6398 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6399 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6400 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6401 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6402 (ins RC:$src1, x86memop:$src2),
6404 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6405 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6407 (OpVT (OpNode RC:$src1,
6408 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6411 let Predicates = [HasAVX] in {
6412 let isCommutable = 0 in
6413 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6415 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6416 memopv2i64, i128mem, 0>, VEX_4V;
6417 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6418 memopv2i64, i128mem, 0>, VEX_4V;
6419 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6420 memopv2i64, i128mem, 0>, VEX_4V;
6421 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6422 memopv2i64, i128mem, 0>, VEX_4V;
6423 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6424 memopv2i64, i128mem, 0>, VEX_4V;
6425 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6426 memopv2i64, i128mem, 0>, VEX_4V;
6427 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6428 memopv2i64, i128mem, 0>, VEX_4V;
6429 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6430 memopv2i64, i128mem, 0>, VEX_4V;
6431 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6435 let Predicates = [HasAVX2] in {
6436 let isCommutable = 0 in
6437 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6438 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6439 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6440 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6441 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6442 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6443 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6444 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6445 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6446 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6447 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6448 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6449 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6450 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6451 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6452 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6453 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6454 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6455 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6456 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6459 let Constraints = "$src1 = $dst" in {
6460 let isCommutable = 0 in
6461 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6462 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6463 memopv2i64, i128mem>;
6464 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6465 memopv2i64, i128mem>;
6466 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6467 memopv2i64, i128mem>;
6468 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6469 memopv2i64, i128mem>;
6470 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6471 memopv2i64, i128mem>;
6472 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6473 memopv2i64, i128mem>;
6474 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6475 memopv2i64, i128mem>;
6476 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6477 memopv2i64, i128mem>;
6478 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6481 let Predicates = [HasAVX] in {
6482 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6483 memopv2i64, i128mem, 0>, VEX_4V;
6484 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6485 memopv2i64, i128mem, 0>, VEX_4V;
6487 let Predicates = [HasAVX2] in {
6488 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6489 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6490 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6491 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6494 let Constraints = "$src1 = $dst" in {
6495 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6496 memopv2i64, i128mem>;
6497 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6498 memopv2i64, i128mem>;
6501 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6502 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6503 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6504 X86MemOperand x86memop, bit Is2Addr = 1> {
6505 let isCommutable = 1 in
6506 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6507 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6509 !strconcat(OpcodeStr,
6510 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6511 !strconcat(OpcodeStr,
6512 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6513 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6515 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6516 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6518 !strconcat(OpcodeStr,
6519 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6520 !strconcat(OpcodeStr,
6521 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6524 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6528 let Predicates = [HasAVX] in {
6529 let isCommutable = 0 in {
6530 let ExeDomain = SSEPackedSingle in {
6531 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6532 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6533 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6534 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6535 f256mem, 0>, VEX_4V, VEX_L;
6537 let ExeDomain = SSEPackedDouble in {
6538 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6539 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6540 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6541 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6542 f256mem, 0>, VEX_4V, VEX_L;
6544 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6545 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6546 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6547 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6549 let ExeDomain = SSEPackedSingle in
6550 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6551 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6552 let ExeDomain = SSEPackedDouble in
6553 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6554 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6555 let ExeDomain = SSEPackedSingle in
6556 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6557 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6560 let Predicates = [HasAVX2] in {
6561 let isCommutable = 0 in {
6562 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6563 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6564 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6565 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6569 let Constraints = "$src1 = $dst" in {
6570 let isCommutable = 0 in {
6571 let ExeDomain = SSEPackedSingle in
6572 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6573 VR128, memopv4f32, f128mem>;
6574 let ExeDomain = SSEPackedDouble in
6575 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6576 VR128, memopv2f64, f128mem>;
6577 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6578 VR128, memopv2i64, i128mem>;
6579 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6580 VR128, memopv2i64, i128mem>;
6582 let ExeDomain = SSEPackedSingle in
6583 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6584 VR128, memopv4f32, f128mem>;
6585 let ExeDomain = SSEPackedDouble in
6586 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6587 VR128, memopv2f64, f128mem>;
6590 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6591 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6592 RegisterClass RC, X86MemOperand x86memop,
6593 PatFrag mem_frag, Intrinsic IntId> {
6594 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6595 (ins RC:$src1, RC:$src2, RC:$src3),
6596 !strconcat(OpcodeStr,
6597 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6598 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6599 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6601 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6602 (ins RC:$src1, x86memop:$src2, RC:$src3),
6603 !strconcat(OpcodeStr,
6604 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6606 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6608 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6611 let Predicates = [HasAVX] in {
6612 let ExeDomain = SSEPackedDouble in {
6613 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6614 memopv2f64, int_x86_sse41_blendvpd>;
6615 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6616 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6617 } // ExeDomain = SSEPackedDouble
6618 let ExeDomain = SSEPackedSingle in {
6619 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6620 memopv4f32, int_x86_sse41_blendvps>;
6621 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6622 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6623 } // ExeDomain = SSEPackedSingle
6624 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6625 memopv2i64, int_x86_sse41_pblendvb>;
6628 let Predicates = [HasAVX2] in {
6629 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6630 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6633 let Predicates = [HasAVX] in {
6634 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6635 (v16i8 VR128:$src2))),
6636 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6637 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6638 (v4i32 VR128:$src2))),
6639 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6640 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6641 (v4f32 VR128:$src2))),
6642 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6643 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6644 (v2i64 VR128:$src2))),
6645 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6646 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6647 (v2f64 VR128:$src2))),
6648 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6649 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6650 (v8i32 VR256:$src2))),
6651 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6652 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6653 (v8f32 VR256:$src2))),
6654 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6655 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6656 (v4i64 VR256:$src2))),
6657 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6658 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6659 (v4f64 VR256:$src2))),
6660 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6662 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6664 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6665 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6667 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6669 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6671 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6672 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6674 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6675 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6677 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6680 let Predicates = [HasAVX2] in {
6681 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6682 (v32i8 VR256:$src2))),
6683 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6684 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6686 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6689 /// SS41I_ternary_int - SSE 4.1 ternary operator
6690 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6691 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6692 X86MemOperand x86memop, Intrinsic IntId> {
6693 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6694 (ins VR128:$src1, VR128:$src2),
6695 !strconcat(OpcodeStr,
6696 "\t{$src2, $dst|$dst, $src2}"),
6697 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6700 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6701 (ins VR128:$src1, x86memop:$src2),
6702 !strconcat(OpcodeStr,
6703 "\t{$src2, $dst|$dst, $src2}"),
6706 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6710 let ExeDomain = SSEPackedDouble in
6711 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6712 int_x86_sse41_blendvpd>;
6713 let ExeDomain = SSEPackedSingle in
6714 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6715 int_x86_sse41_blendvps>;
6716 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6717 int_x86_sse41_pblendvb>;
6719 // Aliases with the implicit xmm0 argument
6720 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6721 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6722 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6723 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6724 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6725 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6726 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6727 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6728 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6729 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6730 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6731 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6733 let Predicates = [UseSSE41] in {
6734 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6735 (v16i8 VR128:$src2))),
6736 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6737 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6738 (v4i32 VR128:$src2))),
6739 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6740 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6741 (v4f32 VR128:$src2))),
6742 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6743 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6744 (v2i64 VR128:$src2))),
6745 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6746 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6747 (v2f64 VR128:$src2))),
6748 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6750 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6752 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6753 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6755 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6756 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6758 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6762 let Predicates = [HasAVX] in
6763 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6764 "vmovntdqa\t{$src, $dst|$dst, $src}",
6765 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6767 let Predicates = [HasAVX2] in
6768 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6769 "vmovntdqa\t{$src, $dst|$dst, $src}",
6770 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6772 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6773 "movntdqa\t{$src, $dst|$dst, $src}",
6774 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6777 //===----------------------------------------------------------------------===//
6778 // SSE4.2 - Compare Instructions
6779 //===----------------------------------------------------------------------===//
6781 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6782 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6783 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6784 X86MemOperand x86memop, bit Is2Addr = 1> {
6785 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6786 (ins RC:$src1, RC:$src2),
6788 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6789 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6790 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6792 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6793 (ins RC:$src1, x86memop:$src2),
6795 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6796 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6798 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6801 let Predicates = [HasAVX] in
6802 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6803 memopv2i64, i128mem, 0>, VEX_4V;
6805 let Predicates = [HasAVX2] in
6806 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6807 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6809 let Constraints = "$src1 = $dst" in
6810 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6811 memopv2i64, i128mem>;
6813 //===----------------------------------------------------------------------===//
6814 // SSE4.2 - String/text Processing Instructions
6815 //===----------------------------------------------------------------------===//
6817 // Packed Compare Implicit Length Strings, Return Mask
6818 multiclass pseudo_pcmpistrm<string asm> {
6819 def REG : PseudoI<(outs VR128:$dst),
6820 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6821 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6823 def MEM : PseudoI<(outs VR128:$dst),
6824 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6825 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
6826 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6829 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6830 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6831 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
6834 multiclass pcmpistrm_SS42AI<string asm> {
6835 def rr : SS42AI<0x62, MRMSrcReg, (outs),
6836 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6837 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6840 def rm :SS42AI<0x62, MRMSrcMem, (outs),
6841 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6842 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6846 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6847 let Predicates = [HasAVX] in
6848 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
6849 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
6852 // Packed Compare Explicit Length Strings, Return Mask
6853 multiclass pseudo_pcmpestrm<string asm> {
6854 def REG : PseudoI<(outs VR128:$dst),
6855 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6856 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6857 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6858 def MEM : PseudoI<(outs VR128:$dst),
6859 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6860 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
6861 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
6864 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6865 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6866 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
6869 multiclass SS42AI_pcmpestrm<string asm> {
6870 def rr : SS42AI<0x60, MRMSrcReg, (outs),
6871 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6872 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6875 def rm : SS42AI<0x60, MRMSrcMem, (outs),
6876 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6877 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6881 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6882 let Predicates = [HasAVX] in
6883 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
6884 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
6887 // Packed Compare Implicit Length Strings, Return Index
6888 multiclass pseudo_pcmpistri<string asm> {
6889 def REG : PseudoI<(outs GR32:$dst),
6890 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6891 [(set GR32:$dst, EFLAGS,
6892 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
6893 def MEM : PseudoI<(outs GR32:$dst),
6894 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6895 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
6896 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6899 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6900 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
6901 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
6904 multiclass SS42AI_pcmpistri<string asm> {
6905 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6906 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6907 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6910 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6911 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6912 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6916 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
6917 let Predicates = [HasAVX] in
6918 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
6919 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
6922 // Packed Compare Explicit Length Strings, Return Index
6923 multiclass pseudo_pcmpestri<string asm> {
6924 def REG : PseudoI<(outs GR32:$dst),
6925 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6926 [(set GR32:$dst, EFLAGS,
6927 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6928 def MEM : PseudoI<(outs GR32:$dst),
6929 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6930 [(set GR32:$dst, EFLAGS,
6931 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
6935 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6936 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
6937 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
6940 multiclass SS42AI_pcmpestri<string asm> {
6941 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6942 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6943 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6946 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6947 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6948 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6952 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6953 let Predicates = [HasAVX] in
6954 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
6955 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
6958 //===----------------------------------------------------------------------===//
6959 // SSE4.2 - CRC Instructions
6960 //===----------------------------------------------------------------------===//
6962 // No CRC instructions have AVX equivalents
6964 // crc intrinsic instruction
6965 // This set of instructions are only rm, the only difference is the size
6967 let Constraints = "$src1 = $dst" in {
6968 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6969 (ins GR32:$src1, i8mem:$src2),
6970 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6972 (int_x86_sse42_crc32_32_8 GR32:$src1,
6973 (load addr:$src2)))]>;
6974 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6975 (ins GR32:$src1, GR8:$src2),
6976 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6978 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6979 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6980 (ins GR32:$src1, i16mem:$src2),
6981 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6983 (int_x86_sse42_crc32_32_16 GR32:$src1,
6984 (load addr:$src2)))]>,
6986 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6987 (ins GR32:$src1, GR16:$src2),
6988 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6992 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6993 (ins GR32:$src1, i32mem:$src2),
6994 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6996 (int_x86_sse42_crc32_32_32 GR32:$src1,
6997 (load addr:$src2)))]>;
6998 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6999 (ins GR32:$src1, GR32:$src2),
7000 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7002 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7003 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7004 (ins GR64:$src1, i8mem:$src2),
7005 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7007 (int_x86_sse42_crc32_64_8 GR64:$src1,
7008 (load addr:$src2)))]>,
7010 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7011 (ins GR64:$src1, GR8:$src2),
7012 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7014 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7016 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7017 (ins GR64:$src1, i64mem:$src2),
7018 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7020 (int_x86_sse42_crc32_64_64 GR64:$src1,
7021 (load addr:$src2)))]>,
7023 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7024 (ins GR64:$src1, GR64:$src2),
7025 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7027 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7031 //===----------------------------------------------------------------------===//
7032 // AES-NI Instructions
7033 //===----------------------------------------------------------------------===//
7035 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7036 Intrinsic IntId128, bit Is2Addr = 1> {
7037 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7038 (ins VR128:$src1, VR128:$src2),
7040 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7041 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7042 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7044 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7045 (ins VR128:$src1, i128mem:$src2),
7047 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7048 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7050 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7053 // Perform One Round of an AES Encryption/Decryption Flow
7054 let Predicates = [HasAVX, HasAES] in {
7055 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7056 int_x86_aesni_aesenc, 0>, VEX_4V;
7057 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7058 int_x86_aesni_aesenclast, 0>, VEX_4V;
7059 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7060 int_x86_aesni_aesdec, 0>, VEX_4V;
7061 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7062 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7065 let Constraints = "$src1 = $dst" in {
7066 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7067 int_x86_aesni_aesenc>;
7068 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7069 int_x86_aesni_aesenclast>;
7070 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7071 int_x86_aesni_aesdec>;
7072 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7073 int_x86_aesni_aesdeclast>;
7076 // Perform the AES InvMixColumn Transformation
7077 let Predicates = [HasAVX, HasAES] in {
7078 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7080 "vaesimc\t{$src1, $dst|$dst, $src1}",
7082 (int_x86_aesni_aesimc VR128:$src1))]>,
7084 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7085 (ins i128mem:$src1),
7086 "vaesimc\t{$src1, $dst|$dst, $src1}",
7087 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7090 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7092 "aesimc\t{$src1, $dst|$dst, $src1}",
7094 (int_x86_aesni_aesimc VR128:$src1))]>,
7096 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7097 (ins i128mem:$src1),
7098 "aesimc\t{$src1, $dst|$dst, $src1}",
7099 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7102 // AES Round Key Generation Assist
7103 let Predicates = [HasAVX, HasAES] in {
7104 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7105 (ins VR128:$src1, i8imm:$src2),
7106 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7108 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7110 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7111 (ins i128mem:$src1, i8imm:$src2),
7112 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7114 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7117 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7118 (ins VR128:$src1, i8imm:$src2),
7119 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7121 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7123 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7124 (ins i128mem:$src1, i8imm:$src2),
7125 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7127 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7130 //===----------------------------------------------------------------------===//
7131 // PCLMUL Instructions
7132 //===----------------------------------------------------------------------===//
7134 // AVX carry-less Multiplication instructions
7135 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7136 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7137 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7139 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7141 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7142 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7143 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7144 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7145 (memopv2i64 addr:$src2), imm:$src3))]>;
7147 // Carry-less Multiplication instructions
7148 let Constraints = "$src1 = $dst" in {
7149 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7150 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7151 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7153 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7155 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7156 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7157 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7158 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7159 (memopv2i64 addr:$src2), imm:$src3))]>;
7160 } // Constraints = "$src1 = $dst"
7163 multiclass pclmul_alias<string asm, int immop> {
7164 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7165 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7167 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7168 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7170 def : InstAlias<!strconcat("vpclmul", asm,
7171 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7172 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7174 def : InstAlias<!strconcat("vpclmul", asm,
7175 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7176 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7178 defm : pclmul_alias<"hqhq", 0x11>;
7179 defm : pclmul_alias<"hqlq", 0x01>;
7180 defm : pclmul_alias<"lqhq", 0x10>;
7181 defm : pclmul_alias<"lqlq", 0x00>;
7183 //===----------------------------------------------------------------------===//
7184 // SSE4A Instructions
7185 //===----------------------------------------------------------------------===//
7187 let Predicates = [HasSSE4A] in {
7189 let Constraints = "$src = $dst" in {
7190 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7191 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7192 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7193 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7194 imm:$idx))]>, TB, OpSize;
7195 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7196 (ins VR128:$src, VR128:$mask),
7197 "extrq\t{$mask, $src|$src, $mask}",
7198 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7199 VR128:$mask))]>, TB, OpSize;
7201 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7202 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7203 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7204 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7205 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7206 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7207 (ins VR128:$src, VR128:$mask),
7208 "insertq\t{$mask, $src|$src, $mask}",
7209 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7210 VR128:$mask))]>, XD;
7213 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7214 "movntss\t{$src, $dst|$dst, $src}",
7215 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7217 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7218 "movntsd\t{$src, $dst|$dst, $src}",
7219 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7222 //===----------------------------------------------------------------------===//
7224 //===----------------------------------------------------------------------===//
7226 //===----------------------------------------------------------------------===//
7227 // VBROADCAST - Load from memory and broadcast to all elements of the
7228 // destination operand
7230 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7231 X86MemOperand x86memop, Intrinsic Int> :
7232 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7233 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7234 [(set RC:$dst, (Int addr:$src))]>, VEX;
7236 // AVX2 adds register forms
7237 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7239 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7240 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7241 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7243 let ExeDomain = SSEPackedSingle in {
7244 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7245 int_x86_avx_vbroadcast_ss>;
7246 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7247 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7249 let ExeDomain = SSEPackedDouble in
7250 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7251 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7252 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7253 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7255 let ExeDomain = SSEPackedSingle in {
7256 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7257 int_x86_avx2_vbroadcast_ss_ps>;
7258 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7259 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7261 let ExeDomain = SSEPackedDouble in
7262 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7263 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7265 let Predicates = [HasAVX2] in
7266 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7267 int_x86_avx2_vbroadcasti128>, VEX_L;
7269 let Predicates = [HasAVX] in
7270 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7271 (VBROADCASTF128 addr:$src)>;
7274 //===----------------------------------------------------------------------===//
7275 // VINSERTF128 - Insert packed floating-point values
7277 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7278 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7279 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7280 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7283 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7284 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7285 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7289 let Predicates = [HasAVX] in {
7290 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7292 (VINSERTF128rr VR256:$src1, VR128:$src2,
7293 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7294 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7296 (VINSERTF128rr VR256:$src1, VR128:$src2,
7297 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7299 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7301 (VINSERTF128rm VR256:$src1, addr:$src2,
7302 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7303 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7305 (VINSERTF128rm VR256:$src1, addr:$src2,
7306 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7309 let Predicates = [HasAVX1Only] in {
7310 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7312 (VINSERTF128rr VR256:$src1, VR128:$src2,
7313 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7314 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7316 (VINSERTF128rr VR256:$src1, VR128:$src2,
7317 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7318 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7320 (VINSERTF128rr VR256:$src1, VR128:$src2,
7321 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7322 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7324 (VINSERTF128rr VR256:$src1, VR128:$src2,
7325 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7327 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7329 (VINSERTF128rm VR256:$src1, addr:$src2,
7330 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7331 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7332 (bc_v4i32 (memopv2i64 addr:$src2)),
7334 (VINSERTF128rm VR256:$src1, addr:$src2,
7335 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7336 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7337 (bc_v16i8 (memopv2i64 addr:$src2)),
7339 (VINSERTF128rm VR256:$src1, addr:$src2,
7340 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7341 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7342 (bc_v8i16 (memopv2i64 addr:$src2)),
7344 (VINSERTF128rm VR256:$src1, addr:$src2,
7345 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7348 //===----------------------------------------------------------------------===//
7349 // VEXTRACTF128 - Extract packed floating-point values
7351 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7352 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7353 (ins VR256:$src1, i8imm:$src2),
7354 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7357 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7358 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7359 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7364 let Predicates = [HasAVX] in {
7365 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7366 (v4f32 (VEXTRACTF128rr
7367 (v8f32 VR256:$src1),
7368 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7369 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7370 (v2f64 (VEXTRACTF128rr
7371 (v4f64 VR256:$src1),
7372 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7374 def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
7375 (iPTR imm))), addr:$dst),
7376 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7377 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7378 def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
7379 (iPTR imm))), addr:$dst),
7380 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7381 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7384 let Predicates = [HasAVX1Only] in {
7385 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7386 (v2i64 (VEXTRACTF128rr
7387 (v4i64 VR256:$src1),
7388 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7389 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7390 (v4i32 (VEXTRACTF128rr
7391 (v8i32 VR256:$src1),
7392 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7393 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7394 (v8i16 (VEXTRACTF128rr
7395 (v16i16 VR256:$src1),
7396 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7397 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7398 (v16i8 (VEXTRACTF128rr
7399 (v32i8 VR256:$src1),
7400 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7402 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7403 (iPTR imm))), addr:$dst),
7404 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7405 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7406 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7407 (iPTR imm))), addr:$dst),
7408 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7409 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7410 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7411 (iPTR imm))), addr:$dst),
7412 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7413 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7414 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7415 (iPTR imm))), addr:$dst),
7416 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7417 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7420 //===----------------------------------------------------------------------===//
7421 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7423 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7424 Intrinsic IntLd, Intrinsic IntLd256,
7425 Intrinsic IntSt, Intrinsic IntSt256> {
7426 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7427 (ins VR128:$src1, f128mem:$src2),
7428 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7429 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7431 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7432 (ins VR256:$src1, f256mem:$src2),
7433 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7434 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7436 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7437 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7438 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7439 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7440 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7441 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7443 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7446 let ExeDomain = SSEPackedSingle in
7447 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7448 int_x86_avx_maskload_ps,
7449 int_x86_avx_maskload_ps_256,
7450 int_x86_avx_maskstore_ps,
7451 int_x86_avx_maskstore_ps_256>;
7452 let ExeDomain = SSEPackedDouble in
7453 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7454 int_x86_avx_maskload_pd,
7455 int_x86_avx_maskload_pd_256,
7456 int_x86_avx_maskstore_pd,
7457 int_x86_avx_maskstore_pd_256>;
7459 //===----------------------------------------------------------------------===//
7460 // VPERMIL - Permute Single and Double Floating-Point Values
7462 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7463 RegisterClass RC, X86MemOperand x86memop_f,
7464 X86MemOperand x86memop_i, PatFrag i_frag,
7465 Intrinsic IntVar, ValueType vt> {
7466 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7467 (ins RC:$src1, RC:$src2),
7468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7469 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7470 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7471 (ins RC:$src1, x86memop_i:$src2),
7472 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7473 [(set RC:$dst, (IntVar RC:$src1,
7474 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7476 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7477 (ins RC:$src1, i8imm:$src2),
7478 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7479 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7480 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7481 (ins x86memop_f:$src1, i8imm:$src2),
7482 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7484 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7487 let ExeDomain = SSEPackedSingle in {
7488 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7489 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7490 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7491 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7493 let ExeDomain = SSEPackedDouble in {
7494 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7495 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7496 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7497 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7500 let Predicates = [HasAVX] in {
7501 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7502 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7503 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7504 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7505 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7507 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7508 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7509 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7511 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7512 (VPERMILPDri VR128:$src1, imm:$imm)>;
7513 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7514 (VPERMILPDmi addr:$src1, imm:$imm)>;
7517 //===----------------------------------------------------------------------===//
7518 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7520 let ExeDomain = SSEPackedSingle in {
7521 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7522 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7523 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7524 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7525 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7526 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7527 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7528 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7529 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7530 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7533 let Predicates = [HasAVX] in {
7534 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7535 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7536 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7537 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7538 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7541 let Predicates = [HasAVX1Only] in {
7542 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7543 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7544 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7545 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7546 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7547 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7548 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7549 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7551 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7552 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7553 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7554 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7555 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7556 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7557 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7558 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7559 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7560 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7561 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7562 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7565 //===----------------------------------------------------------------------===//
7566 // VZERO - Zero YMM registers
7568 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7569 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7570 // Zero All YMM registers
7571 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7572 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7574 // Zero Upper bits of YMM registers
7575 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7576 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7579 //===----------------------------------------------------------------------===//
7580 // Half precision conversion instructions
7581 //===----------------------------------------------------------------------===//
7582 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7583 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7584 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7585 [(set RC:$dst, (Int VR128:$src))]>,
7587 let neverHasSideEffects = 1, mayLoad = 1 in
7588 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7589 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7592 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7593 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7594 (ins RC:$src1, i32i8imm:$src2),
7595 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7596 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7598 let neverHasSideEffects = 1, mayStore = 1 in
7599 def mr : Ii8<0x1D, MRMDestMem, (outs),
7600 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7601 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7605 let Predicates = [HasAVX, HasF16C] in {
7606 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7607 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7608 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7609 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7612 //===----------------------------------------------------------------------===//
7613 // AVX2 Instructions
7614 //===----------------------------------------------------------------------===//
7616 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7617 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7618 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7619 X86MemOperand x86memop> {
7620 let isCommutable = 1 in
7621 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7622 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7623 !strconcat(OpcodeStr,
7624 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7625 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7627 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7628 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7629 !strconcat(OpcodeStr,
7630 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7633 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7637 let isCommutable = 0 in {
7638 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7639 VR128, memopv2i64, i128mem>;
7640 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7641 VR256, memopv4i64, i256mem>, VEX_L;
7644 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7646 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7647 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7649 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7651 //===----------------------------------------------------------------------===//
7652 // VPBROADCAST - Load from memory and broadcast to all elements of the
7653 // destination operand
7655 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7656 X86MemOperand x86memop, PatFrag ld_frag,
7657 Intrinsic Int128, Intrinsic Int256> {
7658 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7659 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7660 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7661 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7662 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7664 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7665 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7666 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7667 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7668 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7669 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7671 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7675 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7676 int_x86_avx2_pbroadcastb_128,
7677 int_x86_avx2_pbroadcastb_256>;
7678 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7679 int_x86_avx2_pbroadcastw_128,
7680 int_x86_avx2_pbroadcastw_256>;
7681 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7682 int_x86_avx2_pbroadcastd_128,
7683 int_x86_avx2_pbroadcastd_256>;
7684 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7685 int_x86_avx2_pbroadcastq_128,
7686 int_x86_avx2_pbroadcastq_256>;
7688 let Predicates = [HasAVX2] in {
7689 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7690 (VPBROADCASTBrm addr:$src)>;
7691 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7692 (VPBROADCASTBYrm addr:$src)>;
7693 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7694 (VPBROADCASTWrm addr:$src)>;
7695 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7696 (VPBROADCASTWYrm addr:$src)>;
7697 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7698 (VPBROADCASTDrm addr:$src)>;
7699 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7700 (VPBROADCASTDYrm addr:$src)>;
7701 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7702 (VPBROADCASTQrm addr:$src)>;
7703 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7704 (VPBROADCASTQYrm addr:$src)>;
7706 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7707 (VPBROADCASTBrr VR128:$src)>;
7708 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7709 (VPBROADCASTBYrr VR128:$src)>;
7710 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7711 (VPBROADCASTWrr VR128:$src)>;
7712 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7713 (VPBROADCASTWYrr VR128:$src)>;
7714 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7715 (VPBROADCASTDrr VR128:$src)>;
7716 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7717 (VPBROADCASTDYrr VR128:$src)>;
7718 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7719 (VPBROADCASTQrr VR128:$src)>;
7720 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7721 (VPBROADCASTQYrr VR128:$src)>;
7722 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7723 (VBROADCASTSSrr VR128:$src)>;
7724 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7725 (VBROADCASTSSYrr VR128:$src)>;
7726 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7727 (VPBROADCASTQrr VR128:$src)>;
7728 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7729 (VBROADCASTSDYrr VR128:$src)>;
7731 // Provide fallback in case the load node that is used in the patterns above
7732 // is used by additional users, which prevents the pattern selection.
7733 let AddedComplexity = 20 in {
7734 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7735 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7736 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7737 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7738 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7739 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7741 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7742 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7743 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7744 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7745 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7746 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7750 // AVX1 broadcast patterns
7751 let Predicates = [HasAVX1Only] in {
7752 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7753 (VBROADCASTSSYrm addr:$src)>;
7754 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7755 (VBROADCASTSDYrm addr:$src)>;
7756 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7757 (VBROADCASTSSrm addr:$src)>;
7760 let Predicates = [HasAVX] in {
7761 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7762 (VBROADCASTSSYrm addr:$src)>;
7763 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7764 (VBROADCASTSDYrm addr:$src)>;
7765 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7766 (VBROADCASTSSrm addr:$src)>;
7768 // Provide fallback in case the load node that is used in the patterns above
7769 // is used by additional users, which prevents the pattern selection.
7770 let AddedComplexity = 20 in {
7771 // 128bit broadcasts:
7772 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7773 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7774 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7775 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7776 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7777 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7778 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7779 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7780 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7781 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7783 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7784 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7785 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7786 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7787 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7788 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7789 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7790 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7791 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7792 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7796 //===----------------------------------------------------------------------===//
7797 // VPERM - Permute instructions
7800 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7802 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7803 (ins VR256:$src1, VR256:$src2),
7804 !strconcat(OpcodeStr,
7805 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7807 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
7809 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7810 (ins VR256:$src1, i256mem:$src2),
7811 !strconcat(OpcodeStr,
7812 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7814 (OpVT (X86VPermv VR256:$src1,
7815 (bitconvert (mem_frag addr:$src2)))))]>,
7819 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7820 let ExeDomain = SSEPackedSingle in
7821 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7823 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7825 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7826 (ins VR256:$src1, i8imm:$src2),
7827 !strconcat(OpcodeStr,
7828 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7830 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
7832 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7833 (ins i256mem:$src1, i8imm:$src2),
7834 !strconcat(OpcodeStr,
7835 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7837 (OpVT (X86VPermi (mem_frag addr:$src1),
7838 (i8 imm:$src2))))]>, VEX, VEX_L;
7841 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7842 let ExeDomain = SSEPackedDouble in
7843 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7845 //===----------------------------------------------------------------------===//
7846 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7848 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7849 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7850 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7851 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7852 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7853 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7854 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7855 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7856 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7857 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7859 let Predicates = [HasAVX2] in {
7860 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7861 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7862 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7863 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7864 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7865 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7867 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7869 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7870 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7871 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7872 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7873 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7875 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7879 //===----------------------------------------------------------------------===//
7880 // VINSERTI128 - Insert packed integer values
7882 let neverHasSideEffects = 1 in {
7883 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7884 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7885 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7888 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7889 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7890 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7894 let Predicates = [HasAVX2] in {
7895 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7897 (VINSERTI128rr VR256:$src1, VR128:$src2,
7898 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7899 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7901 (VINSERTI128rr VR256:$src1, VR128:$src2,
7902 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7903 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7905 (VINSERTI128rr VR256:$src1, VR128:$src2,
7906 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7907 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7909 (VINSERTI128rr VR256:$src1, VR128:$src2,
7910 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7912 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7914 (VINSERTI128rm VR256:$src1, addr:$src2,
7915 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7916 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7917 (bc_v4i32 (memopv2i64 addr:$src2)),
7919 (VINSERTI128rm VR256:$src1, addr:$src2,
7920 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7921 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7922 (bc_v16i8 (memopv2i64 addr:$src2)),
7924 (VINSERTI128rm VR256:$src1, addr:$src2,
7925 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7926 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7927 (bc_v8i16 (memopv2i64 addr:$src2)),
7929 (VINSERTI128rm VR256:$src1, addr:$src2,
7930 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7933 //===----------------------------------------------------------------------===//
7934 // VEXTRACTI128 - Extract packed integer values
7936 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7937 (ins VR256:$src1, i8imm:$src2),
7938 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7940 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7942 let neverHasSideEffects = 1, mayStore = 1 in
7943 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7944 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7945 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7948 let Predicates = [HasAVX2] in {
7949 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7950 (v2i64 (VEXTRACTI128rr
7951 (v4i64 VR256:$src1),
7952 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7953 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7954 (v4i32 (VEXTRACTI128rr
7955 (v8i32 VR256:$src1),
7956 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7957 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7958 (v8i16 (VEXTRACTI128rr
7959 (v16i16 VR256:$src1),
7960 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7961 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7962 (v16i8 (VEXTRACTI128rr
7963 (v32i8 VR256:$src1),
7964 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7966 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7967 (iPTR imm))), addr:$dst),
7968 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7969 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7970 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7971 (iPTR imm))), addr:$dst),
7972 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7973 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7974 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7975 (iPTR imm))), addr:$dst),
7976 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7977 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7978 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7979 (iPTR imm))), addr:$dst),
7980 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7981 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7984 //===----------------------------------------------------------------------===//
7985 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7987 multiclass avx2_pmovmask<string OpcodeStr,
7988 Intrinsic IntLd128, Intrinsic IntLd256,
7989 Intrinsic IntSt128, Intrinsic IntSt256> {
7990 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7991 (ins VR128:$src1, i128mem:$src2),
7992 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7993 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7994 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7995 (ins VR256:$src1, i256mem:$src2),
7996 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7997 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7999 def mr : AVX28I<0x8e, MRMDestMem, (outs),
8000 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
8001 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8002 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
8003 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
8004 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
8005 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8006 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8009 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8010 int_x86_avx2_maskload_d,
8011 int_x86_avx2_maskload_d_256,
8012 int_x86_avx2_maskstore_d,
8013 int_x86_avx2_maskstore_d_256>;
8014 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8015 int_x86_avx2_maskload_q,
8016 int_x86_avx2_maskload_q_256,
8017 int_x86_avx2_maskstore_q,
8018 int_x86_avx2_maskstore_q_256>, VEX_W;
8021 //===----------------------------------------------------------------------===//
8022 // Variable Bit Shifts
8024 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8025 ValueType vt128, ValueType vt256> {
8026 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8027 (ins VR128:$src1, VR128:$src2),
8028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8030 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8032 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8033 (ins VR128:$src1, i128mem:$src2),
8034 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8036 (vt128 (OpNode VR128:$src1,
8037 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8039 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8040 (ins VR256:$src1, VR256:$src2),
8041 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8043 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8045 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8046 (ins VR256:$src1, i256mem:$src2),
8047 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8049 (vt256 (OpNode VR256:$src1,
8050 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8054 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8055 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8056 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8057 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8058 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8060 //===----------------------------------------------------------------------===//
8061 // VGATHER - GATHER Operations
8062 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8063 X86MemOperand memop128, X86MemOperand memop256> {
8064 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8065 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8066 !strconcat(OpcodeStr,
8067 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8069 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8070 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8071 !strconcat(OpcodeStr,
8072 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8073 []>, VEX_4VOp3, VEX_L;
8076 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8077 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8078 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8079 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8080 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8081 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8082 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8083 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8084 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;