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 = $dst" in {
2907 defm MAXC: basic_sse12_fp_binop_s<0x5F, "max", X86fmaxc, SSE_ALU_ITINS_S>;
2908 defm MINC: basic_sse12_fp_binop_s<0x5D, "min", X86fminc, SSE_ALU_ITINS_S>;
2913 /// In addition, we also have a special variant of the scalar form here to
2914 /// represent the associated intrinsic operation. This form is unlike the
2915 /// plain scalar form, in that it takes an entire vector (instead of a
2916 /// scalar) and leaves the top elements undefined.
2918 /// And, we have a special variant form for a full-vector intrinsic form.
2920 def SSE_SQRTP : OpndItins<
2921 IIC_SSE_SQRTP_RR, IIC_SSE_SQRTP_RM
2924 def SSE_SQRTS : OpndItins<
2925 IIC_SSE_SQRTS_RR, IIC_SSE_SQRTS_RM
2928 def SSE_RCPP : OpndItins<
2929 IIC_SSE_RCPP_RR, IIC_SSE_RCPP_RM
2932 def SSE_RCPS : OpndItins<
2933 IIC_SSE_RCPS_RR, IIC_SSE_RCPS_RM
2936 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2937 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2938 SDNode OpNode, Intrinsic F32Int, OpndItins itins> {
2939 let Predicates = [HasAVX], hasSideEffects = 0 in {
2940 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
2941 (ins FR32:$src1, FR32:$src2),
2942 !strconcat(!strconcat("v", OpcodeStr),
2943 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2944 []>, VEX_4V, VEX_LIG;
2945 let mayLoad = 1 in {
2946 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
2947 (ins FR32:$src1,f32mem:$src2),
2948 !strconcat(OpcodeStr,
2949 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2950 []>, VEX_4V, VEX_LIG;
2951 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2952 (ins VR128:$src1, ssmem:$src2),
2953 !strconcat(!strconcat("v", OpcodeStr),
2954 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2955 []>, VEX_4V, VEX_LIG;
2959 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2960 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2961 [(set FR32:$dst, (OpNode FR32:$src))]>;
2962 // For scalar unary operations, fold a load into the operation
2963 // only in OptForSize mode. It eliminates an instruction, but it also
2964 // eliminates a whole-register clobber (the load), so it introduces a
2965 // partial register update condition.
2966 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2967 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2968 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
2969 Requires<[UseSSE1, OptForSize]>;
2970 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2971 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2972 [(set VR128:$dst, (F32Int VR128:$src))], itins.rr>;
2973 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2974 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2975 [(set VR128:$dst, (F32Int sse_load_f32:$src))], itins.rm>;
2978 /// sse1_fp_unop_s_rw - SSE1 unops where vector form has a read-write operand.
2979 multiclass sse1_fp_unop_rw<bits<8> opc, string OpcodeStr, SDNode OpNode,
2981 let Predicates = [HasAVX], hasSideEffects = 0 in {
2982 def V#NAME#SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst),
2983 (ins FR32:$src1, FR32:$src2),
2984 !strconcat(!strconcat("v", OpcodeStr),
2985 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2986 []>, VEX_4V, VEX_LIG;
2987 let mayLoad = 1 in {
2988 def V#NAME#SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst),
2989 (ins FR32:$src1,f32mem:$src2),
2990 !strconcat(!strconcat("v", OpcodeStr),
2991 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2992 []>, VEX_4V, VEX_LIG;
2993 def V#NAME#SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2994 (ins VR128:$src1, ssmem:$src2),
2995 !strconcat(!strconcat("v", OpcodeStr),
2996 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2997 []>, VEX_4V, VEX_LIG;
3001 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
3002 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3003 [(set FR32:$dst, (OpNode FR32:$src))]>;
3004 // For scalar unary operations, fold a load into the operation
3005 // only in OptForSize mode. It eliminates an instruction, but it also
3006 // eliminates a whole-register clobber (the load), so it introduces a
3007 // partial register update condition.
3008 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
3009 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
3010 [(set FR32:$dst, (OpNode (load addr:$src)))], itins.rm>, XS,
3011 Requires<[UseSSE1, OptForSize]>;
3012 let Constraints = "$src1 = $dst" in {
3013 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
3014 (ins VR128:$src1, VR128:$src2),
3015 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3017 let mayLoad = 1, hasSideEffects = 0 in
3018 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
3019 (ins VR128:$src1, ssmem:$src2),
3020 !strconcat(OpcodeStr, "ss\t{$src2, $dst|$dst, $src2}"),
3025 /// sse1_fp_unop_p - SSE1 unops in packed form.
3026 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
3028 let Predicates = [HasAVX] in {
3029 def V#NAME#PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3030 !strconcat(!strconcat("v", OpcodeStr),
3031 "ps\t{$src, $dst|$dst, $src}"),
3032 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))],
3034 def V#NAME#PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3035 !strconcat(!strconcat("v", OpcodeStr),
3036 "ps\t{$src, $dst|$dst, $src}"),
3037 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))],
3039 def V#NAME#PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3040 !strconcat(!strconcat("v", OpcodeStr),
3041 "ps\t{$src, $dst|$dst, $src}"),
3042 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))],
3043 itins.rr>, VEX, VEX_L;
3044 def V#NAME#PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3045 !strconcat(!strconcat("v", OpcodeStr),
3046 "ps\t{$src, $dst|$dst, $src}"),
3047 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))],
3048 itins.rm>, VEX, VEX_L;
3051 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3052 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3053 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))], itins.rr>;
3054 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3055 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3056 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))], itins.rm>;
3059 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3060 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3061 Intrinsic V4F32Int, Intrinsic V8F32Int,
3063 let Predicates = [HasAVX] in {
3064 def V#NAME#PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3065 !strconcat(!strconcat("v", OpcodeStr),
3066 "ps\t{$src, $dst|$dst, $src}"),
3067 [(set VR128:$dst, (V4F32Int VR128:$src))],
3069 def V#NAME#PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3070 !strconcat(!strconcat("v", OpcodeStr),
3071 "ps\t{$src, $dst|$dst, $src}"),
3072 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3074 def V#NAME#PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3075 !strconcat(!strconcat("v", OpcodeStr),
3076 "ps\t{$src, $dst|$dst, $src}"),
3077 [(set VR256:$dst, (V8F32Int VR256:$src))],
3078 itins.rr>, VEX, VEX_L;
3079 def V#NAME#PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst),
3081 !strconcat(!strconcat("v", OpcodeStr),
3082 "ps\t{$src, $dst|$dst, $src}"),
3083 [(set VR256:$dst, (V8F32Int (memopv8f32 addr:$src)))],
3084 itins.rm>, VEX, VEX_L;
3087 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3088 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3089 [(set VR128:$dst, (V4F32Int VR128:$src))],
3091 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3092 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3093 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))],
3097 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3098 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3099 SDNode OpNode, Intrinsic F64Int, OpndItins itins> {
3100 let Predicates = [HasAVX], hasSideEffects = 0 in {
3101 def V#NAME#SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst),
3102 (ins FR64:$src1, FR64:$src2),
3103 !strconcat(!strconcat("v", OpcodeStr),
3104 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3105 []>, VEX_4V, VEX_LIG;
3106 let mayLoad = 1 in {
3107 def V#NAME#SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
3108 (ins FR64:$src1,f64mem:$src2),
3109 !strconcat(OpcodeStr,
3110 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3111 []>, VEX_4V, VEX_LIG;
3112 def V#NAME#SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3113 (ins VR128:$src1, sdmem:$src2),
3114 !strconcat(!strconcat("v", OpcodeStr),
3115 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3116 []>, VEX_4V, VEX_LIG;
3120 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3121 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3122 [(set FR64:$dst, (OpNode FR64:$src))], itins.rr>;
3123 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3124 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3125 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3126 [(set FR64:$dst, (OpNode (load addr:$src)))], itins.rm>, XD,
3127 Requires<[UseSSE2, OptForSize]>;
3128 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3129 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3130 [(set VR128:$dst, (F64Int VR128:$src))], itins.rr>;
3131 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3132 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3133 [(set VR128:$dst, (F64Int sse_load_f64:$src))], itins.rm>;
3136 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3137 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3138 SDNode OpNode, OpndItins itins> {
3139 let Predicates = [HasAVX] in {
3140 def V#NAME#PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3141 !strconcat(!strconcat("v", OpcodeStr),
3142 "pd\t{$src, $dst|$dst, $src}"),
3143 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))],
3145 def V#NAME#PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3146 !strconcat(!strconcat("v", OpcodeStr),
3147 "pd\t{$src, $dst|$dst, $src}"),
3148 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))],
3150 def V#NAME#PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3151 !strconcat(!strconcat("v", OpcodeStr),
3152 "pd\t{$src, $dst|$dst, $src}"),
3153 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))],
3154 itins.rr>, VEX, VEX_L;
3155 def V#NAME#PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3156 !strconcat(!strconcat("v", OpcodeStr),
3157 "pd\t{$src, $dst|$dst, $src}"),
3158 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))],
3159 itins.rm>, VEX, VEX_L;
3162 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3163 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3164 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))], itins.rr>;
3165 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3166 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3167 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))], itins.rm>;
3171 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss,
3173 sse1_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>,
3174 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd,
3176 sse2_fp_unop_p<0x51, "sqrt", fsqrt, SSE_SQRTP>;
3178 // Reciprocal approximations. Note that these typically require refinement
3179 // in order to obtain suitable precision.
3180 defm RSQRT : sse1_fp_unop_rw<0x52, "rsqrt", X86frsqrt, SSE_SQRTS>,
3181 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, SSE_SQRTP>,
3182 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps,
3183 int_x86_avx_rsqrt_ps_256, SSE_SQRTP>;
3184 defm RCP : sse1_fp_unop_rw<0x53, "rcp", X86frcp, SSE_RCPS>,
3185 sse1_fp_unop_p<0x53, "rcp", X86frcp, SSE_RCPP>,
3186 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps,
3187 int_x86_avx_rcp_ps_256, SSE_RCPP>;
3189 def : Pat<(f32 (fsqrt FR32:$src)),
3190 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3191 def : Pat<(f32 (fsqrt (load addr:$src))),
3192 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3193 Requires<[HasAVX, OptForSize]>;
3194 def : Pat<(f64 (fsqrt FR64:$src)),
3195 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3196 def : Pat<(f64 (fsqrt (load addr:$src))),
3197 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3198 Requires<[HasAVX, OptForSize]>;
3200 def : Pat<(f32 (X86frsqrt FR32:$src)),
3201 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3202 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3203 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3204 Requires<[HasAVX, OptForSize]>;
3206 def : Pat<(f32 (X86frcp FR32:$src)),
3207 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3208 def : Pat<(f32 (X86frcp (load addr:$src))),
3209 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3210 Requires<[HasAVX, OptForSize]>;
3212 let Predicates = [HasAVX] in {
3213 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3214 (COPY_TO_REGCLASS (VSQRTSSr (f32 (IMPLICIT_DEF)),
3215 (COPY_TO_REGCLASS VR128:$src, FR32)),
3217 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3218 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3220 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3221 (COPY_TO_REGCLASS (VSQRTSDr (f64 (IMPLICIT_DEF)),
3222 (COPY_TO_REGCLASS VR128:$src, FR64)),
3224 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3225 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3227 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3228 (COPY_TO_REGCLASS (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3229 (COPY_TO_REGCLASS VR128:$src, FR32)),
3231 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3232 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3234 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3235 (COPY_TO_REGCLASS (VRCPSSr (f32 (IMPLICIT_DEF)),
3236 (COPY_TO_REGCLASS VR128:$src, FR32)),
3238 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3239 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3242 // Reciprocal approximations. Note that these typically require refinement
3243 // in order to obtain suitable precision.
3244 let Predicates = [UseSSE1] in {
3245 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3246 (RSQRTSSr_Int VR128:$src, VR128:$src)>;
3247 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3248 (RCPSSr_Int VR128:$src, VR128:$src)>;
3251 // There is no f64 version of the reciprocal approximation instructions.
3253 //===----------------------------------------------------------------------===//
3254 // SSE 1 & 2 - Non-temporal stores
3255 //===----------------------------------------------------------------------===//
3257 let AddedComplexity = 400 in { // Prefer non-temporal versions
3258 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3259 (ins f128mem:$dst, VR128:$src),
3260 "movntps\t{$src, $dst|$dst, $src}",
3261 [(alignednontemporalstore (v4f32 VR128:$src),
3263 IIC_SSE_MOVNT>, VEX;
3264 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3265 (ins f128mem:$dst, VR128:$src),
3266 "movntpd\t{$src, $dst|$dst, $src}",
3267 [(alignednontemporalstore (v2f64 VR128:$src),
3269 IIC_SSE_MOVNT>, VEX;
3271 let ExeDomain = SSEPackedInt in
3272 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3273 (ins f128mem:$dst, VR128:$src),
3274 "movntdq\t{$src, $dst|$dst, $src}",
3275 [(alignednontemporalstore (v2i64 VR128:$src),
3277 IIC_SSE_MOVNT>, VEX;
3279 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3280 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3282 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3283 (ins f256mem:$dst, VR256:$src),
3284 "movntps\t{$src, $dst|$dst, $src}",
3285 [(alignednontemporalstore (v8f32 VR256:$src),
3287 IIC_SSE_MOVNT>, VEX, VEX_L;
3288 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3289 (ins f256mem:$dst, VR256:$src),
3290 "movntpd\t{$src, $dst|$dst, $src}",
3291 [(alignednontemporalstore (v4f64 VR256:$src),
3293 IIC_SSE_MOVNT>, VEX, VEX_L;
3294 let ExeDomain = SSEPackedInt in
3295 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3296 (ins f256mem:$dst, VR256:$src),
3297 "movntdq\t{$src, $dst|$dst, $src}",
3298 [(alignednontemporalstore (v4i64 VR256:$src),
3300 IIC_SSE_MOVNT>, VEX, VEX_L;
3303 let AddedComplexity = 400 in { // Prefer non-temporal versions
3304 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3305 "movntps\t{$src, $dst|$dst, $src}",
3306 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)],
3308 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3309 "movntpd\t{$src, $dst|$dst, $src}",
3310 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)],
3313 let ExeDomain = SSEPackedInt in
3314 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3315 "movntdq\t{$src, $dst|$dst, $src}",
3316 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)],
3319 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3320 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[UseSSE2]>;
3322 // There is no AVX form for instructions below this point
3323 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3324 "movnti{l}\t{$src, $dst|$dst, $src}",
3325 [(nontemporalstore (i32 GR32:$src), addr:$dst)],
3327 TB, Requires<[HasSSE2]>;
3328 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3329 "movnti{q}\t{$src, $dst|$dst, $src}",
3330 [(nontemporalstore (i64 GR64:$src), addr:$dst)],
3332 TB, Requires<[HasSSE2]>;
3335 //===----------------------------------------------------------------------===//
3336 // SSE 1 & 2 - Prefetch and memory fence
3337 //===----------------------------------------------------------------------===//
3339 // Prefetch intrinsic.
3340 let Predicates = [HasSSE1] in {
3341 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3342 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))],
3343 IIC_SSE_PREFETCH>, TB;
3344 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3345 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))],
3346 IIC_SSE_PREFETCH>, TB;
3347 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3348 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))],
3349 IIC_SSE_PREFETCH>, TB;
3350 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3351 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))],
3352 IIC_SSE_PREFETCH>, TB;
3356 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3357 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)],
3358 IIC_SSE_PREFETCH>, TB, Requires<[HasSSE2]>;
3360 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3361 // was introduced with SSE2, it's backward compatible.
3362 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", [], IIC_SSE_PAUSE>, REP;
3364 // Load, store, and memory fence
3365 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3366 "sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
3367 TB, Requires<[HasSSE1]>;
3368 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3369 "lfence", [(int_x86_sse2_lfence)], IIC_SSE_LFENCE>,
3370 TB, Requires<[HasSSE2]>;
3371 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3372 "mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
3373 TB, Requires<[HasSSE2]>;
3375 def : Pat<(X86SFence), (SFENCE)>;
3376 def : Pat<(X86LFence), (LFENCE)>;
3377 def : Pat<(X86MFence), (MFENCE)>;
3379 //===----------------------------------------------------------------------===//
3380 // SSE 1 & 2 - Load/Store XCSR register
3381 //===----------------------------------------------------------------------===//
3383 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3384 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3385 IIC_SSE_LDMXCSR>, VEX;
3386 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3387 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3388 IIC_SSE_STMXCSR>, VEX;
3390 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3391 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)],
3393 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3394 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)],
3397 //===---------------------------------------------------------------------===//
3398 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3399 //===---------------------------------------------------------------------===//
3401 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3403 let neverHasSideEffects = 1 in {
3404 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3405 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3407 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3408 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>,
3410 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3411 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3413 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3414 "movdqu\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVU_P_RR>,
3419 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3420 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3421 "movdqa\t{$src, $dst|$dst, $src}", [],
3424 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3425 "movdqa\t{$src, $dst|$dst, $src}", [],
3426 IIC_SSE_MOVA_P_RR>, VEX, VEX_L;
3427 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3428 "movdqu\t{$src, $dst|$dst, $src}", [],
3431 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3432 "movdqu\t{$src, $dst|$dst, $src}", [],
3433 IIC_SSE_MOVU_P_RR>, VEX, VEX_L;
3436 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3437 neverHasSideEffects = 1 in {
3438 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3439 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3441 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3442 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RM>,
3444 let Predicates = [HasAVX] in {
3445 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3446 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3448 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3449 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_RM>,
3454 let mayStore = 1, neverHasSideEffects = 1 in {
3455 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3456 (ins i128mem:$dst, VR128:$src),
3457 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3459 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3460 (ins i256mem:$dst, VR256:$src),
3461 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_MR>,
3463 let Predicates = [HasAVX] in {
3464 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3465 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3467 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3468 "vmovdqu\t{$src, $dst|$dst, $src}",[], IIC_SSE_MOVU_P_MR>,
3473 let neverHasSideEffects = 1 in
3474 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3475 "movdqa\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVA_P_RR>;
3477 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3478 "movdqu\t{$src, $dst|$dst, $src}",
3479 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3482 let isCodeGenOnly = 1, hasSideEffects = 0 in {
3483 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3484 "movdqa\t{$src, $dst|$dst, $src}", [],
3487 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3488 "movdqu\t{$src, $dst|$dst, $src}",
3489 [], IIC_SSE_MOVU_P_RR>, XS, Requires<[UseSSE2]>;
3492 let canFoldAsLoad = 1, mayLoad = 1, isReMaterializable = 1,
3493 neverHasSideEffects = 1 in {
3494 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3495 "movdqa\t{$src, $dst|$dst, $src}",
3496 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/],
3498 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3499 "movdqu\t{$src, $dst|$dst, $src}",
3500 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/],
3502 XS, Requires<[UseSSE2]>;
3505 let mayStore = 1 in {
3506 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3507 "movdqa\t{$src, $dst|$dst, $src}",
3508 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/],
3510 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3511 "movdqu\t{$src, $dst|$dst, $src}",
3512 [/*(store (v2i64 VR128:$src), addr:$dst)*/],
3514 XS, Requires<[UseSSE2]>;
3517 } // ExeDomain = SSEPackedInt
3519 let Predicates = [HasAVX] in {
3520 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3521 (VMOVDQUmr addr:$dst, VR128:$src)>;
3522 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3523 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3525 let Predicates = [UseSSE2] in
3526 def : Pat<(int_x86_sse2_storeu_dq addr:$dst, VR128:$src),
3527 (MOVDQUmr addr:$dst, VR128:$src)>;
3529 //===---------------------------------------------------------------------===//
3530 // SSE2 - Packed Integer Arithmetic Instructions
3531 //===---------------------------------------------------------------------===//
3533 def SSE_PMADD : OpndItins<
3534 IIC_SSE_PMADD, IIC_SSE_PMADD
3537 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3539 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3540 RegisterClass RC, PatFrag memop_frag,
3541 X86MemOperand x86memop,
3543 bit IsCommutable = 0,
3545 let isCommutable = IsCommutable in
3546 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3547 (ins RC:$src1, RC:$src2),
3549 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3550 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3551 [(set RC:$dst, (IntId RC:$src1, RC:$src2))], itins.rr>;
3552 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3553 (ins RC:$src1, x86memop:$src2),
3555 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3556 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3557 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))],
3561 multiclass PDI_binop_all_int<bits<8> opc, string OpcodeStr, Intrinsic IntId128,
3562 Intrinsic IntId256, OpndItins itins,
3563 bit IsCommutable = 0> {
3564 let Predicates = [HasAVX] in
3565 defm V#NAME# : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId128,
3566 VR128, memopv2i64, i128mem, itins,
3567 IsCommutable, 0>, VEX_4V;
3569 let Constraints = "$src1 = $dst" in
3570 defm #NAME# : PDI_binop_rm_int<opc, OpcodeStr, IntId128, VR128, memopv2i64,
3571 i128mem, itins, IsCommutable, 1>;
3573 let Predicates = [HasAVX2] in
3574 defm V#NAME#Y : PDI_binop_rm_int<opc, !strconcat("v", OpcodeStr), IntId256,
3575 VR256, memopv4i64, i256mem, itins,
3576 IsCommutable, 0>, VEX_4V, VEX_L;
3579 multiclass PDI_binop_rmi<bits<8> opc, bits<8> opc2, Format ImmForm,
3580 string OpcodeStr, SDNode OpNode,
3581 SDNode OpNode2, RegisterClass RC,
3582 ValueType DstVT, ValueType SrcVT, PatFrag bc_frag,
3583 ShiftOpndItins itins,
3585 // src2 is always 128-bit
3586 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3587 (ins RC:$src1, VR128:$src2),
3589 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3590 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3591 [(set RC:$dst, (DstVT (OpNode RC:$src1, (SrcVT VR128:$src2))))],
3593 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3594 (ins RC:$src1, i128mem:$src2),
3596 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3597 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3598 [(set RC:$dst, (DstVT (OpNode RC:$src1,
3599 (bc_frag (memopv2i64 addr:$src2)))))], itins.rm>;
3600 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3601 (ins RC:$src1, i32i8imm:$src2),
3603 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3604 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3605 [(set RC:$dst, (DstVT (OpNode2 RC:$src1, (i32 imm:$src2))))], itins.ri>;
3608 /// PDI_binop_rm2 - Simple SSE2 binary operator with different src and dst types
3609 multiclass PDI_binop_rm2<bits<8> opc, string OpcodeStr, SDNode OpNode,
3610 ValueType DstVT, ValueType SrcVT, RegisterClass RC,
3611 PatFrag memop_frag, X86MemOperand x86memop,
3613 bit IsCommutable = 0, bit Is2Addr = 1> {
3614 let isCommutable = IsCommutable in
3615 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3616 (ins RC:$src1, RC:$src2),
3618 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3619 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3620 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1), RC:$src2)))]>;
3621 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3622 (ins RC:$src1, x86memop:$src2),
3624 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3625 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3626 [(set RC:$dst, (DstVT (OpNode (SrcVT RC:$src1),
3627 (bitconvert (memop_frag addr:$src2)))))]>;
3629 } // ExeDomain = SSEPackedInt
3631 defm PADDB : PDI_binop_all<0xFC, "paddb", add, v16i8, v32i8,
3632 SSE_INTALU_ITINS_P, 1>;
3633 defm PADDW : PDI_binop_all<0xFD, "paddw", add, v8i16, v16i16,
3634 SSE_INTALU_ITINS_P, 1>;
3635 defm PADDD : PDI_binop_all<0xFE, "paddd", add, v4i32, v8i32,
3636 SSE_INTALU_ITINS_P, 1>;
3637 defm PADDQ : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
3638 SSE_INTALUQ_ITINS_P, 1>;
3639 defm PMULLW : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
3640 SSE_INTMUL_ITINS_P, 1>;
3641 defm PSUBB : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
3642 SSE_INTALU_ITINS_P, 0>;
3643 defm PSUBW : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
3644 SSE_INTALU_ITINS_P, 0>;
3645 defm PSUBD : PDI_binop_all<0xFA, "psubd", sub, v4i32, v8i32,
3646 SSE_INTALU_ITINS_P, 0>;
3647 defm PSUBQ : PDI_binop_all<0xFB, "psubq", sub, v2i64, v4i64,
3648 SSE_INTALUQ_ITINS_P, 0>;
3649 defm PSUBUSB : PDI_binop_all<0xD8, "psubusb", X86subus, v16i8, v32i8,
3650 SSE_INTALU_ITINS_P, 0>;
3651 defm PSUBUSW : PDI_binop_all<0xD9, "psubusw", X86subus, v8i16, v16i16,
3652 SSE_INTALU_ITINS_P, 0>;
3653 defm PMINUB : PDI_binop_all<0xDA, "pminub", X86umin, v16i8, v32i8,
3654 SSE_INTALU_ITINS_P, 1>;
3655 defm PMINSW : PDI_binop_all<0xEA, "pminsw", X86smin, v8i16, v16i16,
3656 SSE_INTALU_ITINS_P, 1>;
3657 defm PMAXUB : PDI_binop_all<0xDE, "pmaxub", X86umax, v16i8, v32i8,
3658 SSE_INTALU_ITINS_P, 1>;
3659 defm PMAXSW : PDI_binop_all<0xEE, "pmaxsw", X86smax, v8i16, v16i16,
3660 SSE_INTALU_ITINS_P, 1>;
3663 defm PSUBSB : PDI_binop_all_int<0xE8, "psubsb", int_x86_sse2_psubs_b,
3664 int_x86_avx2_psubs_b, SSE_INTALU_ITINS_P, 0>;
3665 defm PSUBSW : PDI_binop_all_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3666 int_x86_avx2_psubs_w, SSE_INTALU_ITINS_P, 0>;
3667 defm PADDSB : PDI_binop_all_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3668 int_x86_avx2_padds_b, SSE_INTALU_ITINS_P, 1>;
3669 defm PADDSW : PDI_binop_all_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3670 int_x86_avx2_padds_w, SSE_INTALU_ITINS_P, 1>;
3671 defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3672 int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
3673 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3674 int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
3675 defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3676 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
3677 defm PMULHW : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3678 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
3679 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3680 int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
3681 defm PAVGB : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3682 int_x86_avx2_pavg_b, SSE_INTALU_ITINS_P, 1>;
3683 defm PAVGW : PDI_binop_all_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3684 int_x86_avx2_pavg_w, SSE_INTALU_ITINS_P, 1>;
3685 defm PSADBW : PDI_binop_all_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3686 int_x86_avx2_psad_bw, SSE_INTALU_ITINS_P, 1>;
3688 let Predicates = [HasAVX] in
3689 defm VPMULUDQ : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v2i64, v4i32, VR128,
3690 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1, 0>,
3692 let Predicates = [HasAVX2] in
3693 defm VPMULUDQY : PDI_binop_rm2<0xF4, "vpmuludq", X86pmuludq, v4i64, v8i32,
3694 VR256, memopv4i64, i256mem,
3695 SSE_INTMUL_ITINS_P, 1, 0>, VEX_4V, VEX_L;
3696 let Constraints = "$src1 = $dst" in
3697 defm PMULUDQ : PDI_binop_rm2<0xF4, "pmuludq", X86pmuludq, v2i64, v4i32, VR128,
3698 memopv2i64, i128mem, SSE_INTMUL_ITINS_P, 1>;
3700 //===---------------------------------------------------------------------===//
3701 // SSE2 - Packed Integer Logical Instructions
3702 //===---------------------------------------------------------------------===//
3704 let Predicates = [HasAVX] in {
3705 defm VPSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3706 VR128, v8i16, v8i16, bc_v8i16,
3707 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3708 defm VPSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3709 VR128, v4i32, v4i32, bc_v4i32,
3710 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3711 defm VPSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3712 VR128, v2i64, v2i64, bc_v2i64,
3713 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3715 defm VPSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3716 VR128, v8i16, v8i16, bc_v8i16,
3717 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3718 defm VPSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3719 VR128, v4i32, v4i32, bc_v4i32,
3720 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3721 defm VPSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3722 VR128, v2i64, v2i64, bc_v2i64,
3723 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3725 defm VPSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3726 VR128, v8i16, v8i16, bc_v8i16,
3727 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3728 defm VPSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3729 VR128, v4i32, v4i32, bc_v4i32,
3730 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V;
3732 let ExeDomain = SSEPackedInt in {
3733 // 128-bit logical shifts.
3734 def VPSLLDQri : PDIi8<0x73, MRM7r,
3735 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3736 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3738 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>,
3740 def VPSRLDQri : PDIi8<0x73, MRM3r,
3741 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3742 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3744 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>,
3746 // PSRADQri doesn't exist in SSE[1-3].
3748 } // Predicates = [HasAVX]
3750 let Predicates = [HasAVX2] in {
3751 defm VPSLLWY : PDI_binop_rmi<0xF1, 0x71, MRM6r, "vpsllw", X86vshl, X86vshli,
3752 VR256, v16i16, v8i16, bc_v8i16,
3753 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3754 defm VPSLLDY : PDI_binop_rmi<0xF2, 0x72, MRM6r, "vpslld", X86vshl, X86vshli,
3755 VR256, v8i32, v4i32, bc_v4i32,
3756 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3757 defm VPSLLQY : PDI_binop_rmi<0xF3, 0x73, MRM6r, "vpsllq", X86vshl, X86vshli,
3758 VR256, v4i64, v2i64, bc_v2i64,
3759 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3761 defm VPSRLWY : PDI_binop_rmi<0xD1, 0x71, MRM2r, "vpsrlw", X86vsrl, X86vsrli,
3762 VR256, v16i16, v8i16, bc_v8i16,
3763 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3764 defm VPSRLDY : PDI_binop_rmi<0xD2, 0x72, MRM2r, "vpsrld", X86vsrl, X86vsrli,
3765 VR256, v8i32, v4i32, bc_v4i32,
3766 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3767 defm VPSRLQY : PDI_binop_rmi<0xD3, 0x73, MRM2r, "vpsrlq", X86vsrl, X86vsrli,
3768 VR256, v4i64, v2i64, bc_v2i64,
3769 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3771 defm VPSRAWY : PDI_binop_rmi<0xE1, 0x71, MRM4r, "vpsraw", X86vsra, X86vsrai,
3772 VR256, v16i16, v8i16, bc_v8i16,
3773 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3774 defm VPSRADY : PDI_binop_rmi<0xE2, 0x72, MRM4r, "vpsrad", X86vsra, X86vsrai,
3775 VR256, v8i32, v4i32, bc_v4i32,
3776 SSE_INTSHIFT_ITINS_P, 0>, VEX_4V, VEX_L;
3778 let ExeDomain = SSEPackedInt in {
3779 // 256-bit logical shifts.
3780 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3781 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3782 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3784 (int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2))]>,
3786 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3787 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3788 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3790 (int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2))]>,
3792 // PSRADQYri doesn't exist in SSE[1-3].
3794 } // Predicates = [HasAVX2]
3796 let Constraints = "$src1 = $dst" in {
3797 defm PSLLW : PDI_binop_rmi<0xF1, 0x71, MRM6r, "psllw", X86vshl, X86vshli,
3798 VR128, v8i16, v8i16, bc_v8i16,
3799 SSE_INTSHIFT_ITINS_P>;
3800 defm PSLLD : PDI_binop_rmi<0xF2, 0x72, MRM6r, "pslld", X86vshl, X86vshli,
3801 VR128, v4i32, v4i32, bc_v4i32,
3802 SSE_INTSHIFT_ITINS_P>;
3803 defm PSLLQ : PDI_binop_rmi<0xF3, 0x73, MRM6r, "psllq", X86vshl, X86vshli,
3804 VR128, v2i64, v2i64, bc_v2i64,
3805 SSE_INTSHIFT_ITINS_P>;
3807 defm PSRLW : PDI_binop_rmi<0xD1, 0x71, MRM2r, "psrlw", X86vsrl, X86vsrli,
3808 VR128, v8i16, v8i16, bc_v8i16,
3809 SSE_INTSHIFT_ITINS_P>;
3810 defm PSRLD : PDI_binop_rmi<0xD2, 0x72, MRM2r, "psrld", X86vsrl, X86vsrli,
3811 VR128, v4i32, v4i32, bc_v4i32,
3812 SSE_INTSHIFT_ITINS_P>;
3813 defm PSRLQ : PDI_binop_rmi<0xD3, 0x73, MRM2r, "psrlq", X86vsrl, X86vsrli,
3814 VR128, v2i64, v2i64, bc_v2i64,
3815 SSE_INTSHIFT_ITINS_P>;
3817 defm PSRAW : PDI_binop_rmi<0xE1, 0x71, MRM4r, "psraw", X86vsra, X86vsrai,
3818 VR128, v8i16, v8i16, bc_v8i16,
3819 SSE_INTSHIFT_ITINS_P>;
3820 defm PSRAD : PDI_binop_rmi<0xE2, 0x72, MRM4r, "psrad", X86vsra, X86vsrai,
3821 VR128, v4i32, v4i32, bc_v4i32,
3822 SSE_INTSHIFT_ITINS_P>;
3824 let ExeDomain = SSEPackedInt in {
3825 // 128-bit logical shifts.
3826 def PSLLDQri : PDIi8<0x73, MRM7r,
3827 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3828 "pslldq\t{$src2, $dst|$dst, $src2}",
3830 (int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2))]>;
3831 def PSRLDQri : PDIi8<0x73, MRM3r,
3832 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3833 "psrldq\t{$src2, $dst|$dst, $src2}",
3835 (int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2))]>;
3836 // PSRADQri doesn't exist in SSE[1-3].
3838 } // Constraints = "$src1 = $dst"
3840 let Predicates = [HasAVX] in {
3841 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3842 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3843 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3844 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3845 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3846 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3848 // Shift up / down and insert zero's.
3849 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3850 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3851 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3852 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3855 let Predicates = [HasAVX2] in {
3856 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3857 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3858 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3859 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3862 let Predicates = [UseSSE2] in {
3863 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3864 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3865 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3866 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3867 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3868 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3870 // Shift up / down and insert zero's.
3871 def : Pat<(v2i64 (X86vshldq VR128:$src, (i8 imm:$amt))),
3872 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3873 def : Pat<(v2i64 (X86vshrdq VR128:$src, (i8 imm:$amt))),
3874 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3877 //===---------------------------------------------------------------------===//
3878 // SSE2 - Packed Integer Comparison Instructions
3879 //===---------------------------------------------------------------------===//
3881 defm PCMPEQB : PDI_binop_all<0x74, "pcmpeqb", X86pcmpeq, v16i8, v32i8,
3882 SSE_INTALU_ITINS_P, 1>;
3883 defm PCMPEQW : PDI_binop_all<0x75, "pcmpeqw", X86pcmpeq, v8i16, v16i16,
3884 SSE_INTALU_ITINS_P, 1>;
3885 defm PCMPEQD : PDI_binop_all<0x76, "pcmpeqd", X86pcmpeq, v4i32, v8i32,
3886 SSE_INTALU_ITINS_P, 1>;
3887 defm PCMPGTB : PDI_binop_all<0x64, "pcmpgtb", X86pcmpgt, v16i8, v32i8,
3888 SSE_INTALU_ITINS_P, 0>;
3889 defm PCMPGTW : PDI_binop_all<0x65, "pcmpgtw", X86pcmpgt, v8i16, v16i16,
3890 SSE_INTALU_ITINS_P, 0>;
3891 defm PCMPGTD : PDI_binop_all<0x66, "pcmpgtd", X86pcmpgt, v4i32, v8i32,
3892 SSE_INTALU_ITINS_P, 0>;
3894 //===---------------------------------------------------------------------===//
3895 // SSE2 - Packed Integer Pack Instructions
3896 //===---------------------------------------------------------------------===//
3898 defm PACKSSWB : PDI_binop_all_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
3899 int_x86_avx2_packsswb, SSE_INTALU_ITINS_P, 0>;
3900 defm PACKSSDW : PDI_binop_all_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
3901 int_x86_avx2_packssdw, SSE_INTALU_ITINS_P, 0>;
3902 defm PACKUSWB : PDI_binop_all_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
3903 int_x86_avx2_packuswb, SSE_INTALU_ITINS_P, 0>;
3905 //===---------------------------------------------------------------------===//
3906 // SSE2 - Packed Integer Shuffle Instructions
3907 //===---------------------------------------------------------------------===//
3909 let ExeDomain = SSEPackedInt in {
3910 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, SDNode OpNode> {
3911 def ri : Ii8<0x70, MRMSrcReg,
3912 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3913 !strconcat(OpcodeStr,
3914 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3915 [(set VR128:$dst, (vt (OpNode VR128:$src1, (i8 imm:$src2))))],
3917 def mi : Ii8<0x70, MRMSrcMem,
3918 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3919 !strconcat(OpcodeStr,
3920 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3922 (vt (OpNode (bitconvert (memopv2i64 addr:$src1)),
3927 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, SDNode OpNode> {
3928 def Yri : Ii8<0x70, MRMSrcReg,
3929 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
3930 !strconcat(OpcodeStr,
3931 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3932 [(set VR256:$dst, (vt (OpNode VR256:$src1, (i8 imm:$src2))))]>;
3933 def Ymi : Ii8<0x70, MRMSrcMem,
3934 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
3935 !strconcat(OpcodeStr,
3936 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3938 (vt (OpNode (bitconvert (memopv4i64 addr:$src1)),
3939 (i8 imm:$src2))))]>;
3941 } // ExeDomain = SSEPackedInt
3943 let Predicates = [HasAVX] in {
3944 let AddedComplexity = 5 in
3945 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, X86PShufd>, TB, OpSize, VEX;
3947 // SSE2 with ImmT == Imm8 and XS prefix.
3948 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, X86PShufhw>, XS, VEX;
3950 // SSE2 with ImmT == Imm8 and XD prefix.
3951 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, X86PShuflw>, XD, VEX;
3953 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3954 (VPSHUFDmi addr:$src1, imm:$imm)>;
3955 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3956 (VPSHUFDri VR128:$src1, imm:$imm)>;
3959 let Predicates = [HasAVX2] in {
3960 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, X86PShufd>,
3961 TB, OpSize, VEX,VEX_L;
3962 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, X86PShufhw>,
3964 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, X86PShuflw>,
3968 let Predicates = [UseSSE2] in {
3969 let AddedComplexity = 5 in
3970 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, X86PShufd>, TB, OpSize;
3972 // SSE2 with ImmT == Imm8 and XS prefix.
3973 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, X86PShufhw>, XS;
3975 // SSE2 with ImmT == Imm8 and XD prefix.
3976 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, X86PShuflw>, XD;
3978 def : Pat<(v4f32 (X86PShufd (memopv4f32 addr:$src1), (i8 imm:$imm))),
3979 (PSHUFDmi addr:$src1, imm:$imm)>;
3980 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3981 (PSHUFDri VR128:$src1, imm:$imm)>;
3984 //===---------------------------------------------------------------------===//
3985 // SSE2 - Packed Integer Unpack Instructions
3986 //===---------------------------------------------------------------------===//
3988 let ExeDomain = SSEPackedInt in {
3989 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3990 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3991 def rr : PDI<opc, MRMSrcReg,
3992 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3994 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3995 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3996 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))],
3998 def rm : PDI<opc, MRMSrcMem,
3999 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4001 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4002 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4003 [(set VR128:$dst, (OpNode VR128:$src1,
4004 (bc_frag (memopv2i64
4009 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4010 SDNode OpNode, PatFrag bc_frag> {
4011 def Yrr : PDI<opc, MRMSrcReg,
4012 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4013 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4014 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4015 def Yrm : PDI<opc, MRMSrcMem,
4016 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4017 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4018 [(set VR256:$dst, (OpNode VR256:$src1,
4019 (bc_frag (memopv4i64 addr:$src2))))]>;
4022 let Predicates = [HasAVX] in {
4023 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4024 bc_v16i8, 0>, VEX_4V;
4025 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4026 bc_v8i16, 0>, VEX_4V;
4027 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4028 bc_v4i32, 0>, VEX_4V;
4029 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4030 bc_v2i64, 0>, VEX_4V;
4032 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4033 bc_v16i8, 0>, VEX_4V;
4034 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4035 bc_v8i16, 0>, VEX_4V;
4036 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4037 bc_v4i32, 0>, VEX_4V;
4038 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4039 bc_v2i64, 0>, VEX_4V;
4042 let Predicates = [HasAVX2] in {
4043 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4044 bc_v32i8>, VEX_4V, VEX_L;
4045 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4046 bc_v16i16>, VEX_4V, VEX_L;
4047 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4048 bc_v8i32>, VEX_4V, VEX_L;
4049 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4050 bc_v4i64>, VEX_4V, VEX_L;
4052 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4053 bc_v32i8>, VEX_4V, VEX_L;
4054 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4055 bc_v16i16>, VEX_4V, VEX_L;
4056 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4057 bc_v8i32>, VEX_4V, VEX_L;
4058 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4059 bc_v4i64>, VEX_4V, VEX_L;
4062 let Constraints = "$src1 = $dst" in {
4063 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4065 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4067 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4069 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4072 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4074 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4076 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4078 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4081 } // ExeDomain = SSEPackedInt
4083 //===---------------------------------------------------------------------===//
4084 // SSE2 - Packed Integer Extract and Insert
4085 //===---------------------------------------------------------------------===//
4087 let ExeDomain = SSEPackedInt in {
4088 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4089 def rri : Ii8<0xC4, MRMSrcReg,
4090 (outs VR128:$dst), (ins VR128:$src1,
4091 GR32:$src2, i32i8imm:$src3),
4093 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4094 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4096 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))], IIC_SSE_PINSRW>;
4097 def rmi : Ii8<0xC4, MRMSrcMem,
4098 (outs VR128:$dst), (ins VR128:$src1,
4099 i16mem:$src2, i32i8imm:$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, (extloadi16 addr:$src2),
4105 imm:$src3))], IIC_SSE_PINSRW>;
4109 let Predicates = [HasAVX] in
4110 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4111 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4112 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4113 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4114 imm:$src2))]>, TB, OpSize, VEX;
4115 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4116 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4117 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4118 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4119 imm:$src2))], IIC_SSE_PEXTRW>;
4122 let Predicates = [HasAVX] in {
4123 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4124 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4125 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4126 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4127 []>, TB, OpSize, VEX_4V;
4130 let Constraints = "$src1 = $dst" in
4131 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[UseSSE2]>;
4133 } // ExeDomain = SSEPackedInt
4135 //===---------------------------------------------------------------------===//
4136 // SSE2 - Packed Mask Creation
4137 //===---------------------------------------------------------------------===//
4139 let ExeDomain = SSEPackedInt in {
4141 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4142 "pmovmskb\t{$src, $dst|$dst, $src}",
4143 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4144 IIC_SSE_MOVMSK>, VEX;
4145 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4146 "pmovmskb\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVMSK>, VEX;
4148 let Predicates = [HasAVX2] in {
4149 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4150 "pmovmskb\t{$src, $dst|$dst, $src}",
4151 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX, VEX_L;
4152 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4153 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4156 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4157 "pmovmskb\t{$src, $dst|$dst, $src}",
4158 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))],
4161 } // ExeDomain = SSEPackedInt
4163 //===---------------------------------------------------------------------===//
4164 // SSE2 - Conditional Store
4165 //===---------------------------------------------------------------------===//
4167 let ExeDomain = SSEPackedInt in {
4170 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4171 (ins VR128:$src, VR128:$mask),
4172 "maskmovdqu\t{$mask, $src|$src, $mask}",
4173 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4174 IIC_SSE_MASKMOV>, VEX;
4176 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4177 (ins VR128:$src, VR128:$mask),
4178 "maskmovdqu\t{$mask, $src|$src, $mask}",
4179 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4180 IIC_SSE_MASKMOV>, VEX;
4183 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4184 "maskmovdqu\t{$mask, $src|$src, $mask}",
4185 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)],
4188 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4189 "maskmovdqu\t{$mask, $src|$src, $mask}",
4190 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)],
4193 } // ExeDomain = SSEPackedInt
4195 //===---------------------------------------------------------------------===//
4196 // SSE2 - Move Doubleword
4197 //===---------------------------------------------------------------------===//
4199 //===---------------------------------------------------------------------===//
4200 // Move Int Doubleword to Packed Double Int
4202 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4203 "movd\t{$src, $dst|$dst, $src}",
4205 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>,
4207 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4208 "movd\t{$src, $dst|$dst, $src}",
4210 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4213 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4214 "mov{d|q}\t{$src, $dst|$dst, $src}",
4216 (v2i64 (scalar_to_vector GR64:$src)))],
4217 IIC_SSE_MOVDQ>, VEX;
4218 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4219 "mov{d|q}\t{$src, $dst|$dst, $src}",
4220 [(set FR64:$dst, (bitconvert GR64:$src))],
4221 IIC_SSE_MOVDQ>, VEX;
4223 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4224 "movd\t{$src, $dst|$dst, $src}",
4226 (v4i32 (scalar_to_vector GR32:$src)))], IIC_SSE_MOVDQ>;
4227 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4228 "movd\t{$src, $dst|$dst, $src}",
4230 (v4i32 (scalar_to_vector (loadi32 addr:$src))))],
4232 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4233 "mov{d|q}\t{$src, $dst|$dst, $src}",
4235 (v2i64 (scalar_to_vector GR64:$src)))],
4237 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4238 "mov{d|q}\t{$src, $dst|$dst, $src}",
4239 [(set FR64:$dst, (bitconvert GR64:$src))],
4242 //===---------------------------------------------------------------------===//
4243 // Move Int Doubleword to Single Scalar
4245 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4246 "movd\t{$src, $dst|$dst, $src}",
4247 [(set FR32:$dst, (bitconvert GR32:$src))],
4248 IIC_SSE_MOVDQ>, VEX;
4250 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4251 "movd\t{$src, $dst|$dst, $src}",
4252 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4255 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4256 "movd\t{$src, $dst|$dst, $src}",
4257 [(set FR32:$dst, (bitconvert GR32:$src))],
4260 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4261 "movd\t{$src, $dst|$dst, $src}",
4262 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))],
4265 //===---------------------------------------------------------------------===//
4266 // Move Packed Doubleword Int to Packed Double Int
4268 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4269 "movd\t{$src, $dst|$dst, $src}",
4270 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4271 (iPTR 0)))], IIC_SSE_MOVD_ToGP>, VEX;
4272 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4273 (ins i32mem:$dst, VR128:$src),
4274 "movd\t{$src, $dst|$dst, $src}",
4275 [(store (i32 (vector_extract (v4i32 VR128:$src),
4276 (iPTR 0))), addr:$dst)], IIC_SSE_MOVDQ>,
4278 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4279 "movd\t{$src, $dst|$dst, $src}",
4280 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4281 (iPTR 0)))], IIC_SSE_MOVD_ToGP>;
4282 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4283 "movd\t{$src, $dst|$dst, $src}",
4284 [(store (i32 (vector_extract (v4i32 VR128:$src),
4285 (iPTR 0))), addr:$dst)],
4288 //===---------------------------------------------------------------------===//
4289 // Move Packed Doubleword Int first element to Doubleword Int
4291 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4292 "vmov{d|q}\t{$src, $dst|$dst, $src}",
4293 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4296 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4298 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4299 "mov{d|q}\t{$src, $dst|$dst, $src}",
4300 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4304 //===---------------------------------------------------------------------===//
4305 // Bitcast FR64 <-> GR64
4307 let Predicates = [HasAVX] in
4308 def VMOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4309 "vmovq\t{$src, $dst|$dst, $src}",
4310 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4312 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4313 "mov{d|q}\t{$src, $dst|$dst, $src}",
4314 [(set GR64:$dst, (bitconvert FR64:$src))],
4315 IIC_SSE_MOVDQ>, VEX;
4316 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4317 "movq\t{$src, $dst|$dst, $src}",
4318 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4319 IIC_SSE_MOVDQ>, VEX;
4321 def MOV64toSDrm : S2SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4322 "movq\t{$src, $dst|$dst, $src}",
4323 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))],
4325 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4326 "mov{d|q}\t{$src, $dst|$dst, $src}",
4327 [(set GR64:$dst, (bitconvert FR64:$src))],
4329 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4330 "movq\t{$src, $dst|$dst, $src}",
4331 [(store (i64 (bitconvert FR64:$src)), addr:$dst)],
4334 //===---------------------------------------------------------------------===//
4335 // Move Scalar Single to Double Int
4337 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4338 "movd\t{$src, $dst|$dst, $src}",
4339 [(set GR32:$dst, (bitconvert FR32:$src))],
4340 IIC_SSE_MOVD_ToGP>, VEX;
4341 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4342 "movd\t{$src, $dst|$dst, $src}",
4343 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4344 IIC_SSE_MOVDQ>, VEX;
4345 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4346 "movd\t{$src, $dst|$dst, $src}",
4347 [(set GR32:$dst, (bitconvert FR32:$src))],
4349 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4350 "movd\t{$src, $dst|$dst, $src}",
4351 [(store (i32 (bitconvert FR32:$src)), addr:$dst)],
4354 //===---------------------------------------------------------------------===//
4355 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4357 let AddedComplexity = 15 in {
4358 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4359 "movd\t{$src, $dst|$dst, $src}",
4360 [(set VR128:$dst, (v4i32 (X86vzmovl
4361 (v4i32 (scalar_to_vector GR32:$src)))))],
4362 IIC_SSE_MOVDQ>, VEX;
4363 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4364 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4365 [(set VR128:$dst, (v2i64 (X86vzmovl
4366 (v2i64 (scalar_to_vector GR64:$src)))))],
4370 let AddedComplexity = 15 in {
4371 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4372 "movd\t{$src, $dst|$dst, $src}",
4373 [(set VR128:$dst, (v4i32 (X86vzmovl
4374 (v4i32 (scalar_to_vector GR32:$src)))))],
4376 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4377 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4378 [(set VR128:$dst, (v2i64 (X86vzmovl
4379 (v2i64 (scalar_to_vector GR64:$src)))))],
4383 let AddedComplexity = 20 in {
4384 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4385 "movd\t{$src, $dst|$dst, $src}",
4387 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4388 (loadi32 addr:$src))))))],
4389 IIC_SSE_MOVDQ>, VEX;
4390 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4391 "movd\t{$src, $dst|$dst, $src}",
4393 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4394 (loadi32 addr:$src))))))],
4398 let Predicates = [HasAVX] in {
4399 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4400 let AddedComplexity = 20 in {
4401 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4402 (VMOVZDI2PDIrm addr:$src)>;
4403 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4404 (VMOVZDI2PDIrm addr:$src)>;
4406 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4407 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4408 (v4i32 (scalar_to_vector GR32:$src)),(iPTR 0)))),
4409 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4410 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4411 (v2i64 (scalar_to_vector GR64:$src)),(iPTR 0)))),
4412 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4415 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4416 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4417 (MOVZDI2PDIrm addr:$src)>;
4418 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4419 (MOVZDI2PDIrm addr:$src)>;
4422 // These are the correct encodings of the instructions so that we know how to
4423 // read correct assembly, even though we continue to emit the wrong ones for
4424 // compatibility with Darwin's buggy assembler.
4425 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4426 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4427 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4428 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4429 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4430 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4431 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4432 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4433 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4434 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4435 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4436 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4438 //===---------------------------------------------------------------------===//
4439 // SSE2 - Move Quadword
4440 //===---------------------------------------------------------------------===//
4442 //===---------------------------------------------------------------------===//
4443 // Move Quadword Int to Packed Quadword Int
4445 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4446 "vmovq\t{$src, $dst|$dst, $src}",
4448 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4449 VEX, Requires<[HasAVX]>;
4450 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4451 "movq\t{$src, $dst|$dst, $src}",
4453 (v2i64 (scalar_to_vector (loadi64 addr:$src))))],
4455 Requires<[UseSSE2]>; // SSE2 instruction with XS Prefix
4457 //===---------------------------------------------------------------------===//
4458 // Move Packed Quadword Int to Quadword Int
4460 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4461 "movq\t{$src, $dst|$dst, $src}",
4462 [(store (i64 (vector_extract (v2i64 VR128:$src),
4463 (iPTR 0))), addr:$dst)],
4464 IIC_SSE_MOVDQ>, VEX;
4465 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4466 "movq\t{$src, $dst|$dst, $src}",
4467 [(store (i64 (vector_extract (v2i64 VR128:$src),
4468 (iPTR 0))), addr:$dst)],
4471 //===---------------------------------------------------------------------===//
4472 // Store / copy lower 64-bits of a XMM register.
4474 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4475 "movq\t{$src, $dst|$dst, $src}",
4476 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4477 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4478 "movq\t{$src, $dst|$dst, $src}",
4479 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)],
4482 let AddedComplexity = 20 in
4483 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4484 "vmovq\t{$src, $dst|$dst, $src}",
4486 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4487 (loadi64 addr:$src))))))],
4489 XS, VEX, Requires<[HasAVX]>;
4491 let AddedComplexity = 20 in
4492 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4493 "movq\t{$src, $dst|$dst, $src}",
4495 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4496 (loadi64 addr:$src))))))],
4498 XS, Requires<[UseSSE2]>;
4500 let Predicates = [HasAVX], AddedComplexity = 20 in {
4501 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4502 (VMOVZQI2PQIrm addr:$src)>;
4503 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4504 (VMOVZQI2PQIrm addr:$src)>;
4505 def : Pat<(v2i64 (X86vzload addr:$src)),
4506 (VMOVZQI2PQIrm addr:$src)>;
4509 let Predicates = [UseSSE2], AddedComplexity = 20 in {
4510 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4511 (MOVZQI2PQIrm addr:$src)>;
4512 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4513 (MOVZQI2PQIrm addr:$src)>;
4514 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4517 let Predicates = [HasAVX] in {
4518 def : Pat<(v4i64 (alignedX86vzload addr:$src)),
4519 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4520 def : Pat<(v4i64 (X86vzload addr:$src)),
4521 (SUBREG_TO_REG (i32 0), (VMOVUPSrm addr:$src), sub_xmm)>;
4524 //===---------------------------------------------------------------------===//
4525 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4526 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4528 let AddedComplexity = 15 in
4529 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4530 "vmovq\t{$src, $dst|$dst, $src}",
4531 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4533 XS, VEX, Requires<[HasAVX]>;
4534 let AddedComplexity = 15 in
4535 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4536 "movq\t{$src, $dst|$dst, $src}",
4537 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))],
4539 XS, Requires<[UseSSE2]>;
4541 let AddedComplexity = 20 in
4542 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4543 "vmovq\t{$src, $dst|$dst, $src}",
4544 [(set VR128:$dst, (v2i64 (X86vzmovl
4545 (loadv2i64 addr:$src))))],
4547 XS, VEX, Requires<[HasAVX]>;
4548 let AddedComplexity = 20 in {
4549 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4550 "movq\t{$src, $dst|$dst, $src}",
4551 [(set VR128:$dst, (v2i64 (X86vzmovl
4552 (loadv2i64 addr:$src))))],
4554 XS, Requires<[UseSSE2]>;
4557 let AddedComplexity = 20 in {
4558 let Predicates = [HasAVX] in {
4559 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4560 (VMOVZPQILo2PQIrm addr:$src)>;
4561 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4562 (VMOVZPQILo2PQIrr VR128:$src)>;
4564 let Predicates = [UseSSE2] in {
4565 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4566 (MOVZPQILo2PQIrm addr:$src)>;
4567 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4568 (MOVZPQILo2PQIrr VR128:$src)>;
4572 // Instructions to match in the assembler
4573 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4574 "movq\t{$src, $dst|$dst, $src}", [],
4575 IIC_SSE_MOVDQ>, VEX, VEX_W;
4576 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4577 "movq\t{$src, $dst|$dst, $src}", [],
4578 IIC_SSE_MOVDQ>, VEX, VEX_W;
4579 // Recognize "movd" with GR64 destination, but encode as a "movq"
4580 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4581 "movd\t{$src, $dst|$dst, $src}", [],
4582 IIC_SSE_MOVDQ>, VEX, VEX_W;
4584 // Instructions for the disassembler
4585 // xr = XMM register
4588 let Predicates = [HasAVX] in
4589 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4590 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4591 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4592 "movq\t{$src, $dst|$dst, $src}", [], IIC_SSE_MOVQ_RR>, XS;
4594 //===---------------------------------------------------------------------===//
4595 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4596 //===---------------------------------------------------------------------===//
4597 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4598 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4599 X86MemOperand x86memop> {
4600 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4601 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4602 [(set RC:$dst, (vt (OpNode RC:$src)))],
4604 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4605 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4606 [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
4610 let Predicates = [HasAVX] in {
4611 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4612 v4f32, VR128, memopv4f32, f128mem>, VEX;
4613 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4614 v4f32, VR128, memopv4f32, f128mem>, VEX;
4615 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4616 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4617 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4618 v8f32, VR256, memopv8f32, f256mem>, VEX, VEX_L;
4620 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4621 memopv4f32, f128mem>;
4622 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4623 memopv4f32, f128mem>;
4625 let Predicates = [HasAVX] in {
4626 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4627 (VMOVSHDUPrr VR128:$src)>;
4628 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4629 (VMOVSHDUPrm addr:$src)>;
4630 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4631 (VMOVSLDUPrr VR128:$src)>;
4632 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4633 (VMOVSLDUPrm addr:$src)>;
4634 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4635 (VMOVSHDUPYrr VR256:$src)>;
4636 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4637 (VMOVSHDUPYrm addr:$src)>;
4638 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4639 (VMOVSLDUPYrr VR256:$src)>;
4640 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4641 (VMOVSLDUPYrm addr:$src)>;
4644 let Predicates = [UseSSE3] in {
4645 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4646 (MOVSHDUPrr VR128:$src)>;
4647 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4648 (MOVSHDUPrm addr:$src)>;
4649 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4650 (MOVSLDUPrr VR128:$src)>;
4651 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4652 (MOVSLDUPrm addr:$src)>;
4655 //===---------------------------------------------------------------------===//
4656 // SSE3 - Replicate Double FP - MOVDDUP
4657 //===---------------------------------------------------------------------===//
4659 multiclass sse3_replicate_dfp<string OpcodeStr> {
4660 let neverHasSideEffects = 1 in
4661 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4662 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4663 [], IIC_SSE_MOV_LH>;
4664 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4665 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4668 (scalar_to_vector (loadf64 addr:$src)))))],
4672 // FIXME: Merge with above classe when there're patterns for the ymm version
4673 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4674 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4675 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4676 [(set VR256:$dst, (v4f64 (X86Movddup VR256:$src)))]>;
4677 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4678 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4681 (scalar_to_vector (loadf64 addr:$src)))))]>;
4684 let Predicates = [HasAVX] in {
4685 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4686 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX, VEX_L;
4689 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4691 let Predicates = [HasAVX] in {
4692 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4693 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4694 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4695 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4696 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4697 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4698 def : Pat<(X86Movddup (bc_v2f64
4699 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4700 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4703 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4704 (VMOVDDUPYrm addr:$src)>;
4705 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4706 (VMOVDDUPYrm addr:$src)>;
4707 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4708 (VMOVDDUPYrm addr:$src)>;
4709 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4710 (VMOVDDUPYrr VR256:$src)>;
4713 let Predicates = [UseSSE3] in {
4714 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4715 (MOVDDUPrm addr:$src)>;
4716 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4717 (MOVDDUPrm addr:$src)>;
4718 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4719 (MOVDDUPrm addr:$src)>;
4720 def : Pat<(X86Movddup (bc_v2f64
4721 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4722 (MOVDDUPrm addr:$src)>;
4725 //===---------------------------------------------------------------------===//
4726 // SSE3 - Move Unaligned Integer
4727 //===---------------------------------------------------------------------===//
4729 let Predicates = [HasAVX] in {
4730 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4731 "vlddqu\t{$src, $dst|$dst, $src}",
4732 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4733 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4734 "vlddqu\t{$src, $dst|$dst, $src}",
4735 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>,
4738 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4739 "lddqu\t{$src, $dst|$dst, $src}",
4740 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))],
4743 //===---------------------------------------------------------------------===//
4744 // SSE3 - Arithmetic
4745 //===---------------------------------------------------------------------===//
4747 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4748 X86MemOperand x86memop, OpndItins itins,
4750 def rr : I<0xD0, MRMSrcReg,
4751 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4753 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4754 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4755 [(set RC:$dst, (Int RC:$src1, RC:$src2))], itins.rr>;
4756 def rm : I<0xD0, MRMSrcMem,
4757 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4759 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4760 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4761 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))], itins.rr>;
4764 let Predicates = [HasAVX] in {
4765 let ExeDomain = SSEPackedSingle in {
4766 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4767 f128mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V;
4768 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4769 f256mem, SSE_ALU_F32P, 0>, TB, XD, VEX_4V, VEX_L;
4771 let ExeDomain = SSEPackedDouble in {
4772 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4773 f128mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V;
4774 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4775 f256mem, SSE_ALU_F64P, 0>, TB, OpSize, VEX_4V, VEX_L;
4778 let Constraints = "$src1 = $dst", Predicates = [UseSSE3] in {
4779 let ExeDomain = SSEPackedSingle in
4780 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4781 f128mem, SSE_ALU_F32P>, TB, XD;
4782 let ExeDomain = SSEPackedDouble in
4783 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4784 f128mem, SSE_ALU_F64P>, TB, OpSize;
4787 //===---------------------------------------------------------------------===//
4788 // SSE3 Instructions
4789 //===---------------------------------------------------------------------===//
4792 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4793 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4794 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4796 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4797 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4798 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4800 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4802 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4803 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4804 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4805 IIC_SSE_HADDSUB_RM>;
4807 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4808 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
4809 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4811 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4812 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4813 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], IIC_SSE_HADDSUB_RR>;
4815 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4817 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4818 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4819 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))],
4820 IIC_SSE_HADDSUB_RM>;
4823 let Predicates = [HasAVX] in {
4824 let ExeDomain = SSEPackedSingle in {
4825 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4826 X86fhadd, 0>, VEX_4V;
4827 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4828 X86fhsub, 0>, VEX_4V;
4829 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4830 X86fhadd, 0>, VEX_4V, VEX_L;
4831 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4832 X86fhsub, 0>, VEX_4V, VEX_L;
4834 let ExeDomain = SSEPackedDouble in {
4835 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4836 X86fhadd, 0>, VEX_4V;
4837 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4838 X86fhsub, 0>, VEX_4V;
4839 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4840 X86fhadd, 0>, VEX_4V, VEX_L;
4841 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4842 X86fhsub, 0>, VEX_4V, VEX_L;
4846 let Constraints = "$src1 = $dst" in {
4847 let ExeDomain = SSEPackedSingle in {
4848 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
4849 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
4851 let ExeDomain = SSEPackedDouble in {
4852 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
4853 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
4857 //===---------------------------------------------------------------------===//
4858 // SSSE3 - Packed Absolute Instructions
4859 //===---------------------------------------------------------------------===//
4862 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4863 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4864 Intrinsic IntId128> {
4865 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4867 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4868 [(set VR128:$dst, (IntId128 VR128:$src))], IIC_SSE_PABS_RR>,
4871 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4873 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4876 (bitconvert (memopv2i64 addr:$src))))], IIC_SSE_PABS_RM>,
4880 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4881 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
4882 Intrinsic IntId256> {
4883 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4885 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4886 [(set VR256:$dst, (IntId256 VR256:$src))]>,
4889 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4891 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4894 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
4897 let Predicates = [HasAVX] in {
4898 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
4899 int_x86_ssse3_pabs_b_128>, VEX;
4900 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
4901 int_x86_ssse3_pabs_w_128>, VEX;
4902 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
4903 int_x86_ssse3_pabs_d_128>, VEX;
4906 let Predicates = [HasAVX2] in {
4907 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
4908 int_x86_avx2_pabs_b>, VEX, VEX_L;
4909 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
4910 int_x86_avx2_pabs_w>, VEX, VEX_L;
4911 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
4912 int_x86_avx2_pabs_d>, VEX, VEX_L;
4915 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
4916 int_x86_ssse3_pabs_b_128>;
4917 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
4918 int_x86_ssse3_pabs_w_128>;
4919 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
4920 int_x86_ssse3_pabs_d_128>;
4922 //===---------------------------------------------------------------------===//
4923 // SSSE3 - Packed Binary Operator Instructions
4924 //===---------------------------------------------------------------------===//
4926 def SSE_PHADDSUBD : OpndItins<
4927 IIC_SSE_PHADDSUBD_RR, IIC_SSE_PHADDSUBD_RM
4929 def SSE_PHADDSUBSW : OpndItins<
4930 IIC_SSE_PHADDSUBSW_RR, IIC_SSE_PHADDSUBSW_RM
4932 def SSE_PHADDSUBW : OpndItins<
4933 IIC_SSE_PHADDSUBW_RR, IIC_SSE_PHADDSUBW_RM
4935 def SSE_PSHUFB : OpndItins<
4936 IIC_SSE_PSHUFB_RR, IIC_SSE_PSHUFB_RM
4938 def SSE_PSIGN : OpndItins<
4939 IIC_SSE_PSIGN_RR, IIC_SSE_PSIGN_RM
4941 def SSE_PMULHRSW : OpndItins<
4942 IIC_SSE_PMULHRSW, IIC_SSE_PMULHRSW
4945 /// SS3I_binop_rm - Simple SSSE3 bin op
4946 multiclass SS3I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4947 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
4948 X86MemOperand x86memop, OpndItins itins,
4950 let isCommutable = 1 in
4951 def rr : SS38I<opc, MRMSrcReg, (outs RC:$dst),
4952 (ins RC:$src1, RC:$src2),
4954 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4955 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4956 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))], itins.rr>,
4958 def rm : SS38I<opc, MRMSrcMem, (outs RC:$dst),
4959 (ins RC:$src1, x86memop:$src2),
4961 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4962 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4964 (OpVT (OpNode RC:$src1,
4965 (bitconvert (memop_frag addr:$src2)))))], itins.rm>, OpSize;
4968 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4969 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4970 Intrinsic IntId128, OpndItins itins,
4972 let isCommutable = 1 in
4973 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4974 (ins VR128:$src1, VR128:$src2),
4976 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4977 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4978 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4980 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4981 (ins VR128:$src1, i128mem:$src2),
4983 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4984 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4986 (IntId128 VR128:$src1,
4987 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
4990 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
4991 Intrinsic IntId256> {
4992 let isCommutable = 1 in
4993 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
4994 (ins VR256:$src1, VR256:$src2),
4995 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4996 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
4998 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
4999 (ins VR256:$src1, i256mem:$src2),
5000 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5002 (IntId256 VR256:$src1,
5003 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5006 let ImmT = NoImm, Predicates = [HasAVX] in {
5007 let isCommutable = 0 in {
5008 defm VPHADDW : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
5009 memopv2i64, i128mem,
5010 SSE_PHADDSUBW, 0>, VEX_4V;
5011 defm VPHADDD : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v4i32, VR128,
5012 memopv2i64, i128mem,
5013 SSE_PHADDSUBD, 0>, VEX_4V;
5014 defm VPHSUBW : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v8i16, VR128,
5015 memopv2i64, i128mem,
5016 SSE_PHADDSUBW, 0>, VEX_4V;
5017 defm VPHSUBD : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v4i32, VR128,
5018 memopv2i64, i128mem,
5019 SSE_PHADDSUBD, 0>, VEX_4V;
5020 defm VPSIGNB : SS3I_binop_rm<0x08, "vpsignb", X86psign, v16i8, VR128,
5021 memopv2i64, i128mem,
5022 SSE_PSIGN, 0>, VEX_4V;
5023 defm VPSIGNW : SS3I_binop_rm<0x09, "vpsignw", X86psign, v8i16, VR128,
5024 memopv2i64, i128mem,
5025 SSE_PSIGN, 0>, VEX_4V;
5026 defm VPSIGND : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v4i32, VR128,
5027 memopv2i64, i128mem,
5028 SSE_PSIGN, 0>, VEX_4V;
5029 defm VPSHUFB : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
5030 memopv2i64, i128mem,
5031 SSE_PSHUFB, 0>, VEX_4V;
5032 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5033 int_x86_ssse3_phadd_sw_128,
5034 SSE_PHADDSUBSW, 0>, VEX_4V;
5035 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5036 int_x86_ssse3_phsub_sw_128,
5037 SSE_PHADDSUBSW, 0>, VEX_4V;
5038 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5039 int_x86_ssse3_pmadd_ub_sw_128,
5040 SSE_PMADD, 0>, VEX_4V;
5042 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5043 int_x86_ssse3_pmul_hr_sw_128,
5044 SSE_PMULHRSW, 0>, VEX_4V;
5047 let ImmT = NoImm, Predicates = [HasAVX2] in {
5048 let isCommutable = 0 in {
5049 defm VPHADDWY : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
5050 memopv4i64, i256mem,
5051 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5052 defm VPHADDDY : SS3I_binop_rm<0x02, "vphaddd", X86hadd, v8i32, VR256,
5053 memopv4i64, i256mem,
5054 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5055 defm VPHSUBWY : SS3I_binop_rm<0x05, "vphsubw", X86hsub, v16i16, VR256,
5056 memopv4i64, i256mem,
5057 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5058 defm VPHSUBDY : SS3I_binop_rm<0x06, "vphsubd", X86hsub, v8i32, VR256,
5059 memopv4i64, i256mem,
5060 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5061 defm VPSIGNBY : SS3I_binop_rm<0x08, "vpsignb", X86psign, v32i8, VR256,
5062 memopv4i64, i256mem,
5063 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5064 defm VPSIGNWY : SS3I_binop_rm<0x09, "vpsignw", X86psign, v16i16, VR256,
5065 memopv4i64, i256mem,
5066 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5067 defm VPSIGNDY : SS3I_binop_rm<0x0A, "vpsignd", X86psign, v8i32, VR256,
5068 memopv4i64, i256mem,
5069 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5070 defm VPSHUFBY : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
5071 memopv4i64, i256mem,
5072 SSE_PHADDSUBW, 0>, VEX_4V, VEX_L;
5073 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5074 int_x86_avx2_phadd_sw>, VEX_4V, VEX_L;
5075 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5076 int_x86_avx2_phsub_sw>, VEX_4V, VEX_L;
5077 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5078 int_x86_avx2_pmadd_ub_sw>, VEX_4V, VEX_L;
5080 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5081 int_x86_avx2_pmul_hr_sw>, VEX_4V, VEX_L;
5084 // None of these have i8 immediate fields.
5085 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5086 let isCommutable = 0 in {
5087 defm PHADDW : SS3I_binop_rm<0x01, "phaddw", X86hadd, v8i16, VR128,
5088 memopv2i64, i128mem, SSE_PHADDSUBW>;
5089 defm PHADDD : SS3I_binop_rm<0x02, "phaddd", X86hadd, v4i32, VR128,
5090 memopv2i64, i128mem, SSE_PHADDSUBD>;
5091 defm PHSUBW : SS3I_binop_rm<0x05, "phsubw", X86hsub, v8i16, VR128,
5092 memopv2i64, i128mem, SSE_PHADDSUBW>;
5093 defm PHSUBD : SS3I_binop_rm<0x06, "phsubd", X86hsub, v4i32, VR128,
5094 memopv2i64, i128mem, SSE_PHADDSUBD>;
5095 defm PSIGNB : SS3I_binop_rm<0x08, "psignb", X86psign, v16i8, VR128,
5096 memopv2i64, i128mem, SSE_PSIGN>;
5097 defm PSIGNW : SS3I_binop_rm<0x09, "psignw", X86psign, v8i16, VR128,
5098 memopv2i64, i128mem, SSE_PSIGN>;
5099 defm PSIGND : SS3I_binop_rm<0x0A, "psignd", X86psign, v4i32, VR128,
5100 memopv2i64, i128mem, SSE_PSIGN>;
5101 defm PSHUFB : SS3I_binop_rm<0x00, "pshufb", X86pshufb, v16i8, VR128,
5102 memopv2i64, i128mem, SSE_PSHUFB>;
5103 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5104 int_x86_ssse3_phadd_sw_128,
5106 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5107 int_x86_ssse3_phsub_sw_128,
5109 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5110 int_x86_ssse3_pmadd_ub_sw_128, SSE_PMADD>;
5112 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5113 int_x86_ssse3_pmul_hr_sw_128,
5117 //===---------------------------------------------------------------------===//
5118 // SSSE3 - Packed Align Instruction Patterns
5119 //===---------------------------------------------------------------------===//
5121 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5122 let neverHasSideEffects = 1 in {
5123 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5124 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5126 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5128 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5129 [], IIC_SSE_PALIGNR>, OpSize;
5131 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5132 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5134 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5136 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5137 [], IIC_SSE_PALIGNR>, OpSize;
5141 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5142 let neverHasSideEffects = 1 in {
5143 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5144 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5146 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5149 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5150 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5152 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5157 let Predicates = [HasAVX] in
5158 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5159 let Predicates = [HasAVX2] in
5160 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V, VEX_L;
5161 let Constraints = "$src1 = $dst", Predicates = [UseSSSE3] in
5162 defm PALIGN : ssse3_palign<"palignr">;
5164 let Predicates = [HasAVX2] in {
5165 def : Pat<(v8i32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5166 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5167 def : Pat<(v8f32 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5168 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5169 def : Pat<(v16i16 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5170 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5171 def : Pat<(v32i8 (X86PAlign VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5172 (VPALIGNR256rr VR256:$src2, VR256:$src1, imm:$imm)>;
5175 let Predicates = [HasAVX] in {
5176 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5177 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5178 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5179 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5180 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5181 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5182 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5183 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5186 let Predicates = [UseSSSE3] in {
5187 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5188 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5189 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5190 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5191 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5192 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5193 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5194 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5197 //===---------------------------------------------------------------------===//
5198 // SSSE3 - Thread synchronization
5199 //===---------------------------------------------------------------------===//
5201 let usesCustomInserter = 1 in {
5202 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5203 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5204 Requires<[HasSSE3]>;
5207 let Uses = [EAX, ECX, EDX] in
5208 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", [], IIC_SSE_MONITOR>,
5209 TB, Requires<[HasSSE3]>;
5210 let Uses = [ECX, EAX] in
5211 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait",
5212 [(int_x86_sse3_mwait ECX, EAX)], IIC_SSE_MWAIT>,
5213 TB, Requires<[HasSSE3]>;
5215 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5216 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5218 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5219 Requires<[In32BitMode]>;
5220 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5221 Requires<[In64BitMode]>;
5223 //===----------------------------------------------------------------------===//
5224 // SSE4.1 - Packed Move with Sign/Zero Extend
5225 //===----------------------------------------------------------------------===//
5227 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5228 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5229 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5230 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5232 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5233 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5235 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5239 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5241 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5242 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5243 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5245 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5246 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5247 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5250 let Predicates = [HasAVX] in {
5251 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5253 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5255 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5257 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5259 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5261 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5265 let Predicates = [HasAVX2] in {
5266 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5267 int_x86_avx2_pmovsxbw>, VEX, VEX_L;
5268 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5269 int_x86_avx2_pmovsxwd>, VEX, VEX_L;
5270 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5271 int_x86_avx2_pmovsxdq>, VEX, VEX_L;
5272 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5273 int_x86_avx2_pmovzxbw>, VEX, VEX_L;
5274 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5275 int_x86_avx2_pmovzxwd>, VEX, VEX_L;
5276 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5277 int_x86_avx2_pmovzxdq>, VEX, VEX_L;
5280 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5281 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5282 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5283 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5284 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5285 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5287 let Predicates = [HasAVX] in {
5288 // Common patterns involving scalar load.
5289 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5290 (VPMOVSXBWrm addr:$src)>;
5291 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5292 (VPMOVSXBWrm addr:$src)>;
5293 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5294 (VPMOVSXBWrm addr:$src)>;
5296 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5297 (VPMOVSXWDrm addr:$src)>;
5298 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5299 (VPMOVSXWDrm addr:$src)>;
5300 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5301 (VPMOVSXWDrm addr:$src)>;
5303 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5304 (VPMOVSXDQrm addr:$src)>;
5305 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5306 (VPMOVSXDQrm addr:$src)>;
5307 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5308 (VPMOVSXDQrm addr:$src)>;
5310 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5311 (VPMOVZXBWrm addr:$src)>;
5312 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5313 (VPMOVZXBWrm addr:$src)>;
5314 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5315 (VPMOVZXBWrm addr:$src)>;
5317 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5318 (VPMOVZXWDrm addr:$src)>;
5319 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5320 (VPMOVZXWDrm addr:$src)>;
5321 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5322 (VPMOVZXWDrm addr:$src)>;
5324 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5325 (VPMOVZXDQrm addr:$src)>;
5326 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5327 (VPMOVZXDQrm addr:$src)>;
5328 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5329 (VPMOVZXDQrm addr:$src)>;
5332 let Predicates = [UseSSE41] in {
5333 // Common patterns involving scalar load.
5334 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5335 (PMOVSXBWrm addr:$src)>;
5336 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5337 (PMOVSXBWrm addr:$src)>;
5338 def : Pat<(int_x86_sse41_pmovsxbw (bc_v16i8 (loadv2i64 addr:$src))),
5339 (PMOVSXBWrm addr:$src)>;
5341 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5342 (PMOVSXWDrm addr:$src)>;
5343 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5344 (PMOVSXWDrm addr:$src)>;
5345 def : Pat<(int_x86_sse41_pmovsxwd (bc_v8i16 (loadv2i64 addr:$src))),
5346 (PMOVSXWDrm addr:$src)>;
5348 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5349 (PMOVSXDQrm addr:$src)>;
5350 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5351 (PMOVSXDQrm addr:$src)>;
5352 def : Pat<(int_x86_sse41_pmovsxdq (bc_v4i32 (loadv2i64 addr:$src))),
5353 (PMOVSXDQrm addr:$src)>;
5355 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5356 (PMOVZXBWrm addr:$src)>;
5357 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5358 (PMOVZXBWrm addr:$src)>;
5359 def : Pat<(int_x86_sse41_pmovzxbw (bc_v16i8 (loadv2i64 addr:$src))),
5360 (PMOVZXBWrm addr:$src)>;
5362 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5363 (PMOVZXWDrm addr:$src)>;
5364 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5365 (PMOVZXWDrm addr:$src)>;
5366 def : Pat<(int_x86_sse41_pmovzxwd (bc_v8i16 (loadv2i64 addr:$src))),
5367 (PMOVZXWDrm addr:$src)>;
5369 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5370 (PMOVZXDQrm addr:$src)>;
5371 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5372 (PMOVZXDQrm addr:$src)>;
5373 def : Pat<(int_x86_sse41_pmovzxdq (bc_v4i32 (loadv2i64 addr:$src))),
5374 (PMOVZXDQrm addr:$src)>;
5377 let Predicates = [HasAVX2] in {
5378 let AddedComplexity = 15 in {
5379 def : Pat<(v4i64 (X86vzmovly (v4i32 VR128:$src))),
5380 (VPMOVZXDQYrr VR128:$src)>;
5381 def : Pat<(v8i32 (X86vzmovly (v8i16 VR128:$src))),
5382 (VPMOVZXWDYrr VR128:$src)>;
5385 def : Pat<(v4i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQYrr VR128:$src)>;
5386 def : Pat<(v8i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDYrr VR128:$src)>;
5389 let Predicates = [HasAVX] in {
5390 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (VPMOVSXDQrr VR128:$src)>;
5391 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (VPMOVSXWDrr VR128:$src)>;
5394 let Predicates = [UseSSE41] in {
5395 def : Pat<(v2i64 (X86vsmovl (v4i32 VR128:$src))), (PMOVSXDQrr VR128:$src)>;
5396 def : Pat<(v4i32 (X86vsmovl (v8i16 VR128:$src))), (PMOVSXWDrr VR128:$src)>;
5400 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5401 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5402 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5403 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5405 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5406 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5408 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5412 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5414 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5415 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5416 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5418 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5419 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5421 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5425 let Predicates = [HasAVX] in {
5426 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5428 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5430 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5432 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5436 let Predicates = [HasAVX2] in {
5437 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5438 int_x86_avx2_pmovsxbd>, VEX, VEX_L;
5439 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5440 int_x86_avx2_pmovsxwq>, VEX, VEX_L;
5441 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5442 int_x86_avx2_pmovzxbd>, VEX, VEX_L;
5443 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5444 int_x86_avx2_pmovzxwq>, VEX, VEX_L;
5447 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5448 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5449 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5450 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5452 let Predicates = [HasAVX] in {
5453 // Common patterns involving scalar load
5454 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5455 (VPMOVSXBDrm addr:$src)>;
5456 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5457 (VPMOVSXWQrm addr:$src)>;
5459 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5460 (VPMOVZXBDrm addr:$src)>;
5461 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5462 (VPMOVZXWQrm addr:$src)>;
5465 let Predicates = [UseSSE41] in {
5466 // Common patterns involving scalar load
5467 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5468 (PMOVSXBDrm addr:$src)>;
5469 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5470 (PMOVSXWQrm addr:$src)>;
5472 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5473 (PMOVZXBDrm addr:$src)>;
5474 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5475 (PMOVZXWQrm addr:$src)>;
5478 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5479 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5480 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5481 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5483 // Expecting a i16 load any extended to i32 value.
5484 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5485 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5486 [(set VR128:$dst, (IntId (bitconvert
5487 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5491 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5493 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5495 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5497 // Expecting a i16 load any extended to i32 value.
5498 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5499 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5500 [(set VR256:$dst, (IntId (bitconvert
5501 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5505 let Predicates = [HasAVX] in {
5506 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5508 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5511 let Predicates = [HasAVX2] in {
5512 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5513 int_x86_avx2_pmovsxbq>, VEX, VEX_L;
5514 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5515 int_x86_avx2_pmovzxbq>, VEX, VEX_L;
5517 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5518 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5520 let Predicates = [HasAVX2] in {
5521 def : Pat<(v8i32 (X86vsmovl (v8i16 (bitconvert (v2i64 (load addr:$src)))))),
5522 (VPMOVSXWDYrm addr:$src)>;
5523 def : Pat<(v4i64 (X86vsmovl (v4i32 (bitconvert (v2i64 (load addr:$src)))))),
5524 (VPMOVSXDQYrm addr:$src)>;
5526 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2i64
5527 (scalar_to_vector (loadi64 addr:$src))))))),
5528 (VPMOVSXBDYrm addr:$src)>;
5529 def : Pat<(v8i32 (X86vsext (v16i8 (bitconvert (v2f64
5530 (scalar_to_vector (loadf64 addr:$src))))))),
5531 (VPMOVSXBDYrm addr:$src)>;
5533 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2i64
5534 (scalar_to_vector (loadi64 addr:$src))))))),
5535 (VPMOVSXWQYrm addr:$src)>;
5536 def : Pat<(v4i64 (X86vsext (v8i16 (bitconvert (v2f64
5537 (scalar_to_vector (loadf64 addr:$src))))))),
5538 (VPMOVSXWQYrm addr:$src)>;
5540 def : Pat<(v4i64 (X86vsext (v16i8 (bitconvert (v4i32
5541 (scalar_to_vector (loadi32 addr:$src))))))),
5542 (VPMOVSXBQYrm addr:$src)>;
5545 let Predicates = [HasAVX] in {
5546 // Common patterns involving scalar load
5547 def : Pat<(int_x86_sse41_pmovsxbq
5548 (bitconvert (v4i32 (X86vzmovl
5549 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5550 (VPMOVSXBQrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovzxbq
5553 (bitconvert (v4i32 (X86vzmovl
5554 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5555 (VPMOVZXBQrm addr:$src)>;
5558 let Predicates = [UseSSE41] in {
5559 // Common patterns involving scalar load
5560 def : Pat<(int_x86_sse41_pmovsxbq
5561 (bitconvert (v4i32 (X86vzmovl
5562 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5563 (PMOVSXBQrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovzxbq
5566 (bitconvert (v4i32 (X86vzmovl
5567 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5568 (PMOVZXBQrm addr:$src)>;
5570 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5571 (scalar_to_vector (loadi64 addr:$src))))))),
5572 (PMOVSXWDrm addr:$src)>;
5573 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5574 (scalar_to_vector (loadf64 addr:$src))))))),
5575 (PMOVSXWDrm addr:$src)>;
5576 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5577 (scalar_to_vector (loadi32 addr:$src))))))),
5578 (PMOVSXBDrm addr:$src)>;
5579 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5580 (scalar_to_vector (loadi32 addr:$src))))))),
5581 (PMOVSXWQrm addr:$src)>;
5582 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5583 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5584 (PMOVSXBQrm addr:$src)>;
5585 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5586 (scalar_to_vector (loadi64 addr:$src))))))),
5587 (PMOVSXDQrm addr:$src)>;
5588 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5589 (scalar_to_vector (loadf64 addr:$src))))))),
5590 (PMOVSXDQrm addr:$src)>;
5591 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5592 (scalar_to_vector (loadi64 addr:$src))))))),
5593 (PMOVSXBWrm addr:$src)>;
5594 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5595 (scalar_to_vector (loadf64 addr:$src))))))),
5596 (PMOVSXBWrm addr:$src)>;
5599 let Predicates = [HasAVX2] in {
5600 def : Pat<(v16i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWYrr VR128:$src)>;
5601 def : Pat<(v8i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDYrr VR128:$src)>;
5602 def : Pat<(v4i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQYrr VR128:$src)>;
5604 def : Pat<(v8i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDYrr VR128:$src)>;
5605 def : Pat<(v4i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQYrr VR128:$src)>;
5607 def : Pat<(v4i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQYrr VR128:$src)>;
5609 def : Pat<(v16i16 (X86vzext (v32i8 VR256:$src))),
5610 (VPMOVZXBWYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5611 def : Pat<(v8i32 (X86vzext (v32i8 VR256:$src))),
5612 (VPMOVZXBDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5613 def : Pat<(v4i64 (X86vzext (v32i8 VR256:$src))),
5614 (VPMOVZXBQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5616 def : Pat<(v8i32 (X86vzext (v16i16 VR256:$src))),
5617 (VPMOVZXWDYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5618 def : Pat<(v4i64 (X86vzext (v16i16 VR256:$src))),
5619 (VPMOVZXWQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5621 def : Pat<(v4i64 (X86vzext (v8i32 VR256:$src))),
5622 (VPMOVZXDQYrr (EXTRACT_SUBREG VR256:$src, sub_xmm))>;
5625 let Predicates = [HasAVX] in {
5626 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBWrr VR128:$src)>;
5627 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBDrr VR128:$src)>;
5628 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (VPMOVZXBQrr VR128:$src)>;
5630 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWDrr VR128:$src)>;
5631 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (VPMOVZXWQrr VR128:$src)>;
5633 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (VPMOVZXDQrr VR128:$src)>;
5635 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5636 (VPMOVZXBWrm addr:$src)>;
5637 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5638 (VPMOVZXBWrm addr:$src)>;
5639 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5640 (VPMOVZXBDrm addr:$src)>;
5641 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5642 (VPMOVZXBQrm addr:$src)>;
5644 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5645 (VPMOVZXWDrm addr:$src)>;
5646 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5647 (VPMOVZXWDrm addr:$src)>;
5648 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5649 (VPMOVZXWQrm addr:$src)>;
5651 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5652 (VPMOVZXDQrm addr:$src)>;
5653 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5654 (VPMOVZXDQrm addr:$src)>;
5655 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5656 (VPMOVZXDQrm addr:$src)>;
5658 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2i64
5659 (scalar_to_vector (loadi64 addr:$src))))))),
5660 (VPMOVSXWDrm addr:$src)>;
5661 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2i64
5662 (scalar_to_vector (loadi64 addr:$src))))))),
5663 (VPMOVSXDQrm addr:$src)>;
5664 def : Pat<(v4i32 (X86vsext (v8i16 (bitconvert (v2f64
5665 (scalar_to_vector (loadf64 addr:$src))))))),
5666 (VPMOVSXWDrm addr:$src)>;
5667 def : Pat<(v2i64 (X86vsext (v4i32 (bitconvert (v2f64
5668 (scalar_to_vector (loadf64 addr:$src))))))),
5669 (VPMOVSXDQrm addr:$src)>;
5670 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2i64
5671 (scalar_to_vector (loadi64 addr:$src))))))),
5672 (VPMOVSXBWrm addr:$src)>;
5673 def : Pat<(v8i16 (X86vsext (v16i8 (bitconvert (v2f64
5674 (scalar_to_vector (loadf64 addr:$src))))))),
5675 (VPMOVSXBWrm addr:$src)>;
5677 def : Pat<(v4i32 (X86vsext (v16i8 (bitconvert (v4i32
5678 (scalar_to_vector (loadi32 addr:$src))))))),
5679 (VPMOVSXBDrm addr:$src)>;
5680 def : Pat<(v2i64 (X86vsext (v8i16 (bitconvert (v4i32
5681 (scalar_to_vector (loadi32 addr:$src))))))),
5682 (VPMOVSXWQrm addr:$src)>;
5683 def : Pat<(v2i64 (X86vsext (v16i8 (bitconvert (v4i32
5684 (scalar_to_vector (extloadi32i16 addr:$src))))))),
5685 (VPMOVSXBQrm addr:$src)>;
5688 let Predicates = [UseSSE41] in {
5689 def : Pat<(v8i16 (X86vzext (v16i8 VR128:$src))), (PMOVZXBWrr VR128:$src)>;
5690 def : Pat<(v4i32 (X86vzext (v16i8 VR128:$src))), (PMOVZXBDrr VR128:$src)>;
5691 def : Pat<(v2i64 (X86vzext (v16i8 VR128:$src))), (PMOVZXBQrr VR128:$src)>;
5693 def : Pat<(v4i32 (X86vzext (v8i16 VR128:$src))), (PMOVZXWDrr VR128:$src)>;
5694 def : Pat<(v2i64 (X86vzext (v8i16 VR128:$src))), (PMOVZXWQrr VR128:$src)>;
5696 def : Pat<(v2i64 (X86vzext (v4i32 VR128:$src))), (PMOVZXDQrr VR128:$src)>;
5698 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5699 (PMOVZXBWrm addr:$src)>;
5700 def : Pat<(v8i16 (X86vzext (v16i8 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5701 (PMOVZXBWrm addr:$src)>;
5702 def : Pat<(v4i32 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5703 (PMOVZXBDrm addr:$src)>;
5704 def : Pat<(v2i64 (X86vzext (v16i8 (bitconvert (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))),
5705 (PMOVZXBQrm addr:$src)>;
5707 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5708 (PMOVZXWDrm addr:$src)>;
5709 def : Pat<(v4i32 (X86vzext (v8i16 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5710 (PMOVZXWDrm addr:$src)>;
5711 def : Pat<(v2i64 (X86vzext (v8i16 (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5712 (PMOVZXWQrm addr:$src)>;
5714 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))),
5715 (PMOVZXDQrm addr:$src)>;
5716 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2f64 (scalar_to_vector (loadf64 addr:$src))))))),
5717 (PMOVZXDQrm addr:$src)>;
5718 def : Pat<(v2i64 (X86vzext (v4i32 (bitconvert (v2i64 (X86vzload addr:$src)))))),
5719 (PMOVZXDQrm addr:$src)>;
5722 //===----------------------------------------------------------------------===//
5723 // SSE4.1 - Extract Instructions
5724 //===----------------------------------------------------------------------===//
5726 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5727 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5728 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5729 (ins VR128:$src1, i32i8imm:$src2),
5730 !strconcat(OpcodeStr,
5731 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5732 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5734 let neverHasSideEffects = 1, mayStore = 1 in
5735 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5736 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5737 !strconcat(OpcodeStr,
5738 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5741 // There's an AssertZext in the way of writing the store pattern
5742 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5745 let Predicates = [HasAVX] in {
5746 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5747 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5748 (ins VR128:$src1, i32i8imm:$src2),
5749 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5752 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5755 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5756 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5757 let neverHasSideEffects = 1, mayStore = 1 in
5758 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5759 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5760 !strconcat(OpcodeStr,
5761 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5764 // There's an AssertZext in the way of writing the store pattern
5765 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5768 let Predicates = [HasAVX] in
5769 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5771 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5774 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5775 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5776 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5777 (ins VR128:$src1, i32i8imm:$src2),
5778 !strconcat(OpcodeStr,
5779 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5781 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5782 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5783 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5784 !strconcat(OpcodeStr,
5785 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5786 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5787 addr:$dst)]>, OpSize;
5790 let Predicates = [HasAVX] in
5791 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5793 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5795 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5796 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5797 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5798 (ins VR128:$src1, i32i8imm:$src2),
5799 !strconcat(OpcodeStr,
5800 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5802 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5803 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5804 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5805 !strconcat(OpcodeStr,
5806 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5807 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5808 addr:$dst)]>, OpSize, REX_W;
5811 let Predicates = [HasAVX] in
5812 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5814 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5816 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5818 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5819 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5820 (ins VR128:$src1, i32i8imm:$src2),
5821 !strconcat(OpcodeStr,
5822 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5824 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5826 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5827 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5828 !strconcat(OpcodeStr,
5829 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5830 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5831 addr:$dst)]>, OpSize;
5834 let ExeDomain = SSEPackedSingle in {
5835 let Predicates = [HasAVX] in {
5836 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5837 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5838 (ins VR128:$src1, i32i8imm:$src2),
5839 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5842 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5845 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5846 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5849 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5851 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5854 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5855 Requires<[UseSSE41]>;
5857 //===----------------------------------------------------------------------===//
5858 // SSE4.1 - Insert Instructions
5859 //===----------------------------------------------------------------------===//
5861 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5862 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5863 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5865 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5867 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5869 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5870 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5871 (ins VR128:$src1, i8mem:$src2, i32i8imm:$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, (extloadi8 addr:$src2),
5878 imm:$src3))]>, OpSize;
5881 let Predicates = [HasAVX] in
5882 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5883 let Constraints = "$src1 = $dst" in
5884 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5886 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5887 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5888 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5890 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5892 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5894 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5896 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5897 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5899 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5901 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5903 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5904 imm:$src3)))]>, OpSize;
5907 let Predicates = [HasAVX] in
5908 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5909 let Constraints = "$src1 = $dst" in
5910 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5912 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5913 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5914 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5916 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5918 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5920 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5922 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5923 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5925 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5927 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5929 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5930 imm:$src3)))]>, OpSize;
5933 let Predicates = [HasAVX] in
5934 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5935 let Constraints = "$src1 = $dst" in
5936 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5938 // insertps has a few different modes, there's the first two here below which
5939 // are optimized inserts that won't zero arbitrary elements in the destination
5940 // vector. The next one matches the intrinsic and could zero arbitrary elements
5941 // in the target vector.
5942 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5943 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5944 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5946 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5948 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5950 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5952 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5953 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5955 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5957 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5959 (X86insrtps VR128:$src1,
5960 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5961 imm:$src3))]>, OpSize;
5964 let ExeDomain = SSEPackedSingle in {
5965 let Predicates = [HasAVX] in
5966 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5967 let Constraints = "$src1 = $dst" in
5968 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5971 //===----------------------------------------------------------------------===//
5972 // SSE4.1 - Round Instructions
5973 //===----------------------------------------------------------------------===//
5975 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5976 X86MemOperand x86memop, RegisterClass RC,
5977 PatFrag mem_frag32, PatFrag mem_frag64,
5978 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5979 let ExeDomain = SSEPackedSingle in {
5980 // Intrinsic operation, reg.
5981 // Vector intrinsic operation, reg
5982 def PSr : SS4AIi8<opcps, MRMSrcReg,
5983 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5984 !strconcat(OpcodeStr,
5985 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5986 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5989 // Vector intrinsic operation, mem
5990 def PSm : SS4AIi8<opcps, MRMSrcMem,
5991 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5992 !strconcat(OpcodeStr,
5993 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5995 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5997 } // ExeDomain = SSEPackedSingle
5999 let ExeDomain = SSEPackedDouble in {
6000 // Vector intrinsic operation, reg
6001 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6002 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6003 !strconcat(OpcodeStr,
6004 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6005 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6008 // Vector intrinsic operation, mem
6009 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6010 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6011 !strconcat(OpcodeStr,
6012 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6014 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6016 } // ExeDomain = SSEPackedDouble
6019 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6022 Intrinsic F64Int, bit Is2Addr = 1> {
6023 let ExeDomain = GenericDomain in {
6025 let hasSideEffects = 0 in
6026 def SSr : SS4AIi8<opcss, MRMSrcReg,
6027 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6029 !strconcat(OpcodeStr,
6030 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6031 !strconcat(OpcodeStr,
6032 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6035 // Intrinsic operation, reg.
6036 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6037 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6039 !strconcat(OpcodeStr,
6040 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6041 !strconcat(OpcodeStr,
6042 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6043 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6046 // Intrinsic operation, mem.
6047 def SSm : SS4AIi8<opcss, MRMSrcMem,
6048 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6050 !strconcat(OpcodeStr,
6051 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6052 !strconcat(OpcodeStr,
6053 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6055 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6059 let hasSideEffects = 0 in
6060 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6061 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6063 !strconcat(OpcodeStr,
6064 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6065 !strconcat(OpcodeStr,
6066 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6069 // Intrinsic operation, reg.
6070 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6071 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6073 !strconcat(OpcodeStr,
6074 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6075 !strconcat(OpcodeStr,
6076 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6077 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6080 // Intrinsic operation, mem.
6081 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6082 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6084 !strconcat(OpcodeStr,
6085 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6086 !strconcat(OpcodeStr,
6087 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6089 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6091 } // ExeDomain = GenericDomain
6094 // FP round - roundss, roundps, roundsd, roundpd
6095 let Predicates = [HasAVX] in {
6097 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6098 memopv4f32, memopv2f64,
6099 int_x86_sse41_round_ps,
6100 int_x86_sse41_round_pd>, VEX;
6101 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6102 memopv8f32, memopv4f64,
6103 int_x86_avx_round_ps_256,
6104 int_x86_avx_round_pd_256>, VEX, VEX_L;
6105 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6106 int_x86_sse41_round_ss,
6107 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6109 def : Pat<(ffloor FR32:$src),
6110 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6111 def : Pat<(f64 (ffloor FR64:$src)),
6112 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6113 def : Pat<(f32 (fnearbyint FR32:$src)),
6114 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6115 def : Pat<(f64 (fnearbyint FR64:$src)),
6116 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6117 def : Pat<(f32 (fceil FR32:$src)),
6118 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6119 def : Pat<(f64 (fceil FR64:$src)),
6120 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6121 def : Pat<(f32 (frint FR32:$src)),
6122 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6123 def : Pat<(f64 (frint FR64:$src)),
6124 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6125 def : Pat<(f32 (ftrunc FR32:$src)),
6126 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6127 def : Pat<(f64 (ftrunc FR64:$src)),
6128 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6130 def : Pat<(v4f32 (ffloor VR128:$src)),
6131 (VROUNDPSr VR128:$src, (i32 0x1))>;
6132 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6133 (VROUNDPSr VR128:$src, (i32 0xC))>;
6134 def : Pat<(v4f32 (fceil VR128:$src)),
6135 (VROUNDPSr VR128:$src, (i32 0x2))>;
6136 def : Pat<(v4f32 (frint VR128:$src)),
6137 (VROUNDPSr VR128:$src, (i32 0x4))>;
6138 def : Pat<(v4f32 (ftrunc VR128:$src)),
6139 (VROUNDPSr VR128:$src, (i32 0x3))>;
6141 def : Pat<(v2f64 (ffloor VR128:$src)),
6142 (VROUNDPDr VR128:$src, (i32 0x1))>;
6143 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6144 (VROUNDPDr VR128:$src, (i32 0xC))>;
6145 def : Pat<(v2f64 (fceil VR128:$src)),
6146 (VROUNDPDr VR128:$src, (i32 0x2))>;
6147 def : Pat<(v2f64 (frint VR128:$src)),
6148 (VROUNDPDr VR128:$src, (i32 0x4))>;
6149 def : Pat<(v2f64 (ftrunc VR128:$src)),
6150 (VROUNDPDr VR128:$src, (i32 0x3))>;
6152 def : Pat<(v8f32 (ffloor VR256:$src)),
6153 (VROUNDYPSr VR256:$src, (i32 0x1))>;
6154 def : Pat<(v8f32 (fnearbyint VR256:$src)),
6155 (VROUNDYPSr VR256:$src, (i32 0xC))>;
6156 def : Pat<(v8f32 (fceil VR256:$src)),
6157 (VROUNDYPSr VR256:$src, (i32 0x2))>;
6158 def : Pat<(v8f32 (frint VR256:$src)),
6159 (VROUNDYPSr VR256:$src, (i32 0x4))>;
6160 def : Pat<(v8f32 (ftrunc VR256:$src)),
6161 (VROUNDYPSr VR256:$src, (i32 0x3))>;
6163 def : Pat<(v4f64 (ffloor VR256:$src)),
6164 (VROUNDYPDr VR256:$src, (i32 0x1))>;
6165 def : Pat<(v4f64 (fnearbyint VR256:$src)),
6166 (VROUNDYPDr VR256:$src, (i32 0xC))>;
6167 def : Pat<(v4f64 (fceil VR256:$src)),
6168 (VROUNDYPDr VR256:$src, (i32 0x2))>;
6169 def : Pat<(v4f64 (frint VR256:$src)),
6170 (VROUNDYPDr VR256:$src, (i32 0x4))>;
6171 def : Pat<(v4f64 (ftrunc VR256:$src)),
6172 (VROUNDYPDr VR256:$src, (i32 0x3))>;
6175 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6176 memopv4f32, memopv2f64,
6177 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6178 let Constraints = "$src1 = $dst" in
6179 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6180 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6182 let Predicates = [UseSSE41] in {
6183 def : Pat<(ffloor FR32:$src),
6184 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6185 def : Pat<(f64 (ffloor FR64:$src)),
6186 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6187 def : Pat<(f32 (fnearbyint FR32:$src)),
6188 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6189 def : Pat<(f64 (fnearbyint FR64:$src)),
6190 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6191 def : Pat<(f32 (fceil FR32:$src)),
6192 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6193 def : Pat<(f64 (fceil FR64:$src)),
6194 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6195 def : Pat<(f32 (frint FR32:$src)),
6196 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6197 def : Pat<(f64 (frint FR64:$src)),
6198 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6199 def : Pat<(f32 (ftrunc FR32:$src)),
6200 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6201 def : Pat<(f64 (ftrunc FR64:$src)),
6202 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6204 def : Pat<(v4f32 (ffloor VR128:$src)),
6205 (ROUNDPSr VR128:$src, (i32 0x1))>;
6206 def : Pat<(v4f32 (fnearbyint VR128:$src)),
6207 (ROUNDPSr VR128:$src, (i32 0xC))>;
6208 def : Pat<(v4f32 (fceil VR128:$src)),
6209 (ROUNDPSr VR128:$src, (i32 0x2))>;
6210 def : Pat<(v4f32 (frint VR128:$src)),
6211 (ROUNDPSr VR128:$src, (i32 0x4))>;
6212 def : Pat<(v4f32 (ftrunc VR128:$src)),
6213 (ROUNDPSr VR128:$src, (i32 0x3))>;
6215 def : Pat<(v2f64 (ffloor VR128:$src)),
6216 (ROUNDPDr VR128:$src, (i32 0x1))>;
6217 def : Pat<(v2f64 (fnearbyint VR128:$src)),
6218 (ROUNDPDr VR128:$src, (i32 0xC))>;
6219 def : Pat<(v2f64 (fceil VR128:$src)),
6220 (ROUNDPDr VR128:$src, (i32 0x2))>;
6221 def : Pat<(v2f64 (frint VR128:$src)),
6222 (ROUNDPDr VR128:$src, (i32 0x4))>;
6223 def : Pat<(v2f64 (ftrunc VR128:$src)),
6224 (ROUNDPDr VR128:$src, (i32 0x3))>;
6227 //===----------------------------------------------------------------------===//
6228 // SSE4.1 - Packed Bit Test
6229 //===----------------------------------------------------------------------===//
6231 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6232 // the intel intrinsic that corresponds to this.
6233 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6234 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6235 "vptest\t{$src2, $src1|$src1, $src2}",
6236 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6238 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6239 "vptest\t{$src2, $src1|$src1, $src2}",
6240 [(set EFLAGS,(X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6243 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6244 "vptest\t{$src2, $src1|$src1, $src2}",
6245 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6247 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6248 "vptest\t{$src2, $src1|$src1, $src2}",
6249 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6253 let Defs = [EFLAGS] in {
6254 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6255 "ptest\t{$src2, $src1|$src1, $src2}",
6256 [(set EFLAGS, (X86ptest VR128:$src1, (v2i64 VR128:$src2)))]>,
6258 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6259 "ptest\t{$src2, $src1|$src1, $src2}",
6260 [(set EFLAGS, (X86ptest VR128:$src1, (memopv2i64 addr:$src2)))]>,
6264 // The bit test instructions below are AVX only
6265 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6266 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6267 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6268 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6269 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6270 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6271 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6272 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6276 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6277 let ExeDomain = SSEPackedSingle in {
6278 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6279 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>,
6282 let ExeDomain = SSEPackedDouble in {
6283 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6284 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>,
6289 //===----------------------------------------------------------------------===//
6290 // SSE4.1 - Misc Instructions
6291 //===----------------------------------------------------------------------===//
6293 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6294 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6295 "popcnt{w}\t{$src, $dst|$dst, $src}",
6296 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6298 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6299 "popcnt{w}\t{$src, $dst|$dst, $src}",
6300 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6301 (implicit EFLAGS)]>, OpSize, XS;
6303 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6304 "popcnt{l}\t{$src, $dst|$dst, $src}",
6305 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6307 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6308 "popcnt{l}\t{$src, $dst|$dst, $src}",
6309 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6310 (implicit EFLAGS)]>, XS;
6312 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6313 "popcnt{q}\t{$src, $dst|$dst, $src}",
6314 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6316 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6317 "popcnt{q}\t{$src, $dst|$dst, $src}",
6318 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6319 (implicit EFLAGS)]>, XS;
6324 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6325 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6326 Intrinsic IntId128> {
6327 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6329 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6330 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6331 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6333 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6336 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6339 let Predicates = [HasAVX] in
6340 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6341 int_x86_sse41_phminposuw>, VEX;
6342 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6343 int_x86_sse41_phminposuw>;
6345 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6346 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6347 Intrinsic IntId128, bit Is2Addr = 1> {
6348 let isCommutable = 1 in
6349 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6350 (ins VR128:$src1, VR128:$src2),
6352 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6353 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6354 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6355 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6356 (ins VR128:$src1, i128mem:$src2),
6358 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6359 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6361 (IntId128 VR128:$src1,
6362 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6365 /// SS41I_binop_rm_int_y - Simple SSE 4.1 binary operator
6366 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6367 Intrinsic IntId256> {
6368 let isCommutable = 1 in
6369 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6370 (ins VR256:$src1, VR256:$src2),
6371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6372 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6373 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6374 (ins VR256:$src1, i256mem:$src2),
6375 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6377 (IntId256 VR256:$src1,
6378 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6382 /// SS48I_binop_rm - Simple SSE41 binary operator.
6383 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6384 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6385 X86MemOperand x86memop, bit Is2Addr = 1> {
6386 let isCommutable = 1 in
6387 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst),
6388 (ins RC:$src1, RC:$src2),
6390 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6391 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6392 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>, OpSize;
6393 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst),
6394 (ins RC:$src1, x86memop:$src2),
6396 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6399 (OpVT (OpNode RC:$src1,
6400 (bitconvert (memop_frag addr:$src2)))))]>, OpSize;
6403 let Predicates = [HasAVX] in {
6404 let isCommutable = 0 in
6405 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6407 defm VPMINSB : SS48I_binop_rm<0x38, "vpminsb", X86smin, v16i8, VR128,
6408 memopv2i64, i128mem, 0>, VEX_4V;
6409 defm VPMINSD : SS48I_binop_rm<0x39, "vpminsd", X86smin, v4i32, VR128,
6410 memopv2i64, i128mem, 0>, VEX_4V;
6411 defm VPMINUD : SS48I_binop_rm<0x3B, "vpminud", X86umin, v4i32, VR128,
6412 memopv2i64, i128mem, 0>, VEX_4V;
6413 defm VPMINUW : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v8i16, VR128,
6414 memopv2i64, i128mem, 0>, VEX_4V;
6415 defm VPMAXSB : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v16i8, VR128,
6416 memopv2i64, i128mem, 0>, VEX_4V;
6417 defm VPMAXSD : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v4i32, VR128,
6418 memopv2i64, i128mem, 0>, VEX_4V;
6419 defm VPMAXUD : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v4i32, VR128,
6420 memopv2i64, i128mem, 0>, VEX_4V;
6421 defm VPMAXUW : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v8i16, VR128,
6422 memopv2i64, i128mem, 0>, VEX_4V;
6423 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6427 let Predicates = [HasAVX2] in {
6428 let isCommutable = 0 in
6429 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6430 int_x86_avx2_packusdw>, VEX_4V, VEX_L;
6431 defm VPMINSBY : SS48I_binop_rm<0x38, "vpminsb", X86smin, v32i8, VR256,
6432 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6433 defm VPMINSDY : SS48I_binop_rm<0x39, "vpminsd", X86smin, v8i32, VR256,
6434 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6435 defm VPMINUDY : SS48I_binop_rm<0x3B, "vpminud", X86umin, v8i32, VR256,
6436 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6437 defm VPMINUWY : SS48I_binop_rm<0x3A, "vpminuw", X86umin, v16i16, VR256,
6438 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6439 defm VPMAXSBY : SS48I_binop_rm<0x3C, "vpmaxsb", X86smax, v32i8, VR256,
6440 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6441 defm VPMAXSDY : SS48I_binop_rm<0x3D, "vpmaxsd", X86smax, v8i32, VR256,
6442 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6443 defm VPMAXUDY : SS48I_binop_rm<0x3F, "vpmaxud", X86umax, v8i32, VR256,
6444 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6445 defm VPMAXUWY : SS48I_binop_rm<0x3E, "vpmaxuw", X86umax, v16i16, VR256,
6446 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6447 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6448 int_x86_avx2_pmul_dq>, VEX_4V, VEX_L;
6451 let Constraints = "$src1 = $dst" in {
6452 let isCommutable = 0 in
6453 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6454 defm PMINSB : SS48I_binop_rm<0x38, "pminsb", X86smin, v16i8, VR128,
6455 memopv2i64, i128mem>;
6456 defm PMINSD : SS48I_binop_rm<0x39, "pminsd", X86smin, v4i32, VR128,
6457 memopv2i64, i128mem>;
6458 defm PMINUD : SS48I_binop_rm<0x3B, "pminud", X86umin, v4i32, VR128,
6459 memopv2i64, i128mem>;
6460 defm PMINUW : SS48I_binop_rm<0x3A, "pminuw", X86umin, v8i16, VR128,
6461 memopv2i64, i128mem>;
6462 defm PMAXSB : SS48I_binop_rm<0x3C, "pmaxsb", X86smax, v16i8, VR128,
6463 memopv2i64, i128mem>;
6464 defm PMAXSD : SS48I_binop_rm<0x3D, "pmaxsd", X86smax, v4i32, VR128,
6465 memopv2i64, i128mem>;
6466 defm PMAXUD : SS48I_binop_rm<0x3F, "pmaxud", X86umax, v4i32, VR128,
6467 memopv2i64, i128mem>;
6468 defm PMAXUW : SS48I_binop_rm<0x3E, "pmaxuw", X86umax, v8i16, VR128,
6469 memopv2i64, i128mem>;
6470 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6473 let Predicates = [HasAVX] in {
6474 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, VR128,
6475 memopv2i64, i128mem, 0>, VEX_4V;
6476 defm VPCMPEQQ : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v2i64, VR128,
6477 memopv2i64, i128mem, 0>, VEX_4V;
6479 let Predicates = [HasAVX2] in {
6480 defm VPMULLDY : SS48I_binop_rm<0x40, "vpmulld", mul, v8i32, VR256,
6481 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6482 defm VPCMPEQQY : SS48I_binop_rm<0x29, "vpcmpeqq", X86pcmpeq, v4i64, VR256,
6483 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6486 let Constraints = "$src1 = $dst" in {
6487 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32, VR128,
6488 memopv2i64, i128mem>;
6489 defm PCMPEQQ : SS48I_binop_rm<0x29, "pcmpeqq", X86pcmpeq, v2i64, VR128,
6490 memopv2i64, i128mem>;
6493 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6494 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6495 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6496 X86MemOperand x86memop, bit Is2Addr = 1> {
6497 let isCommutable = 1 in
6498 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6499 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6501 !strconcat(OpcodeStr,
6502 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6503 !strconcat(OpcodeStr,
6504 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6505 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6507 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6508 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6510 !strconcat(OpcodeStr,
6511 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6512 !strconcat(OpcodeStr,
6513 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6516 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6520 let Predicates = [HasAVX] in {
6521 let isCommutable = 0 in {
6522 let ExeDomain = SSEPackedSingle in {
6523 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6524 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6525 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6526 int_x86_avx_blend_ps_256, VR256, memopv8f32,
6527 f256mem, 0>, VEX_4V, VEX_L;
6529 let ExeDomain = SSEPackedDouble in {
6530 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6531 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6532 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6533 int_x86_avx_blend_pd_256,VR256, memopv4f64,
6534 f256mem, 0>, VEX_4V, VEX_L;
6536 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6537 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6538 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6539 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6541 let ExeDomain = SSEPackedSingle in
6542 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6543 VR128, memopv4f32, f128mem, 0>, VEX_4V;
6544 let ExeDomain = SSEPackedDouble in
6545 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6546 VR128, memopv2f64, f128mem, 0>, VEX_4V;
6547 let ExeDomain = SSEPackedSingle in
6548 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6549 VR256, memopv8f32, i256mem, 0>, VEX_4V, VEX_L;
6552 let Predicates = [HasAVX2] in {
6553 let isCommutable = 0 in {
6554 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6555 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6556 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6557 VR256, memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6561 let Constraints = "$src1 = $dst" in {
6562 let isCommutable = 0 in {
6563 let ExeDomain = SSEPackedSingle in
6564 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6565 VR128, memopv4f32, f128mem>;
6566 let ExeDomain = SSEPackedDouble in
6567 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6568 VR128, memopv2f64, f128mem>;
6569 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6570 VR128, memopv2i64, i128mem>;
6571 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6572 VR128, memopv2i64, i128mem>;
6574 let ExeDomain = SSEPackedSingle in
6575 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6576 VR128, memopv4f32, f128mem>;
6577 let ExeDomain = SSEPackedDouble in
6578 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6579 VR128, memopv2f64, f128mem>;
6582 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6583 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6584 RegisterClass RC, X86MemOperand x86memop,
6585 PatFrag mem_frag, Intrinsic IntId> {
6586 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6587 (ins RC:$src1, RC:$src2, RC:$src3),
6588 !strconcat(OpcodeStr,
6589 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6590 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6591 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6593 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6594 (ins RC:$src1, x86memop:$src2, RC:$src3),
6595 !strconcat(OpcodeStr,
6596 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6598 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6600 IIC_DEFAULT, SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6603 let Predicates = [HasAVX] in {
6604 let ExeDomain = SSEPackedDouble in {
6605 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, f128mem,
6606 memopv2f64, int_x86_sse41_blendvpd>;
6607 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, f256mem,
6608 memopv4f64, int_x86_avx_blendv_pd_256>, VEX_L;
6609 } // ExeDomain = SSEPackedDouble
6610 let ExeDomain = SSEPackedSingle in {
6611 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, f128mem,
6612 memopv4f32, int_x86_sse41_blendvps>;
6613 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, f256mem,
6614 memopv8f32, int_x86_avx_blendv_ps_256>, VEX_L;
6615 } // ExeDomain = SSEPackedSingle
6616 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6617 memopv2i64, int_x86_sse41_pblendvb>;
6620 let Predicates = [HasAVX2] in {
6621 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6622 memopv4i64, int_x86_avx2_pblendvb>, VEX_L;
6625 let Predicates = [HasAVX] in {
6626 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6627 (v16i8 VR128:$src2))),
6628 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6629 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6630 (v4i32 VR128:$src2))),
6631 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6632 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6633 (v4f32 VR128:$src2))),
6634 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6635 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6636 (v2i64 VR128:$src2))),
6637 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6638 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6639 (v2f64 VR128:$src2))),
6640 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6641 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6642 (v8i32 VR256:$src2))),
6643 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6644 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6645 (v8f32 VR256:$src2))),
6646 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6647 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6648 (v4i64 VR256:$src2))),
6649 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6650 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6651 (v4f64 VR256:$src2))),
6652 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6654 def : Pat<(v8f32 (X86Blendi (v8f32 VR256:$src1), (v8f32 VR256:$src2),
6656 (VBLENDPSYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6657 def : Pat<(v4f64 (X86Blendi (v4f64 VR256:$src1), (v4f64 VR256:$src2),
6659 (VBLENDPDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6661 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6663 (VPBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6664 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6666 (VBLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6667 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6669 (VBLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6672 let Predicates = [HasAVX2] in {
6673 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6674 (v32i8 VR256:$src2))),
6675 (VPBLENDVBYrr VR256:$src1, VR256:$src2, VR256:$mask)>;
6676 def : Pat<(v16i16 (X86Blendi (v16i16 VR256:$src1), (v16i16 VR256:$src2),
6678 (VPBLENDWYrri VR256:$src1, VR256:$src2, imm:$mask)>;
6681 /// SS41I_ternary_int - SSE 4.1 ternary operator
6682 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6683 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6684 X86MemOperand x86memop, Intrinsic IntId> {
6685 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6686 (ins VR128:$src1, VR128:$src2),
6687 !strconcat(OpcodeStr,
6688 "\t{$src2, $dst|$dst, $src2}"),
6689 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6692 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6693 (ins VR128:$src1, x86memop:$src2),
6694 !strconcat(OpcodeStr,
6695 "\t{$src2, $dst|$dst, $src2}"),
6698 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6702 let ExeDomain = SSEPackedDouble in
6703 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64, f128mem,
6704 int_x86_sse41_blendvpd>;
6705 let ExeDomain = SSEPackedSingle in
6706 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32, f128mem,
6707 int_x86_sse41_blendvps>;
6708 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64, i128mem,
6709 int_x86_sse41_pblendvb>;
6711 // Aliases with the implicit xmm0 argument
6712 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6713 (BLENDVPDrr0 VR128:$dst, VR128:$src2)>;
6714 def : InstAlias<"blendvpd\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6715 (BLENDVPDrm0 VR128:$dst, f128mem:$src2)>;
6716 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6717 (BLENDVPSrr0 VR128:$dst, VR128:$src2)>;
6718 def : InstAlias<"blendvps\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6719 (BLENDVPSrm0 VR128:$dst, f128mem:$src2)>;
6720 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6721 (PBLENDVBrr0 VR128:$dst, VR128:$src2)>;
6722 def : InstAlias<"pblendvb\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}",
6723 (PBLENDVBrm0 VR128:$dst, i128mem:$src2)>;
6725 let Predicates = [UseSSE41] in {
6726 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6727 (v16i8 VR128:$src2))),
6728 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6729 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6730 (v4i32 VR128:$src2))),
6731 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6732 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6733 (v4f32 VR128:$src2))),
6734 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6735 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6736 (v2i64 VR128:$src2))),
6737 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6738 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6739 (v2f64 VR128:$src2))),
6740 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6742 def : Pat<(v8i16 (X86Blendi (v8i16 VR128:$src1), (v8i16 VR128:$src2),
6744 (PBLENDWrri VR128:$src1, VR128:$src2, imm:$mask)>;
6745 def : Pat<(v4f32 (X86Blendi (v4f32 VR128:$src1), (v4f32 VR128:$src2),
6747 (BLENDPSrri VR128:$src1, VR128:$src2, imm:$mask)>;
6748 def : Pat<(v2f64 (X86Blendi (v2f64 VR128:$src1), (v2f64 VR128:$src2),
6750 (BLENDPDrri VR128:$src1, VR128:$src2, imm:$mask)>;
6754 let Predicates = [HasAVX] in
6755 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6756 "vmovntdqa\t{$src, $dst|$dst, $src}",
6757 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6759 let Predicates = [HasAVX2] in
6760 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6761 "vmovntdqa\t{$src, $dst|$dst, $src}",
6762 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6764 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6765 "movntdqa\t{$src, $dst|$dst, $src}",
6766 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6769 //===----------------------------------------------------------------------===//
6770 // SSE4.2 - Compare Instructions
6771 //===----------------------------------------------------------------------===//
6773 /// SS42I_binop_rm - Simple SSE 4.2 binary operator
6774 multiclass SS42I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6775 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
6776 X86MemOperand x86memop, bit Is2Addr = 1> {
6777 def rr : SS428I<opc, MRMSrcReg, (outs RC:$dst),
6778 (ins RC:$src1, RC:$src2),
6780 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6781 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6782 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>,
6784 def rm : SS428I<opc, MRMSrcMem, (outs RC:$dst),
6785 (ins RC:$src1, x86memop:$src2),
6787 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6788 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6790 (OpVT (OpNode RC:$src1, (memop_frag addr:$src2))))]>, OpSize;
6793 let Predicates = [HasAVX] in
6794 defm VPCMPGTQ : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v2i64, VR128,
6795 memopv2i64, i128mem, 0>, VEX_4V;
6797 let Predicates = [HasAVX2] in
6798 defm VPCMPGTQY : SS42I_binop_rm<0x37, "vpcmpgtq", X86pcmpgt, v4i64, VR256,
6799 memopv4i64, i256mem, 0>, VEX_4V, VEX_L;
6801 let Constraints = "$src1 = $dst" in
6802 defm PCMPGTQ : SS42I_binop_rm<0x37, "pcmpgtq", X86pcmpgt, v2i64, VR128,
6803 memopv2i64, i128mem>;
6805 //===----------------------------------------------------------------------===//
6806 // SSE4.2 - String/text Processing Instructions
6807 //===----------------------------------------------------------------------===//
6809 // Packed Compare Implicit Length Strings, Return Mask
6810 multiclass pseudo_pcmpistrm<string asm> {
6811 def REG : PseudoI<(outs VR128:$dst),
6812 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6813 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6815 def MEM : PseudoI<(outs VR128:$dst),
6816 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6817 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1,
6818 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6821 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6822 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6823 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[UseSSE42]>;
6826 multiclass pcmpistrm_SS42AI<string asm> {
6827 def rr : SS42AI<0x62, MRMSrcReg, (outs),
6828 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6829 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6832 def rm :SS42AI<0x62, MRMSrcMem, (outs),
6833 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6834 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6838 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6839 let Predicates = [HasAVX] in
6840 defm VPCMPISTRM128 : pcmpistrm_SS42AI<"vpcmpistrm">, VEX;
6841 defm PCMPISTRM128 : pcmpistrm_SS42AI<"pcmpistrm"> ;
6844 // Packed Compare Explicit Length Strings, Return Mask
6845 multiclass pseudo_pcmpestrm<string asm> {
6846 def REG : PseudoI<(outs VR128:$dst),
6847 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6848 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6849 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6850 def MEM : PseudoI<(outs VR128:$dst),
6851 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6852 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128 VR128:$src1, EAX,
6853 (bc_v16i8 (memopv2i64 addr:$src3)), EDX, imm:$src5))]>;
6856 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6857 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6858 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[UseSSE42]>;
6861 multiclass SS42AI_pcmpestrm<string asm> {
6862 def rr : SS42AI<0x60, MRMSrcReg, (outs),
6863 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6864 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6867 def rm : SS42AI<0x60, MRMSrcMem, (outs),
6868 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6869 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6873 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6874 let Predicates = [HasAVX] in
6875 defm VPCMPESTRM128 : SS42AI_pcmpestrm<"vpcmpestrm">, VEX;
6876 defm PCMPESTRM128 : SS42AI_pcmpestrm<"pcmpestrm">;
6879 // Packed Compare Implicit Length Strings, Return Index
6880 multiclass pseudo_pcmpistri<string asm> {
6881 def REG : PseudoI<(outs GR32:$dst),
6882 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6883 [(set GR32:$dst, EFLAGS,
6884 (X86pcmpistri VR128:$src1, VR128:$src2, imm:$src3))]>;
6885 def MEM : PseudoI<(outs GR32:$dst),
6886 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6887 [(set GR32:$dst, EFLAGS, (X86pcmpistri VR128:$src1,
6888 (bc_v16i8 (memopv2i64 addr:$src2)), imm:$src3))]>;
6891 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6892 defm VPCMPISTRI : pseudo_pcmpistri<"#VPCMPISTRI">, Requires<[HasAVX]>;
6893 defm PCMPISTRI : pseudo_pcmpistri<"#PCMPISTRI">, Requires<[UseSSE42]>;
6896 multiclass SS42AI_pcmpistri<string asm> {
6897 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6898 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6899 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6902 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6903 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6904 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6908 let Defs = [ECX, EFLAGS], neverHasSideEffects = 1 in {
6909 let Predicates = [HasAVX] in
6910 defm VPCMPISTRI : SS42AI_pcmpistri<"vpcmpistri">, VEX;
6911 defm PCMPISTRI : SS42AI_pcmpistri<"pcmpistri">;
6914 // Packed Compare Explicit Length Strings, Return Index
6915 multiclass pseudo_pcmpestri<string asm> {
6916 def REG : PseudoI<(outs GR32:$dst),
6917 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6918 [(set GR32:$dst, EFLAGS,
6919 (X86pcmpestri VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6920 def MEM : PseudoI<(outs GR32:$dst),
6921 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6922 [(set GR32:$dst, EFLAGS,
6923 (X86pcmpestri VR128:$src1, EAX, (bc_v16i8 (memopv2i64 addr:$src3)), EDX,
6927 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6928 defm VPCMPESTRI : pseudo_pcmpestri<"#VPCMPESTRI">, Requires<[HasAVX]>;
6929 defm PCMPESTRI : pseudo_pcmpestri<"#PCMPESTRI">, Requires<[UseSSE42]>;
6932 multiclass SS42AI_pcmpestri<string asm> {
6933 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6934 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6935 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6938 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6939 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6940 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6944 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6945 let Predicates = [HasAVX] in
6946 defm VPCMPESTRI : SS42AI_pcmpestri<"vpcmpestri">, VEX;
6947 defm PCMPESTRI : SS42AI_pcmpestri<"pcmpestri">;
6950 //===----------------------------------------------------------------------===//
6951 // SSE4.2 - CRC Instructions
6952 //===----------------------------------------------------------------------===//
6954 // No CRC instructions have AVX equivalents
6956 // crc intrinsic instruction
6957 // This set of instructions are only rm, the only difference is the size
6959 let Constraints = "$src1 = $dst" in {
6960 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6961 (ins GR32:$src1, i8mem:$src2),
6962 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6964 (int_x86_sse42_crc32_32_8 GR32:$src1,
6965 (load addr:$src2)))]>;
6966 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6967 (ins GR32:$src1, GR8:$src2),
6968 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6970 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6971 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6972 (ins GR32:$src1, i16mem:$src2),
6973 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6975 (int_x86_sse42_crc32_32_16 GR32:$src1,
6976 (load addr:$src2)))]>,
6978 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6979 (ins GR32:$src1, GR16:$src2),
6980 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6982 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6984 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6985 (ins GR32:$src1, i32mem:$src2),
6986 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6988 (int_x86_sse42_crc32_32_32 GR32:$src1,
6989 (load addr:$src2)))]>;
6990 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6991 (ins GR32:$src1, GR32:$src2),
6992 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6994 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6995 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6996 (ins GR64:$src1, i8mem:$src2),
6997 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6999 (int_x86_sse42_crc32_64_8 GR64:$src1,
7000 (load addr:$src2)))]>,
7002 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7003 (ins GR64:$src1, GR8:$src2),
7004 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7006 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7008 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7009 (ins GR64:$src1, i64mem:$src2),
7010 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7012 (int_x86_sse42_crc32_64_64 GR64:$src1,
7013 (load addr:$src2)))]>,
7015 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7016 (ins GR64:$src1, GR64:$src2),
7017 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7019 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7023 //===----------------------------------------------------------------------===//
7024 // AES-NI Instructions
7025 //===----------------------------------------------------------------------===//
7027 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7028 Intrinsic IntId128, bit Is2Addr = 1> {
7029 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7030 (ins VR128:$src1, VR128:$src2),
7032 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7033 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7034 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7036 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7037 (ins VR128:$src1, i128mem:$src2),
7039 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7040 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7042 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7045 // Perform One Round of an AES Encryption/Decryption Flow
7046 let Predicates = [HasAVX, HasAES] in {
7047 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7048 int_x86_aesni_aesenc, 0>, VEX_4V;
7049 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7050 int_x86_aesni_aesenclast, 0>, VEX_4V;
7051 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7052 int_x86_aesni_aesdec, 0>, VEX_4V;
7053 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7054 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7057 let Constraints = "$src1 = $dst" in {
7058 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7059 int_x86_aesni_aesenc>;
7060 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7061 int_x86_aesni_aesenclast>;
7062 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7063 int_x86_aesni_aesdec>;
7064 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7065 int_x86_aesni_aesdeclast>;
7068 // Perform the AES InvMixColumn Transformation
7069 let Predicates = [HasAVX, HasAES] in {
7070 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7072 "vaesimc\t{$src1, $dst|$dst, $src1}",
7074 (int_x86_aesni_aesimc VR128:$src1))]>,
7076 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7077 (ins i128mem:$src1),
7078 "vaesimc\t{$src1, $dst|$dst, $src1}",
7079 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7082 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7084 "aesimc\t{$src1, $dst|$dst, $src1}",
7086 (int_x86_aesni_aesimc VR128:$src1))]>,
7088 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7089 (ins i128mem:$src1),
7090 "aesimc\t{$src1, $dst|$dst, $src1}",
7091 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7094 // AES Round Key Generation Assist
7095 let Predicates = [HasAVX, HasAES] in {
7096 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7097 (ins VR128:$src1, i8imm:$src2),
7098 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7100 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7102 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7103 (ins i128mem:$src1, i8imm:$src2),
7104 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7106 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7109 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7110 (ins VR128:$src1, i8imm:$src2),
7111 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7113 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7115 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7116 (ins i128mem:$src1, i8imm:$src2),
7117 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7119 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7122 //===----------------------------------------------------------------------===//
7123 // PCLMUL Instructions
7124 //===----------------------------------------------------------------------===//
7126 // AVX carry-less Multiplication instructions
7127 def VPCLMULQDQrr : AVXPCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7128 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7129 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7131 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7133 def VPCLMULQDQrm : AVXPCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7134 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7135 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7136 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7137 (memopv2i64 addr:$src2), imm:$src3))]>;
7139 // Carry-less Multiplication instructions
7140 let Constraints = "$src1 = $dst" in {
7141 def PCLMULQDQrr : PCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7142 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7143 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7145 (int_x86_pclmulqdq VR128:$src1, VR128:$src2, imm:$src3))]>;
7147 def PCLMULQDQrm : PCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7148 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7149 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7150 [(set VR128:$dst, (int_x86_pclmulqdq VR128:$src1,
7151 (memopv2i64 addr:$src2), imm:$src3))]>;
7152 } // Constraints = "$src1 = $dst"
7155 multiclass pclmul_alias<string asm, int immop> {
7156 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7157 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7159 def : InstAlias<!strconcat("pclmul", asm, "dq {$src, $dst|$dst, $src}"),
7160 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7162 def : InstAlias<!strconcat("vpclmul", asm,
7163 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7164 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7166 def : InstAlias<!strconcat("vpclmul", asm,
7167 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7168 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7170 defm : pclmul_alias<"hqhq", 0x11>;
7171 defm : pclmul_alias<"hqlq", 0x01>;
7172 defm : pclmul_alias<"lqhq", 0x10>;
7173 defm : pclmul_alias<"lqlq", 0x00>;
7175 //===----------------------------------------------------------------------===//
7176 // SSE4A Instructions
7177 //===----------------------------------------------------------------------===//
7179 let Predicates = [HasSSE4A] in {
7181 let Constraints = "$src = $dst" in {
7182 def EXTRQI : Ii8<0x78, MRM0r, (outs VR128:$dst),
7183 (ins VR128:$src, i8imm:$len, i8imm:$idx),
7184 "extrq\t{$idx, $len, $src|$src, $len, $idx}",
7185 [(set VR128:$dst, (int_x86_sse4a_extrqi VR128:$src, imm:$len,
7186 imm:$idx))]>, TB, OpSize;
7187 def EXTRQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7188 (ins VR128:$src, VR128:$mask),
7189 "extrq\t{$mask, $src|$src, $mask}",
7190 [(set VR128:$dst, (int_x86_sse4a_extrq VR128:$src,
7191 VR128:$mask))]>, TB, OpSize;
7193 def INSERTQI : Ii8<0x78, MRMSrcReg, (outs VR128:$dst),
7194 (ins VR128:$src, VR128:$src2, i8imm:$len, i8imm:$idx),
7195 "insertq\t{$idx, $len, $src2, $src|$src, $src2, $len, $idx}",
7196 [(set VR128:$dst, (int_x86_sse4a_insertqi VR128:$src,
7197 VR128:$src2, imm:$len, imm:$idx))]>, XD;
7198 def INSERTQ : I<0x79, MRMSrcReg, (outs VR128:$dst),
7199 (ins VR128:$src, VR128:$mask),
7200 "insertq\t{$mask, $src|$src, $mask}",
7201 [(set VR128:$dst, (int_x86_sse4a_insertq VR128:$src,
7202 VR128:$mask))]>, XD;
7205 def MOVNTSS : I<0x2B, MRMDestMem, (outs), (ins f32mem:$dst, VR128:$src),
7206 "movntss\t{$src, $dst|$dst, $src}",
7207 [(int_x86_sse4a_movnt_ss addr:$dst, VR128:$src)]>, XS;
7209 def MOVNTSD : I<0x2B, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
7210 "movntsd\t{$src, $dst|$dst, $src}",
7211 [(int_x86_sse4a_movnt_sd addr:$dst, VR128:$src)]>, XD;
7214 //===----------------------------------------------------------------------===//
7216 //===----------------------------------------------------------------------===//
7218 //===----------------------------------------------------------------------===//
7219 // VBROADCAST - Load from memory and broadcast to all elements of the
7220 // destination operand
7222 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7223 X86MemOperand x86memop, Intrinsic Int> :
7224 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7225 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7226 [(set RC:$dst, (Int addr:$src))]>, VEX;
7228 // AVX2 adds register forms
7229 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7231 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7232 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7233 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7235 let ExeDomain = SSEPackedSingle in {
7236 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7237 int_x86_avx_vbroadcast_ss>;
7238 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7239 int_x86_avx_vbroadcast_ss_256>, VEX_L;
7241 let ExeDomain = SSEPackedDouble in
7242 def VBROADCASTSDYrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7243 int_x86_avx_vbroadcast_sd_256>, VEX_L;
7244 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7245 int_x86_avx_vbroadcastf128_pd_256>, VEX_L;
7247 let ExeDomain = SSEPackedSingle in {
7248 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7249 int_x86_avx2_vbroadcast_ss_ps>;
7250 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7251 int_x86_avx2_vbroadcast_ss_ps_256>, VEX_L;
7253 let ExeDomain = SSEPackedDouble in
7254 def VBROADCASTSDYrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7255 int_x86_avx2_vbroadcast_sd_pd_256>, VEX_L;
7257 let Predicates = [HasAVX2] in
7258 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7259 int_x86_avx2_vbroadcasti128>, VEX_L;
7261 let Predicates = [HasAVX] in
7262 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7263 (VBROADCASTF128 addr:$src)>;
7266 //===----------------------------------------------------------------------===//
7267 // VINSERTF128 - Insert packed floating-point values
7269 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7270 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7271 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7272 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7275 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7276 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7277 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7281 let Predicates = [HasAVX] in {
7282 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7284 (VINSERTF128rr VR256:$src1, VR128:$src2,
7285 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7286 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7288 (VINSERTF128rr VR256:$src1, VR128:$src2,
7289 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7291 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (memopv4f32 addr:$src2),
7293 (VINSERTF128rm VR256:$src1, addr:$src2,
7294 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7295 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (memopv2f64 addr:$src2),
7297 (VINSERTF128rm VR256:$src1, addr:$src2,
7298 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7301 let Predicates = [HasAVX1Only] in {
7302 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7304 (VINSERTF128rr VR256:$src1, VR128:$src2,
7305 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7306 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7308 (VINSERTF128rr VR256:$src1, VR128:$src2,
7309 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7310 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7312 (VINSERTF128rr VR256:$src1, VR128:$src2,
7313 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7314 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7316 (VINSERTF128rr VR256:$src1, VR128:$src2,
7317 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7319 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7321 (VINSERTF128rm VR256:$src1, addr:$src2,
7322 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7323 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7324 (bc_v4i32 (memopv2i64 addr:$src2)),
7326 (VINSERTF128rm VR256:$src1, addr:$src2,
7327 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7328 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7329 (bc_v16i8 (memopv2i64 addr:$src2)),
7331 (VINSERTF128rm VR256:$src1, addr:$src2,
7332 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7333 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7334 (bc_v8i16 (memopv2i64 addr:$src2)),
7336 (VINSERTF128rm VR256:$src1, addr:$src2,
7337 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7340 //===----------------------------------------------------------------------===//
7341 // VEXTRACTF128 - Extract packed floating-point values
7343 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7344 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7345 (ins VR256:$src1, i8imm:$src2),
7346 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7349 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7350 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7351 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7356 let Predicates = [HasAVX] in {
7357 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7358 (v4f32 (VEXTRACTF128rr
7359 (v8f32 VR256:$src1),
7360 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7361 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7362 (v2f64 (VEXTRACTF128rr
7363 (v4f64 VR256:$src1),
7364 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7366 def : Pat<(alignedstore (v4f32 (vextractf128_extract:$ext (v8f32 VR256:$src1),
7367 (iPTR imm))), addr:$dst),
7368 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7369 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7370 def : Pat<(alignedstore (v2f64 (vextractf128_extract:$ext (v4f64 VR256:$src1),
7371 (iPTR imm))), addr:$dst),
7372 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7373 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7376 let Predicates = [HasAVX1Only] in {
7377 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7378 (v2i64 (VEXTRACTF128rr
7379 (v4i64 VR256:$src1),
7380 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7381 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7382 (v4i32 (VEXTRACTF128rr
7383 (v8i32 VR256:$src1),
7384 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7385 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7386 (v8i16 (VEXTRACTF128rr
7387 (v16i16 VR256:$src1),
7388 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7389 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7390 (v16i8 (VEXTRACTF128rr
7391 (v32i8 VR256:$src1),
7392 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7394 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7395 (iPTR imm))), addr:$dst),
7396 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7397 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7398 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7399 (iPTR imm))), addr:$dst),
7400 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7401 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7402 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7403 (iPTR imm))), addr:$dst),
7404 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7405 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7406 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7407 (iPTR imm))), addr:$dst),
7408 (VEXTRACTF128mr addr:$dst, VR256:$src1,
7409 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7412 //===----------------------------------------------------------------------===//
7413 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7415 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7416 Intrinsic IntLd, Intrinsic IntLd256,
7417 Intrinsic IntSt, Intrinsic IntSt256> {
7418 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7419 (ins VR128:$src1, f128mem:$src2),
7420 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7421 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7423 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7424 (ins VR256:$src1, f256mem:$src2),
7425 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7426 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7428 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7429 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7430 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7431 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7432 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7433 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7434 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7435 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
7438 let ExeDomain = SSEPackedSingle in
7439 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7440 int_x86_avx_maskload_ps,
7441 int_x86_avx_maskload_ps_256,
7442 int_x86_avx_maskstore_ps,
7443 int_x86_avx_maskstore_ps_256>;
7444 let ExeDomain = SSEPackedDouble in
7445 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7446 int_x86_avx_maskload_pd,
7447 int_x86_avx_maskload_pd_256,
7448 int_x86_avx_maskstore_pd,
7449 int_x86_avx_maskstore_pd_256>;
7451 //===----------------------------------------------------------------------===//
7452 // VPERMIL - Permute Single and Double Floating-Point Values
7454 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7455 RegisterClass RC, X86MemOperand x86memop_f,
7456 X86MemOperand x86memop_i, PatFrag i_frag,
7457 Intrinsic IntVar, ValueType vt> {
7458 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7459 (ins RC:$src1, RC:$src2),
7460 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7461 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7462 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7463 (ins RC:$src1, x86memop_i:$src2),
7464 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7465 [(set RC:$dst, (IntVar RC:$src1,
7466 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7468 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7469 (ins RC:$src1, i8imm:$src2),
7470 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7471 [(set RC:$dst, (vt (X86VPermilp RC:$src1, (i8 imm:$src2))))]>, VEX;
7472 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7473 (ins x86memop_f:$src1, i8imm:$src2),
7474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7476 (vt (X86VPermilp (memop addr:$src1), (i8 imm:$src2))))]>, VEX;
7479 let ExeDomain = SSEPackedSingle in {
7480 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7481 memopv2i64, int_x86_avx_vpermilvar_ps, v4f32>;
7482 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7483 memopv4i64, int_x86_avx_vpermilvar_ps_256, v8f32>, VEX_L;
7485 let ExeDomain = SSEPackedDouble in {
7486 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7487 memopv2i64, int_x86_avx_vpermilvar_pd, v2f64>;
7488 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7489 memopv4i64, int_x86_avx_vpermilvar_pd_256, v4f64>, VEX_L;
7492 let Predicates = [HasAVX] in {
7493 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7494 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7495 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7496 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7497 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7499 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7500 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7501 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7503 def : Pat<(v2i64 (X86VPermilp VR128:$src1, (i8 imm:$imm))),
7504 (VPERMILPDri VR128:$src1, imm:$imm)>;
7505 def : Pat<(v2i64 (X86VPermilp (memopv2i64 addr:$src1), (i8 imm:$imm))),
7506 (VPERMILPDmi addr:$src1, imm:$imm)>;
7509 //===----------------------------------------------------------------------===//
7510 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7512 let ExeDomain = SSEPackedSingle in {
7513 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7514 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7515 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7516 [(set VR256:$dst, (v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7517 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7518 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7519 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7520 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7521 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv8f32 addr:$src2),
7522 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7525 let Predicates = [HasAVX] in {
7526 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7527 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7528 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7529 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7530 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7533 let Predicates = [HasAVX1Only] in {
7534 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7535 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7536 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7537 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7538 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7539 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7540 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7541 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7543 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7544 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7545 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7546 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7547 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7548 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7549 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7550 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7551 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7552 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7553 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7554 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7557 //===----------------------------------------------------------------------===//
7558 // VZERO - Zero YMM registers
7560 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7561 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7562 // Zero All YMM registers
7563 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7564 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7566 // Zero Upper bits of YMM registers
7567 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7568 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7571 //===----------------------------------------------------------------------===//
7572 // Half precision conversion instructions
7573 //===----------------------------------------------------------------------===//
7574 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7575 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7576 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7577 [(set RC:$dst, (Int VR128:$src))]>,
7579 let neverHasSideEffects = 1, mayLoad = 1 in
7580 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7581 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7584 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7585 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7586 (ins RC:$src1, i32i8imm:$src2),
7587 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7588 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7590 let neverHasSideEffects = 1, mayStore = 1 in
7591 def mr : Ii8<0x1D, MRMDestMem, (outs),
7592 (ins x86memop:$dst, RC:$src1, i32i8imm:$src2),
7593 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7597 let Predicates = [HasAVX, HasF16C] in {
7598 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7599 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>, VEX_L;
7600 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7601 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>, VEX_L;
7604 //===----------------------------------------------------------------------===//
7605 // AVX2 Instructions
7606 //===----------------------------------------------------------------------===//
7608 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7609 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7610 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7611 X86MemOperand x86memop> {
7612 let isCommutable = 1 in
7613 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7614 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7615 !strconcat(OpcodeStr,
7616 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7617 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7619 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7620 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7621 !strconcat(OpcodeStr,
7622 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7625 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7629 let isCommutable = 0 in {
7630 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7631 VR128, memopv2i64, i128mem>;
7632 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7633 VR256, memopv4i64, i256mem>, VEX_L;
7636 def : Pat<(v4i32 (X86Blendi (v4i32 VR128:$src1), (v4i32 VR128:$src2),
7638 (VPBLENDDrri VR128:$src1, VR128:$src2, imm:$mask)>;
7639 def : Pat<(v8i32 (X86Blendi (v8i32 VR256:$src1), (v8i32 VR256:$src2),
7641 (VPBLENDDYrri VR256:$src1, VR256:$src2, imm:$mask)>;
7643 //===----------------------------------------------------------------------===//
7644 // VPBROADCAST - Load from memory and broadcast to all elements of the
7645 // destination operand
7647 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7648 X86MemOperand x86memop, PatFrag ld_frag,
7649 Intrinsic Int128, Intrinsic Int256> {
7650 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7651 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7652 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7653 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7654 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7656 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7657 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7658 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7659 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX, VEX_L;
7660 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7663 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>,
7667 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7668 int_x86_avx2_pbroadcastb_128,
7669 int_x86_avx2_pbroadcastb_256>;
7670 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7671 int_x86_avx2_pbroadcastw_128,
7672 int_x86_avx2_pbroadcastw_256>;
7673 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7674 int_x86_avx2_pbroadcastd_128,
7675 int_x86_avx2_pbroadcastd_256>;
7676 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7677 int_x86_avx2_pbroadcastq_128,
7678 int_x86_avx2_pbroadcastq_256>;
7680 let Predicates = [HasAVX2] in {
7681 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7682 (VPBROADCASTBrm addr:$src)>;
7683 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7684 (VPBROADCASTBYrm addr:$src)>;
7685 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7686 (VPBROADCASTWrm addr:$src)>;
7687 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7688 (VPBROADCASTWYrm addr:$src)>;
7689 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7690 (VPBROADCASTDrm addr:$src)>;
7691 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7692 (VPBROADCASTDYrm addr:$src)>;
7693 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7694 (VPBROADCASTQrm addr:$src)>;
7695 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7696 (VPBROADCASTQYrm addr:$src)>;
7698 def : Pat<(v16i8 (X86VBroadcast (v16i8 VR128:$src))),
7699 (VPBROADCASTBrr VR128:$src)>;
7700 def : Pat<(v32i8 (X86VBroadcast (v16i8 VR128:$src))),
7701 (VPBROADCASTBYrr VR128:$src)>;
7702 def : Pat<(v8i16 (X86VBroadcast (v8i16 VR128:$src))),
7703 (VPBROADCASTWrr VR128:$src)>;
7704 def : Pat<(v16i16 (X86VBroadcast (v8i16 VR128:$src))),
7705 (VPBROADCASTWYrr VR128:$src)>;
7706 def : Pat<(v4i32 (X86VBroadcast (v4i32 VR128:$src))),
7707 (VPBROADCASTDrr VR128:$src)>;
7708 def : Pat<(v8i32 (X86VBroadcast (v4i32 VR128:$src))),
7709 (VPBROADCASTDYrr VR128:$src)>;
7710 def : Pat<(v2i64 (X86VBroadcast (v2i64 VR128:$src))),
7711 (VPBROADCASTQrr VR128:$src)>;
7712 def : Pat<(v4i64 (X86VBroadcast (v2i64 VR128:$src))),
7713 (VPBROADCASTQYrr VR128:$src)>;
7714 def : Pat<(v4f32 (X86VBroadcast (v4f32 VR128:$src))),
7715 (VBROADCASTSSrr VR128:$src)>;
7716 def : Pat<(v8f32 (X86VBroadcast (v4f32 VR128:$src))),
7717 (VBROADCASTSSYrr VR128:$src)>;
7718 def : Pat<(v2f64 (X86VBroadcast (v2f64 VR128:$src))),
7719 (VPBROADCASTQrr VR128:$src)>;
7720 def : Pat<(v4f64 (X86VBroadcast (v2f64 VR128:$src))),
7721 (VBROADCASTSDYrr VR128:$src)>;
7723 // Provide fallback in case the load node that is used in the patterns above
7724 // is used by additional users, which prevents the pattern selection.
7725 let AddedComplexity = 20 in {
7726 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7727 (VBROADCASTSSrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7728 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7729 (VBROADCASTSSYrr (COPY_TO_REGCLASS FR32:$src, VR128))>;
7730 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7731 (VBROADCASTSDYrr (COPY_TO_REGCLASS FR64:$src, VR128))>;
7733 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7734 (VBROADCASTSSrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7735 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7736 (VBROADCASTSSYrr (COPY_TO_REGCLASS GR32:$src, VR128))>;
7737 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7738 (VBROADCASTSDYrr (COPY_TO_REGCLASS GR64:$src, VR128))>;
7742 // AVX1 broadcast patterns
7743 let Predicates = [HasAVX1Only] in {
7744 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7745 (VBROADCASTSSYrm addr:$src)>;
7746 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7747 (VBROADCASTSDYrm addr:$src)>;
7748 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7749 (VBROADCASTSSrm addr:$src)>;
7752 let Predicates = [HasAVX] in {
7753 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7754 (VBROADCASTSSYrm addr:$src)>;
7755 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7756 (VBROADCASTSDYrm addr:$src)>;
7757 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7758 (VBROADCASTSSrm addr:$src)>;
7760 // Provide fallback in case the load node that is used in the patterns above
7761 // is used by additional users, which prevents the pattern selection.
7762 let AddedComplexity = 20 in {
7763 // 128bit broadcasts:
7764 def : Pat<(v4f32 (X86VBroadcast FR32:$src)),
7765 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0)>;
7766 def : Pat<(v8f32 (X86VBroadcast FR32:$src)),
7767 (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
7768 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), sub_xmm),
7769 (VPSHUFDri (COPY_TO_REGCLASS FR32:$src, VR128), 0), 1)>;
7770 def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
7771 (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
7772 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), sub_xmm),
7773 (VPSHUFDri (COPY_TO_REGCLASS FR64:$src, VR128), 0x44), 1)>;
7775 def : Pat<(v4i32 (X86VBroadcast GR32:$src)),
7776 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0)>;
7777 def : Pat<(v8i32 (X86VBroadcast GR32:$src)),
7778 (VINSERTF128rr (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)),
7779 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), sub_xmm),
7780 (VPSHUFDri (COPY_TO_REGCLASS GR32:$src, VR128), 0), 1)>;
7781 def : Pat<(v4i64 (X86VBroadcast GR64:$src)),
7782 (VINSERTF128rr (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)),
7783 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), sub_xmm),
7784 (VPSHUFDri (COPY_TO_REGCLASS GR64:$src, VR128), 0x44), 1)>;
7788 //===----------------------------------------------------------------------===//
7789 // VPERM - Permute instructions
7792 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7794 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7795 (ins VR256:$src1, VR256:$src2),
7796 !strconcat(OpcodeStr,
7797 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7799 (OpVT (X86VPermv VR256:$src1, VR256:$src2)))]>,
7801 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7802 (ins VR256:$src1, i256mem:$src2),
7803 !strconcat(OpcodeStr,
7804 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7806 (OpVT (X86VPermv VR256:$src1,
7807 (bitconvert (mem_frag addr:$src2)))))]>,
7811 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, v8i32>;
7812 let ExeDomain = SSEPackedSingle in
7813 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, v8f32>;
7815 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7817 def Yri : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7818 (ins VR256:$src1, i8imm:$src2),
7819 !strconcat(OpcodeStr,
7820 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7822 (OpVT (X86VPermi VR256:$src1, (i8 imm:$src2))))]>,
7824 def Ymi : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7825 (ins i256mem:$src1, i8imm:$src2),
7826 !strconcat(OpcodeStr,
7827 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7829 (OpVT (X86VPermi (mem_frag addr:$src1),
7830 (i8 imm:$src2))))]>, VEX, VEX_L;
7833 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, v4i64>, VEX_W;
7834 let ExeDomain = SSEPackedDouble in
7835 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, v4f64>, VEX_W;
7837 //===----------------------------------------------------------------------===//
7838 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7840 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7841 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7842 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7843 [(set VR256:$dst, (v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2,
7844 (i8 imm:$src3))))]>, VEX_4V, VEX_L;
7845 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7846 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7847 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7848 [(set VR256:$dst, (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7849 (i8 imm:$src3)))]>, VEX_4V, VEX_L;
7851 let Predicates = [HasAVX2] in {
7852 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7853 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7854 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7855 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7856 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7857 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7859 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7861 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7862 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7863 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7864 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7865 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7867 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7871 //===----------------------------------------------------------------------===//
7872 // VINSERTI128 - Insert packed integer values
7874 let neverHasSideEffects = 1 in {
7875 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7876 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7877 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7880 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7881 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7882 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7886 let Predicates = [HasAVX2] in {
7887 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7889 (VINSERTI128rr VR256:$src1, VR128:$src2,
7890 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7891 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7893 (VINSERTI128rr VR256:$src1, VR128:$src2,
7894 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7895 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7897 (VINSERTI128rr VR256:$src1, VR128:$src2,
7898 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7899 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7901 (VINSERTI128rr VR256:$src1, VR128:$src2,
7902 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7904 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (memopv2i64 addr:$src2),
7906 (VINSERTI128rm VR256:$src1, addr:$src2,
7907 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7908 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1),
7909 (bc_v4i32 (memopv2i64 addr:$src2)),
7911 (VINSERTI128rm VR256:$src1, addr:$src2,
7912 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7913 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1),
7914 (bc_v16i8 (memopv2i64 addr:$src2)),
7916 (VINSERTI128rm VR256:$src1, addr:$src2,
7917 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7918 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1),
7919 (bc_v8i16 (memopv2i64 addr:$src2)),
7921 (VINSERTI128rm VR256:$src1, addr:$src2,
7922 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7925 //===----------------------------------------------------------------------===//
7926 // VEXTRACTI128 - Extract packed integer values
7928 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7929 (ins VR256:$src1, i8imm:$src2),
7930 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7932 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7934 let neverHasSideEffects = 1, mayStore = 1 in
7935 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7936 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7937 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7940 let Predicates = [HasAVX2] in {
7941 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7942 (v2i64 (VEXTRACTI128rr
7943 (v4i64 VR256:$src1),
7944 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7945 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7946 (v4i32 (VEXTRACTI128rr
7947 (v8i32 VR256:$src1),
7948 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7949 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7950 (v8i16 (VEXTRACTI128rr
7951 (v16i16 VR256:$src1),
7952 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7953 def : Pat<(vextractf128_extract:$ext VR256:$src1, (iPTR imm)),
7954 (v16i8 (VEXTRACTI128rr
7955 (v32i8 VR256:$src1),
7956 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7958 def : Pat<(alignedstore (v2i64 (vextractf128_extract:$ext (v4i64 VR256:$src1),
7959 (iPTR imm))), addr:$dst),
7960 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7961 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7962 def : Pat<(alignedstore (v4i32 (vextractf128_extract:$ext (v8i32 VR256:$src1),
7963 (iPTR imm))), addr:$dst),
7964 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7965 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7966 def : Pat<(alignedstore (v8i16 (vextractf128_extract:$ext (v16i16 VR256:$src1),
7967 (iPTR imm))), addr:$dst),
7968 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7969 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7970 def : Pat<(alignedstore (v16i8 (vextractf128_extract:$ext (v32i8 VR256:$src1),
7971 (iPTR imm))), addr:$dst),
7972 (VEXTRACTI128mr addr:$dst, VR256:$src1,
7973 (EXTRACT_get_vextractf128_imm VR128:$ext))>;
7976 //===----------------------------------------------------------------------===//
7977 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7979 multiclass avx2_pmovmask<string OpcodeStr,
7980 Intrinsic IntLd128, Intrinsic IntLd256,
7981 Intrinsic IntSt128, Intrinsic IntSt256> {
7982 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7983 (ins VR128:$src1, i128mem:$src2),
7984 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7985 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7986 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7987 (ins VR256:$src1, i256mem:$src2),
7988 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7989 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7991 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7992 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7993 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7994 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7995 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7996 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7997 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7998 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V, VEX_L;
8001 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
8002 int_x86_avx2_maskload_d,
8003 int_x86_avx2_maskload_d_256,
8004 int_x86_avx2_maskstore_d,
8005 int_x86_avx2_maskstore_d_256>;
8006 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
8007 int_x86_avx2_maskload_q,
8008 int_x86_avx2_maskload_q_256,
8009 int_x86_avx2_maskstore_q,
8010 int_x86_avx2_maskstore_q_256>, VEX_W;
8013 //===----------------------------------------------------------------------===//
8014 // Variable Bit Shifts
8016 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
8017 ValueType vt128, ValueType vt256> {
8018 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
8019 (ins VR128:$src1, VR128:$src2),
8020 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8022 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
8024 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
8025 (ins VR128:$src1, i128mem:$src2),
8026 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8028 (vt128 (OpNode VR128:$src1,
8029 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
8031 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
8032 (ins VR256:$src1, VR256:$src2),
8033 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8035 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
8037 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
8038 (ins VR256:$src1, i256mem:$src2),
8039 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
8041 (vt256 (OpNode VR256:$src1,
8042 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
8046 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
8047 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
8048 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
8049 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
8050 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;
8052 //===----------------------------------------------------------------------===//
8053 // VGATHER - GATHER Operations
8054 multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
8055 X86MemOperand memop128, X86MemOperand memop256> {
8056 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst, VR128:$mask_wb),
8057 (ins VR128:$src1, memop128:$src2, VR128:$mask),
8058 !strconcat(OpcodeStr,
8059 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8061 def Yrm : AVX28I<opc, MRMSrcMem, (outs RC256:$dst, RC256:$mask_wb),
8062 (ins RC256:$src1, memop256:$src2, RC256:$mask),
8063 !strconcat(OpcodeStr,
8064 "\t{$mask, $src2, $dst|$dst, $src2, $mask}"),
8065 []>, VEX_4VOp3, VEX_L;
8068 let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
8069 defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
8070 defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
8071 defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
8072 defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
8073 defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
8074 defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
8075 defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
8076 defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;