1 //====- X86InstrSSE.td - Describe the X86 Instruction Set -------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the Evan Cheng and is distributed under
6 // the University of Illinois Open Source 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 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE specific DAG Nodes.
19 //===----------------------------------------------------------------------===//
21 def X86loadp : SDNode<"X86ISD::LOAD_PACK", SDTLoad, [SDNPHasChain]>;
22 def X86loadu : SDNode<"X86ISD::LOAD_UA", SDTLoad, [SDNPHasChain]>;
23 def X86fand : SDNode<"X86ISD::FAND", SDTFPBinOp,
24 [SDNPCommutative, SDNPAssociative]>;
25 def X86fxor : SDNode<"X86ISD::FXOR", SDTFPBinOp,
26 [SDNPCommutative, SDNPAssociative]>;
27 def X86comi : SDNode<"X86ISD::COMI", SDTX86CmpTest,
28 [SDNPHasChain, SDNPOutFlag]>;
29 def X86ucomi : SDNode<"X86ISD::UCOMI", SDTX86CmpTest,
30 [SDNPHasChain, SDNPOutFlag]>;
31 def X86s2vec : SDNode<"X86ISD::S2VEC", SDTypeProfile<1, 1, []>, []>;
32 def X86pextrw : SDNode<"X86ISD::PEXTRW", SDTypeProfile<1, 2, []>, []>;
33 def X86pinsrw : SDNode<"X86ISD::PINSRW", SDTypeProfile<1, 3, []>, []>;
35 //===----------------------------------------------------------------------===//
36 // SSE Complex Patterns
37 //===----------------------------------------------------------------------===//
39 // These are 'extloads' from a scalar to the low element of a vector, zeroing
40 // the top elements. These are used for the SSE 'ss' and 'sd' instruction
42 def sse_load_f32 : ComplexPattern<v4f32, 4, "SelectScalarSSELoad", []>;
43 def sse_load_f64 : ComplexPattern<v2f64, 4, "SelectScalarSSELoad", []>;
45 def ssmem : Operand<v4f32> {
46 let PrintMethod = "printf32mem";
47 let NumMIOperands = 4;
48 let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm);
50 def sdmem : Operand<v2f64> {
51 let PrintMethod = "printf64mem";
52 let NumMIOperands = 4;
53 let MIOperandInfo = (ops ptr_rc, i8imm, ptr_rc, i32imm);
56 //===----------------------------------------------------------------------===//
57 // SSE pattern fragments
58 //===----------------------------------------------------------------------===//
60 def X86loadpf32 : PatFrag<(ops node:$ptr), (f32 (X86loadp node:$ptr))>;
61 def X86loadpf64 : PatFrag<(ops node:$ptr), (f64 (X86loadp node:$ptr))>;
63 def loadv4f32 : PatFrag<(ops node:$ptr), (v4f32 (load node:$ptr))>;
64 def loadv2f64 : PatFrag<(ops node:$ptr), (v2f64 (load node:$ptr))>;
65 def loadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (load node:$ptr))>;
67 def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
68 def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
69 def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
70 def bc_v8i16 : PatFrag<(ops node:$in), (v8i16 (bitconvert node:$in))>;
71 def bc_v4i32 : PatFrag<(ops node:$in), (v4i32 (bitconvert node:$in))>;
72 def bc_v2i64 : PatFrag<(ops node:$in), (v2i64 (bitconvert node:$in))>;
74 def fp32imm0 : PatLeaf<(f32 fpimm), [{
75 return N->isExactlyValue(+0.0);
78 def PSxLDQ_imm : SDNodeXForm<imm, [{
79 // Transformation function: imm >> 3
80 return getI32Imm(N->getValue() >> 3);
83 // SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to PSHUF*,
85 def SHUFFLE_get_shuf_imm : SDNodeXForm<build_vector, [{
86 return getI8Imm(X86::getShuffleSHUFImmediate(N));
89 // SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to
91 def SHUFFLE_get_pshufhw_imm : SDNodeXForm<build_vector, [{
92 return getI8Imm(X86::getShufflePSHUFHWImmediate(N));
95 // SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to
97 def SHUFFLE_get_pshuflw_imm : SDNodeXForm<build_vector, [{
98 return getI8Imm(X86::getShufflePSHUFLWImmediate(N));
101 def SSE_splat_mask : PatLeaf<(build_vector), [{
102 return X86::isSplatMask(N);
103 }], SHUFFLE_get_shuf_imm>;
105 def SSE_splat_v2_mask : PatLeaf<(build_vector), [{
106 return X86::isSplatMask(N);
109 def MOVHLPS_shuffle_mask : PatLeaf<(build_vector), [{
110 return X86::isMOVHLPSMask(N);
113 def MOVHP_shuffle_mask : PatLeaf<(build_vector), [{
114 return X86::isMOVHPMask(N);
117 def MOVLP_shuffle_mask : PatLeaf<(build_vector), [{
118 return X86::isMOVLPMask(N);
121 def MOVL_shuffle_mask : PatLeaf<(build_vector), [{
122 return X86::isMOVLMask(N);
125 def MOVSHDUP_shuffle_mask : PatLeaf<(build_vector), [{
126 return X86::isMOVSHDUPMask(N);
129 def MOVSLDUP_shuffle_mask : PatLeaf<(build_vector), [{
130 return X86::isMOVSLDUPMask(N);
133 def UNPCKL_shuffle_mask : PatLeaf<(build_vector), [{
134 return X86::isUNPCKLMask(N);
137 def UNPCKH_shuffle_mask : PatLeaf<(build_vector), [{
138 return X86::isUNPCKHMask(N);
141 def UNPCKL_v_undef_shuffle_mask : PatLeaf<(build_vector), [{
142 return X86::isUNPCKL_v_undef_Mask(N);
145 def PSHUFD_shuffle_mask : PatLeaf<(build_vector), [{
146 return X86::isPSHUFDMask(N);
147 }], SHUFFLE_get_shuf_imm>;
149 def PSHUFHW_shuffle_mask : PatLeaf<(build_vector), [{
150 return X86::isPSHUFHWMask(N);
151 }], SHUFFLE_get_pshufhw_imm>;
153 def PSHUFLW_shuffle_mask : PatLeaf<(build_vector), [{
154 return X86::isPSHUFLWMask(N);
155 }], SHUFFLE_get_pshuflw_imm>;
157 def SHUFP_unary_shuffle_mask : PatLeaf<(build_vector), [{
158 return X86::isPSHUFDMask(N);
159 }], SHUFFLE_get_shuf_imm>;
161 def SHUFP_shuffle_mask : PatLeaf<(build_vector), [{
162 return X86::isSHUFPMask(N);
163 }], SHUFFLE_get_shuf_imm>;
165 def PSHUFD_binary_shuffle_mask : PatLeaf<(build_vector), [{
166 return X86::isSHUFPMask(N);
167 }], SHUFFLE_get_shuf_imm>;
169 //===----------------------------------------------------------------------===//
170 // SSE scalar FP Instructions
171 //===----------------------------------------------------------------------===//
173 // Instruction templates
174 // SSI - SSE1 instructions with XS prefix.
175 // SDI - SSE2 instructions with XD prefix.
176 // PSI - SSE1 instructions with TB prefix.
177 // PDI - SSE2 instructions with TB and OpSize prefixes.
178 // PSIi8 - SSE1 instructions with ImmT == Imm8 and TB prefix.
179 // PDIi8 - SSE2 instructions with ImmT == Imm8 and TB and OpSize prefixes.
180 // S3I - SSE3 instructions with TB and OpSize prefixes.
181 // S3SI - SSE3 instructions with XS prefix.
182 // S3DI - SSE3 instructions with XD prefix.
183 class SSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
184 : I<o, F, ops, asm, pattern>, XS, Requires<[HasSSE1]>;
185 class SDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
186 : I<o, F, ops, asm, pattern>, XD, Requires<[HasSSE2]>;
187 class PSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
188 : I<o, F, ops, asm, pattern>, TB, Requires<[HasSSE1]>;
189 class PDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
190 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
191 class PSIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
192 : Ii8<o, F, ops, asm, pattern>, TB, Requires<[HasSSE1]>;
193 class PDIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
194 : Ii8<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
196 class S3SI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
197 : I<o, F, ops, asm, pattern>, XS, Requires<[HasSSE3]>;
198 class S3DI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
199 : I<o, F, ops, asm, pattern>, XD, Requires<[HasSSE3]>;
200 class S3I<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
201 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE3]>;
203 //===----------------------------------------------------------------------===//
204 // Helpers for defining instructions that directly correspond to intrinsics.
206 multiclass SS_IntUnary<bits<8> o, string OpcodeStr, Intrinsic IntId> {
207 def r : SSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
208 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
209 [(set VR128:$dst, (v4f32 (IntId VR128:$src)))]>;
210 def m : SSI<o, MRMSrcMem, (ops VR128:$dst, ssmem:$src),
211 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
212 [(set VR128:$dst, (v4f32 (IntId sse_load_f32:$src)))]>;
215 multiclass SD_IntUnary<bits<8> o, string OpcodeStr, Intrinsic IntId> {
216 def r : SDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
217 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
218 [(set VR128:$dst, (v2f64 (IntId VR128:$src)))]>;
219 def m : SDI<o, MRMSrcMem, (ops VR128:$dst, sdmem:$src),
220 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
221 [(set VR128:$dst, (v2f64 (IntId sse_load_f64:$src)))]>;
224 class PS_Intr<bits<8> o, string OpcodeStr, Intrinsic IntId>
225 : PSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
226 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
227 [(set VR128:$dst, (IntId VR128:$src))]>;
228 class PS_Intm<bits<8> o, string OpcodeStr, Intrinsic IntId>
229 : PSI<o, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
230 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
231 [(set VR128:$dst, (IntId (load addr:$src)))]>;
232 class PD_Intr<bits<8> o, string OpcodeStr, Intrinsic IntId>
233 : PDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
234 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
235 [(set VR128:$dst, (IntId VR128:$src))]>;
236 class PD_Intm<bits<8> o, string OpcodeStr, Intrinsic IntId>
237 : PDI<o, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
238 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
239 [(set VR128:$dst, (IntId (load addr:$src)))]>;
241 class PS_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
242 : PSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
243 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
244 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
245 class PS_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
246 : PSI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
247 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
248 [(set VR128:$dst, (IntId VR128:$src1, (load addr:$src2)))]>;
249 class PD_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
250 : PDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
251 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
252 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
253 class PD_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
254 : PDI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
255 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
256 [(set VR128:$dst, (IntId VR128:$src1, (load addr:$src2)))]>;
258 // Some 'special' instructions
259 def IMPLICIT_DEF_FR32 : I<0, Pseudo, (ops FR32:$dst),
260 "#IMPLICIT_DEF $dst",
261 [(set FR32:$dst, (undef))]>, Requires<[HasSSE2]>;
262 def IMPLICIT_DEF_FR64 : I<0, Pseudo, (ops FR64:$dst),
263 "#IMPLICIT_DEF $dst",
264 [(set FR64:$dst, (undef))]>, Requires<[HasSSE2]>;
266 // CMOV* - Used to implement the SSE SELECT DAG operation. Expanded by the
267 // scheduler into a branch sequence.
268 let usesCustomDAGSchedInserter = 1 in { // Expanded by the scheduler.
269 def CMOV_FR32 : I<0, Pseudo,
270 (ops FR32:$dst, FR32:$t, FR32:$f, i8imm:$cond),
271 "#CMOV_FR32 PSEUDO!",
272 [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond))]>;
273 def CMOV_FR64 : I<0, Pseudo,
274 (ops FR64:$dst, FR64:$t, FR64:$f, i8imm:$cond),
275 "#CMOV_FR64 PSEUDO!",
276 [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond))]>;
277 def CMOV_V4F32 : I<0, Pseudo,
278 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
279 "#CMOV_V4F32 PSEUDO!",
281 (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
282 def CMOV_V2F64 : I<0, Pseudo,
283 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
284 "#CMOV_V2F64 PSEUDO!",
286 (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
287 def CMOV_V2I64 : I<0, Pseudo,
288 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
289 "#CMOV_V2I64 PSEUDO!",
291 (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
295 def MOVSSrr : SSI<0x10, MRMSrcReg, (ops FR32:$dst, FR32:$src),
296 "movss {$src, $dst|$dst, $src}", []>;
297 def MOVSSrm : SSI<0x10, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
298 "movss {$src, $dst|$dst, $src}",
299 [(set FR32:$dst, (loadf32 addr:$src))]>;
300 def MOVSDrr : SDI<0x10, MRMSrcReg, (ops FR64:$dst, FR64:$src),
301 "movsd {$src, $dst|$dst, $src}", []>;
302 def MOVSDrm : SDI<0x10, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
303 "movsd {$src, $dst|$dst, $src}",
304 [(set FR64:$dst, (loadf64 addr:$src))]>;
306 def MOVSSmr : SSI<0x11, MRMDestMem, (ops f32mem:$dst, FR32:$src),
307 "movss {$src, $dst|$dst, $src}",
308 [(store FR32:$src, addr:$dst)]>;
309 def MOVSDmr : SDI<0x11, MRMDestMem, (ops f64mem:$dst, FR64:$src),
310 "movsd {$src, $dst|$dst, $src}",
311 [(store FR64:$src, addr:$dst)]>;
313 /// scalar_sse12_fp_binop_rm - Scalar SSE binops come in four basic forms:
314 /// 1. f32 vs f64 - These come in SSE1/SSE2 forms for float/doubles.
315 /// 2. rr vs rm - They include a reg+reg form and a ref+mem form.
317 /// In addition, scalar SSE ops have an intrinsic form. This form is unlike the
318 /// normal form, in that they take an entire vector (instead of a scalar) and
319 /// leave the top elements undefined. This adds another two variants of the
320 /// above permutations, giving us 8 forms for 'instruction'.
322 let isTwoAddress = 1 in {
323 multiclass scalar_sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
324 SDNode OpNode, Intrinsic F32Int,
325 Intrinsic F64Int, bit Commutable = 0> {
326 // Scalar operation, reg+reg.
327 def SSrr : SSI<opc, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
328 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
329 [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
330 let isCommutable = Commutable;
332 def SDrr : SDI<opc, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
333 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
334 [(set FR64:$dst, (OpNode FR64:$src1, FR64:$src2))]> {
335 let isCommutable = Commutable;
337 // Scalar operation, reg+mem.
338 def SSrm : SSI<opc, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
339 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
340 [(set FR32:$dst, (OpNode FR32:$src1, (load addr:$src2)))]>;
341 def SDrm : SDI<opc, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
342 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
343 [(set FR64:$dst, (OpNode FR64:$src1, (load addr:$src2)))]>;
345 // Vector intrinsic operation, reg+reg.
346 def SSrr_Int : SSI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
347 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
348 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]> {
349 let isCommutable = Commutable;
351 def SDrr_Int : SDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
352 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
353 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2))]> {
354 let isCommutable = Commutable;
356 // Vector intrinsic operation, reg+mem.
357 def SSrm_Int : SSI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, ssmem:$src2),
358 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
359 [(set VR128:$dst, (F32Int VR128:$src1,
360 sse_load_f32:$src2))]>;
361 def SDrm_Int : SDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, sdmem:$src2),
362 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
363 [(set VR128:$dst, (F64Int VR128:$src1,
364 sse_load_f64:$src2))]>;
368 // Arithmetic instructions
370 defm ADD : scalar_sse12_fp_binop_rm<0x58, "add", fadd,
371 int_x86_sse_add_ss, int_x86_sse2_add_sd, 1>;
372 defm MUL : scalar_sse12_fp_binop_rm<0x59, "mul", fmul,
373 int_x86_sse_mul_ss, int_x86_sse2_mul_sd, 1>;
374 defm SUB : scalar_sse12_fp_binop_rm<0x5C, "sub", fsub,
375 int_x86_sse_sub_ss, int_x86_sse2_sub_sd>;
376 defm DIV : scalar_sse12_fp_binop_rm<0x5E, "div", fdiv,
377 int_x86_sse_div_ss, int_x86_sse2_div_sd>;
380 def SQRTSSr : SSI<0x51, MRMSrcReg, (ops FR32:$dst, FR32:$src),
381 "sqrtss {$src, $dst|$dst, $src}",
382 [(set FR32:$dst, (fsqrt FR32:$src))]>;
383 def SQRTSSm : SSI<0x51, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
384 "sqrtss {$src, $dst|$dst, $src}",
385 [(set FR32:$dst, (fsqrt (loadf32 addr:$src)))]>;
386 def SQRTSDr : SDI<0x51, MRMSrcReg, (ops FR64:$dst, FR64:$src),
387 "sqrtsd {$src, $dst|$dst, $src}",
388 [(set FR64:$dst, (fsqrt FR64:$src))]>;
389 def SQRTSDm : SDI<0x51, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
390 "sqrtsd {$src, $dst|$dst, $src}",
391 [(set FR64:$dst, (fsqrt (loadf64 addr:$src)))]>;
393 class SS_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
394 : SSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
395 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
396 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, VR128:$src2)))]>;
397 class SS_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
398 : SSI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, ssmem:$src2),
399 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
400 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, sse_load_f32:$src2)))]>;
401 class SD_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
402 : SDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
403 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
404 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, VR128:$src2)))]>;
405 class SD_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
406 : SDI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, sdmem:$src2),
407 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
408 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, sse_load_f64:$src2)))]>;
411 // Aliases to match intrinsics which expect XMM operand(s).
413 defm SQRTSS_Int : SS_IntUnary<0x51, "sqrtss" , int_x86_sse_sqrt_ss>;
414 defm SQRTSD_Int : SD_IntUnary<0x51, "sqrtsd" , int_x86_sse2_sqrt_sd>;
415 defm RSQRTSS_Int : SS_IntUnary<0x52, "rsqrtss", int_x86_sse_rsqrt_ss>;
416 defm RCPSS_Int : SS_IntUnary<0x53, "rcpss" , int_x86_sse_rcp_ss>;
418 let isTwoAddress = 1 in {
419 let isCommutable = 1 in {
420 def Int_MAXSSrr : SS_Intrr<0x5F, "maxss", int_x86_sse_max_ss>;
421 def Int_MAXSDrr : SD_Intrr<0x5F, "maxsd", int_x86_sse2_max_sd>;
422 def Int_MINSSrr : SS_Intrr<0x5D, "minss", int_x86_sse_min_ss>;
423 def Int_MINSDrr : SD_Intrr<0x5D, "minsd", int_x86_sse2_min_sd>;
425 def Int_MAXSSrm : SS_Intrm<0x5F, "maxss", int_x86_sse_max_ss>;
426 def Int_MAXSDrm : SD_Intrm<0x5F, "maxsd", int_x86_sse2_max_sd>;
427 def Int_MINSSrm : SS_Intrm<0x5D, "minss", int_x86_sse_min_ss>;
428 def Int_MINSDrm : SD_Intrm<0x5D, "minsd", int_x86_sse2_min_sd>;
431 // Conversion instructions
432 def CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops GR32:$dst, FR32:$src),
433 "cvttss2si {$src, $dst|$dst, $src}",
434 [(set GR32:$dst, (fp_to_sint FR32:$src))]>;
435 def CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
436 "cvttss2si {$src, $dst|$dst, $src}",
437 [(set GR32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
438 def CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops GR32:$dst, FR64:$src),
439 "cvttsd2si {$src, $dst|$dst, $src}",
440 [(set GR32:$dst, (fp_to_sint FR64:$src))]>;
441 def CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops GR32:$dst, f64mem:$src),
442 "cvttsd2si {$src, $dst|$dst, $src}",
443 [(set GR32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
444 def CVTSD2SSrr: SDI<0x5A, MRMSrcReg, (ops FR32:$dst, FR64:$src),
445 "cvtsd2ss {$src, $dst|$dst, $src}",
446 [(set FR32:$dst, (fround FR64:$src))]>;
447 def CVTSD2SSrm: SDI<0x5A, MRMSrcMem, (ops FR32:$dst, f64mem:$src),
448 "cvtsd2ss {$src, $dst|$dst, $src}",
449 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
450 def CVTSI2SSrr: SSI<0x2A, MRMSrcReg, (ops FR32:$dst, GR32:$src),
451 "cvtsi2ss {$src, $dst|$dst, $src}",
452 [(set FR32:$dst, (sint_to_fp GR32:$src))]>;
453 def CVTSI2SSrm: SSI<0x2A, MRMSrcMem, (ops FR32:$dst, i32mem:$src),
454 "cvtsi2ss {$src, $dst|$dst, $src}",
455 [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
456 def CVTSI2SDrr: SDI<0x2A, MRMSrcReg, (ops FR64:$dst, GR32:$src),
457 "cvtsi2sd {$src, $dst|$dst, $src}",
458 [(set FR64:$dst, (sint_to_fp GR32:$src))]>;
459 def CVTSI2SDrm: SDI<0x2A, MRMSrcMem, (ops FR64:$dst, i32mem:$src),
460 "cvtsi2sd {$src, $dst|$dst, $src}",
461 [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
463 // SSE2 instructions with XS prefix
464 def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
465 "cvtss2sd {$src, $dst|$dst, $src}",
466 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
468 def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
469 "cvtss2sd {$src, $dst|$dst, $src}",
470 [(set FR64:$dst, (extload addr:$src, f32))]>, XS,
473 // Match intrinsics which expect XMM operand(s).
474 def Int_CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (ops GR32:$dst, VR128:$src),
475 "cvtss2si {$src, $dst|$dst, $src}",
476 [(set GR32:$dst, (int_x86_sse_cvtss2si VR128:$src))]>;
477 def Int_CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
478 "cvtss2si {$src, $dst|$dst, $src}",
479 [(set GR32:$dst, (int_x86_sse_cvtss2si
480 (load addr:$src)))]>;
481 def Int_CVTSD2SIrr: SDI<0x2D, MRMSrcReg, (ops GR32:$dst, VR128:$src),
482 "cvtsd2si {$src, $dst|$dst, $src}",
483 [(set GR32:$dst, (int_x86_sse2_cvtsd2si VR128:$src))]>;
484 def Int_CVTSD2SIrm: SDI<0x2D, MRMSrcMem, (ops GR32:$dst, f128mem:$src),
485 "cvtsd2si {$src, $dst|$dst, $src}",
486 [(set GR32:$dst, (int_x86_sse2_cvtsd2si
487 (load addr:$src)))]>;
489 // Aliases for intrinsics
490 def Int_CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops GR32:$dst, VR128:$src),
491 "cvttss2si {$src, $dst|$dst, $src}",
492 [(set GR32:$dst, (int_x86_sse_cvttss2si VR128:$src))]>;
493 def Int_CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
494 "cvttss2si {$src, $dst|$dst, $src}",
495 [(set GR32:$dst, (int_x86_sse_cvttss2si(load addr:$src)))]>;
496 def Int_CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops GR32:$dst, VR128:$src),
497 "cvttsd2si {$src, $dst|$dst, $src}",
498 [(set GR32:$dst, (int_x86_sse2_cvttsd2si VR128:$src))]>;
499 def Int_CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops GR32:$dst, f128mem:$src),
500 "cvttsd2si {$src, $dst|$dst, $src}",
501 [(set GR32:$dst, (int_x86_sse2_cvttsd2si
502 (load addr:$src)))]>;
504 let isTwoAddress = 1 in {
505 def Int_CVTSI2SSrr: SSI<0x2A, MRMSrcReg,
506 (ops VR128:$dst, VR128:$src1, GR32:$src2),
507 "cvtsi2ss {$src2, $dst|$dst, $src2}",
508 [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
510 def Int_CVTSI2SSrm: SSI<0x2A, MRMSrcMem,
511 (ops VR128:$dst, VR128:$src1, i32mem:$src2),
512 "cvtsi2ss {$src2, $dst|$dst, $src2}",
513 [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
514 (loadi32 addr:$src2)))]>;
517 // Comparison instructions
518 let isTwoAddress = 1 in {
519 def CMPSSrr : SSI<0xC2, MRMSrcReg,
520 (ops FR32:$dst, FR32:$src1, FR32:$src, SSECC:$cc),
521 "cmp${cc}ss {$src, $dst|$dst, $src}",
523 def CMPSSrm : SSI<0xC2, MRMSrcMem,
524 (ops FR32:$dst, FR32:$src1, f32mem:$src, SSECC:$cc),
525 "cmp${cc}ss {$src, $dst|$dst, $src}", []>;
526 def CMPSDrr : SDI<0xC2, MRMSrcReg,
527 (ops FR64:$dst, FR64:$src1, FR64:$src, SSECC:$cc),
528 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
529 def CMPSDrm : SDI<0xC2, MRMSrcMem,
530 (ops FR64:$dst, FR64:$src1, f64mem:$src, SSECC:$cc),
531 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
534 def UCOMISSrr: PSI<0x2E, MRMSrcReg, (ops FR32:$src1, FR32:$src2),
535 "ucomiss {$src2, $src1|$src1, $src2}",
536 [(X86cmp FR32:$src1, FR32:$src2)]>;
537 def UCOMISSrm: PSI<0x2E, MRMSrcMem, (ops FR32:$src1, f32mem:$src2),
538 "ucomiss {$src2, $src1|$src1, $src2}",
539 [(X86cmp FR32:$src1, (loadf32 addr:$src2))]>;
540 def UCOMISDrr: PDI<0x2E, MRMSrcReg, (ops FR64:$src1, FR64:$src2),
541 "ucomisd {$src2, $src1|$src1, $src2}",
542 [(X86cmp FR64:$src1, FR64:$src2)]>;
543 def UCOMISDrm: PDI<0x2E, MRMSrcMem, (ops FR64:$src1, f64mem:$src2),
544 "ucomisd {$src2, $src1|$src1, $src2}",
545 [(X86cmp FR64:$src1, (loadf64 addr:$src2))]>;
547 // Aliases to match intrinsics which expect XMM operand(s).
548 let isTwoAddress = 1 in {
549 def Int_CMPSSrr : SSI<0xC2, MRMSrcReg,
550 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
551 "cmp${cc}ss {$src, $dst|$dst, $src}",
552 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
553 VR128:$src, imm:$cc))]>;
554 def Int_CMPSSrm : SSI<0xC2, MRMSrcMem,
555 (ops VR128:$dst, VR128:$src1, f32mem:$src, SSECC:$cc),
556 "cmp${cc}ss {$src, $dst|$dst, $src}",
557 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
558 (load addr:$src), imm:$cc))]>;
559 def Int_CMPSDrr : SDI<0xC2, MRMSrcReg,
560 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
561 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
562 def Int_CMPSDrm : SDI<0xC2, MRMSrcMem,
563 (ops VR128:$dst, VR128:$src1, f64mem:$src, SSECC:$cc),
564 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
567 def Int_UCOMISSrr: PSI<0x2E, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
568 "ucomiss {$src2, $src1|$src1, $src2}",
569 [(X86ucomi (v4f32 VR128:$src1), VR128:$src2)]>;
570 def Int_UCOMISSrm: PSI<0x2E, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
571 "ucomiss {$src2, $src1|$src1, $src2}",
572 [(X86ucomi (v4f32 VR128:$src1), (load addr:$src2))]>;
573 def Int_UCOMISDrr: PDI<0x2E, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
574 "ucomisd {$src2, $src1|$src1, $src2}",
575 [(X86ucomi (v2f64 VR128:$src1), (v2f64 VR128:$src2))]>;
576 def Int_UCOMISDrm: PDI<0x2E, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
577 "ucomisd {$src2, $src1|$src1, $src2}",
578 [(X86ucomi (v2f64 VR128:$src1), (load addr:$src2))]>;
580 def Int_COMISSrr: PSI<0x2F, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
581 "comiss {$src2, $src1|$src1, $src2}",
582 [(X86comi (v4f32 VR128:$src1), VR128:$src2)]>;
583 def Int_COMISSrm: PSI<0x2F, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
584 "comiss {$src2, $src1|$src1, $src2}",
585 [(X86comi (v4f32 VR128:$src1), (load addr:$src2))]>;
586 def Int_COMISDrr: PDI<0x2F, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
587 "comisd {$src2, $src1|$src1, $src2}",
588 [(X86comi (v2f64 VR128:$src1), (v2f64 VR128:$src2))]>;
589 def Int_COMISDrm: PDI<0x2F, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
590 "comisd {$src2, $src1|$src1, $src2}",
591 [(X86comi (v2f64 VR128:$src1), (load addr:$src2))]>;
593 // Aliases of packed instructions for scalar use. These all have names that
596 // Alias instructions that map fld0 to pxor for sse.
597 def FsFLD0SS : I<0xEF, MRMInitReg, (ops FR32:$dst),
598 "pxor $dst, $dst", [(set FR32:$dst, fp32imm0)]>,
599 Requires<[HasSSE1]>, TB, OpSize;
600 def FsFLD0SD : I<0xEF, MRMInitReg, (ops FR64:$dst),
601 "pxor $dst, $dst", [(set FR64:$dst, fp64imm0)]>,
602 Requires<[HasSSE2]>, TB, OpSize;
604 // Alias instructions to do FR32 / FR64 reg-to-reg copy using movaps / movapd.
605 // Upper bits are disregarded.
606 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (ops FR32:$dst, FR32:$src),
607 "movaps {$src, $dst|$dst, $src}", []>;
608 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (ops FR64:$dst, FR64:$src),
609 "movapd {$src, $dst|$dst, $src}", []>;
611 // Alias instructions to load FR32 / FR64 from f128mem using movaps / movapd.
612 // Upper bits are disregarded.
613 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (ops FR32:$dst, f128mem:$src),
614 "movaps {$src, $dst|$dst, $src}",
615 [(set FR32:$dst, (X86loadpf32 addr:$src))]>;
616 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (ops FR64:$dst, f128mem:$src),
617 "movapd {$src, $dst|$dst, $src}",
618 [(set FR64:$dst, (X86loadpf64 addr:$src))]>;
620 // Alias bitwise logical operations using SSE logical ops on packed FP values.
621 let isTwoAddress = 1 in {
622 let isCommutable = 1 in {
623 def FsANDPSrr : PSI<0x54, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
624 "andps {$src2, $dst|$dst, $src2}",
625 [(set FR32:$dst, (X86fand FR32:$src1, FR32:$src2))]>;
626 def FsANDPDrr : PDI<0x54, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
627 "andpd {$src2, $dst|$dst, $src2}",
628 [(set FR64:$dst, (X86fand FR64:$src1, FR64:$src2))]>;
629 def FsORPSrr : PSI<0x56, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
630 "orps {$src2, $dst|$dst, $src2}", []>;
631 def FsORPDrr : PDI<0x56, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
632 "orpd {$src2, $dst|$dst, $src2}", []>;
633 def FsXORPSrr : PSI<0x57, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
634 "xorps {$src2, $dst|$dst, $src2}",
635 [(set FR32:$dst, (X86fxor FR32:$src1, FR32:$src2))]>;
636 def FsXORPDrr : PDI<0x57, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
637 "xorpd {$src2, $dst|$dst, $src2}",
638 [(set FR64:$dst, (X86fxor FR64:$src1, FR64:$src2))]>;
640 def FsANDPSrm : PSI<0x54, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
641 "andps {$src2, $dst|$dst, $src2}",
642 [(set FR32:$dst, (X86fand FR32:$src1,
643 (X86loadpf32 addr:$src2)))]>;
644 def FsANDPDrm : PDI<0x54, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
645 "andpd {$src2, $dst|$dst, $src2}",
646 [(set FR64:$dst, (X86fand FR64:$src1,
647 (X86loadpf64 addr:$src2)))]>;
648 def FsORPSrm : PSI<0x56, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
649 "orps {$src2, $dst|$dst, $src2}", []>;
650 def FsORPDrm : PDI<0x56, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
651 "orpd {$src2, $dst|$dst, $src2}", []>;
652 def FsXORPSrm : PSI<0x57, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
653 "xorps {$src2, $dst|$dst, $src2}",
654 [(set FR32:$dst, (X86fxor FR32:$src1,
655 (X86loadpf32 addr:$src2)))]>;
656 def FsXORPDrm : PDI<0x57, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
657 "xorpd {$src2, $dst|$dst, $src2}",
658 [(set FR64:$dst, (X86fxor FR64:$src1,
659 (X86loadpf64 addr:$src2)))]>;
661 def FsANDNPSrr : PSI<0x55, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
662 "andnps {$src2, $dst|$dst, $src2}", []>;
663 def FsANDNPSrm : PSI<0x55, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
664 "andnps {$src2, $dst|$dst, $src2}", []>;
665 def FsANDNPDrr : PDI<0x55, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
666 "andnpd {$src2, $dst|$dst, $src2}", []>;
667 def FsANDNPDrm : PDI<0x55, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
668 "andnpd {$src2, $dst|$dst, $src2}", []>;
671 //===----------------------------------------------------------------------===//
672 // SSE packed FP Instructions
673 //===----------------------------------------------------------------------===//
675 // Some 'special' instructions
676 def IMPLICIT_DEF_VR128 : I<0, Pseudo, (ops VR128:$dst),
677 "#IMPLICIT_DEF $dst",
678 [(set VR128:$dst, (v4f32 (undef)))]>,
682 def MOVAPSrr : PSI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
683 "movaps {$src, $dst|$dst, $src}", []>;
684 def MOVAPSrm : PSI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
685 "movaps {$src, $dst|$dst, $src}",
686 [(set VR128:$dst, (loadv4f32 addr:$src))]>;
687 def MOVAPDrr : PDI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
688 "movapd {$src, $dst|$dst, $src}", []>;
689 def MOVAPDrm : PDI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
690 "movapd {$src, $dst|$dst, $src}",
691 [(set VR128:$dst, (loadv2f64 addr:$src))]>;
693 def MOVAPSmr : PSI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
694 "movaps {$src, $dst|$dst, $src}",
695 [(store (v4f32 VR128:$src), addr:$dst)]>;
696 def MOVAPDmr : PDI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
697 "movapd {$src, $dst|$dst, $src}",
698 [(store (v2f64 VR128:$src), addr:$dst)]>;
700 def MOVUPSrr : PSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
701 "movups {$src, $dst|$dst, $src}", []>;
702 def MOVUPSrm : PSI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
703 "movups {$src, $dst|$dst, $src}",
704 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
705 def MOVUPSmr : PSI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
706 "movups {$src, $dst|$dst, $src}",
707 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
708 def MOVUPDrr : PDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
709 "movupd {$src, $dst|$dst, $src}", []>;
710 def MOVUPDrm : PDI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
711 "movupd {$src, $dst|$dst, $src}",
712 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
713 def MOVUPDmr : PDI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
714 "movupd {$src, $dst|$dst, $src}",
715 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
717 let isTwoAddress = 1 in {
718 let AddedComplexity = 20 in {
719 def MOVLPSrm : PSI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
720 "movlps {$src2, $dst|$dst, $src2}",
722 (v4f32 (vector_shuffle VR128:$src1,
723 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
724 MOVLP_shuffle_mask)))]>;
725 def MOVLPDrm : PDI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
726 "movlpd {$src2, $dst|$dst, $src2}",
728 (v2f64 (vector_shuffle VR128:$src1,
729 (scalar_to_vector (loadf64 addr:$src2)),
730 MOVLP_shuffle_mask)))]>;
731 def MOVHPSrm : PSI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
732 "movhps {$src2, $dst|$dst, $src2}",
734 (v4f32 (vector_shuffle VR128:$src1,
735 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
736 MOVHP_shuffle_mask)))]>;
737 def MOVHPDrm : PDI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
738 "movhpd {$src2, $dst|$dst, $src2}",
740 (v2f64 (vector_shuffle VR128:$src1,
741 (scalar_to_vector (loadf64 addr:$src2)),
742 MOVHP_shuffle_mask)))]>;
746 def MOVLPSmr : PSI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
747 "movlps {$src, $dst|$dst, $src}",
748 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
749 (iPTR 0))), addr:$dst)]>;
750 def MOVLPDmr : PDI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
751 "movlpd {$src, $dst|$dst, $src}",
752 [(store (f64 (vector_extract (v2f64 VR128:$src),
753 (iPTR 0))), addr:$dst)]>;
755 // v2f64 extract element 1 is always custom lowered to unpack high to low
756 // and extract element 0 so the non-store version isn't too horrible.
757 def MOVHPSmr : PSI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
758 "movhps {$src, $dst|$dst, $src}",
759 [(store (f64 (vector_extract
760 (v2f64 (vector_shuffle
761 (bc_v2f64 (v4f32 VR128:$src)), (undef),
762 UNPCKH_shuffle_mask)), (iPTR 0))),
764 def MOVHPDmr : PDI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
765 "movhpd {$src, $dst|$dst, $src}",
766 [(store (f64 (vector_extract
767 (v2f64 (vector_shuffle VR128:$src, (undef),
768 UNPCKH_shuffle_mask)), (iPTR 0))),
771 let isTwoAddress = 1 in {
772 let AddedComplexity = 20 in {
773 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
774 "movlhps {$src2, $dst|$dst, $src2}",
776 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
777 MOVHP_shuffle_mask)))]>;
779 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
780 "movhlps {$src2, $dst|$dst, $src2}",
782 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
783 MOVHLPS_shuffle_mask)))]>;
787 def MOVSHDUPrr : S3SI<0x16, MRMSrcReg, (ops VR128:$dst, VR128:$src),
788 "movshdup {$src, $dst|$dst, $src}",
789 [(set VR128:$dst, (v4f32 (vector_shuffle
791 MOVSHDUP_shuffle_mask)))]>;
792 def MOVSHDUPrm : S3SI<0x16, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
793 "movshdup {$src, $dst|$dst, $src}",
794 [(set VR128:$dst, (v4f32 (vector_shuffle
795 (loadv4f32 addr:$src), (undef),
796 MOVSHDUP_shuffle_mask)))]>;
798 def MOVSLDUPrr : S3SI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src),
799 "movsldup {$src, $dst|$dst, $src}",
800 [(set VR128:$dst, (v4f32 (vector_shuffle
802 MOVSLDUP_shuffle_mask)))]>;
803 def MOVSLDUPrm : S3SI<0x12, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
804 "movsldup {$src, $dst|$dst, $src}",
805 [(set VR128:$dst, (v4f32 (vector_shuffle
806 (loadv4f32 addr:$src), (undef),
807 MOVSLDUP_shuffle_mask)))]>;
809 def MOVDDUPrr : S3DI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src),
810 "movddup {$src, $dst|$dst, $src}",
811 [(set VR128:$dst, (v2f64 (vector_shuffle
813 SSE_splat_v2_mask)))]>;
814 def MOVDDUPrm : S3DI<0x12, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
815 "movddup {$src, $dst|$dst, $src}",
816 [(set VR128:$dst, (v2f64 (vector_shuffle
817 (scalar_to_vector (loadf64 addr:$src)),
819 SSE_splat_v2_mask)))]>;
821 // SSE2 instructions without OpSize prefix
822 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
823 "cvtdq2ps {$src, $dst|$dst, $src}",
824 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
825 TB, Requires<[HasSSE2]>;
826 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
827 "cvtdq2ps {$src, $dst|$dst, $src}",
828 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
829 (bitconvert (loadv2i64 addr:$src))))]>,
830 TB, Requires<[HasSSE2]>;
832 // SSE2 instructions with XS prefix
833 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
834 "cvtdq2pd {$src, $dst|$dst, $src}",
835 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
836 XS, Requires<[HasSSE2]>;
837 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
838 "cvtdq2pd {$src, $dst|$dst, $src}",
839 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
840 (bitconvert (loadv2i64 addr:$src))))]>,
841 XS, Requires<[HasSSE2]>;
843 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
844 "cvtps2dq {$src, $dst|$dst, $src}",
845 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
846 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
847 "cvtps2dq {$src, $dst|$dst, $src}",
848 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
849 (load addr:$src)))]>;
850 // SSE2 packed instructions with XS prefix
851 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
852 "cvttps2dq {$src, $dst|$dst, $src}",
853 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))]>,
854 XS, Requires<[HasSSE2]>;
855 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
856 "cvttps2dq {$src, $dst|$dst, $src}",
857 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
858 (load addr:$src)))]>,
859 XS, Requires<[HasSSE2]>;
861 // SSE2 packed instructions with XD prefix
862 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
863 "cvtpd2dq {$src, $dst|$dst, $src}",
864 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
865 XD, Requires<[HasSSE2]>;
866 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
867 "cvtpd2dq {$src, $dst|$dst, $src}",
868 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
869 (load addr:$src)))]>,
870 XD, Requires<[HasSSE2]>;
871 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
872 "cvttpd2dq {$src, $dst|$dst, $src}",
873 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
874 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
875 "cvttpd2dq {$src, $dst|$dst, $src}",
876 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
877 (load addr:$src)))]>;
879 // SSE2 instructions without OpSize prefix
880 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
881 "cvtps2pd {$src, $dst|$dst, $src}",
882 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
883 TB, Requires<[HasSSE2]>;
884 def Int_CVTPS2PDrm : I<0x5A, MRMSrcReg, (ops VR128:$dst, f64mem:$src),
885 "cvtps2pd {$src, $dst|$dst, $src}",
886 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
887 (load addr:$src)))]>,
888 TB, Requires<[HasSSE2]>;
890 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
891 "cvtpd2ps {$src, $dst|$dst, $src}",
892 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
893 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, f128mem:$src),
894 "cvtpd2ps {$src, $dst|$dst, $src}",
895 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
896 (load addr:$src)))]>;
898 // Match intrinsics which expect XMM operand(s).
899 // Aliases for intrinsics
900 let isTwoAddress = 1 in {
901 def Int_CVTSI2SDrr: SDI<0x2A, MRMSrcReg,
902 (ops VR128:$dst, VR128:$src1, GR32:$src2),
903 "cvtsi2sd {$src2, $dst|$dst, $src2}",
904 [(set VR128:$dst, (int_x86_sse2_cvtsi2sd VR128:$src1,
906 def Int_CVTSI2SDrm: SDI<0x2A, MRMSrcMem,
907 (ops VR128:$dst, VR128:$src1, i32mem:$src2),
908 "cvtsi2sd {$src2, $dst|$dst, $src2}",
909 [(set VR128:$dst, (int_x86_sse2_cvtsi2sd VR128:$src1,
910 (loadi32 addr:$src2)))]>;
911 def Int_CVTSD2SSrr: SDI<0x5A, MRMSrcReg,
912 (ops VR128:$dst, VR128:$src1, VR128:$src2),
913 "cvtsd2ss {$src2, $dst|$dst, $src2}",
914 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
916 def Int_CVTSD2SSrm: SDI<0x5A, MRMSrcMem,
917 (ops VR128:$dst, VR128:$src1, f64mem:$src2),
918 "cvtsd2ss {$src2, $dst|$dst, $src2}",
919 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
920 (load addr:$src2)))]>;
921 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
922 (ops VR128:$dst, VR128:$src1, VR128:$src2),
923 "cvtss2sd {$src2, $dst|$dst, $src2}",
924 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
927 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
928 (ops VR128:$dst, VR128:$src1, f32mem:$src2),
929 "cvtss2sd {$src2, $dst|$dst, $src2}",
930 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
931 (load addr:$src2)))]>, XS,
935 /// packed_sse12_fp_binop_rm - Packed SSE binops come in four basic forms:
936 /// 1. v4f32 vs v2f64 - These come in SSE1/SSE2 forms for float/doubles.
937 /// 2. rr vs rm - They include a reg+reg form and a ref+mem form.
939 let isTwoAddress = 1 in {
940 multiclass packed_sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
941 SDNode OpNode, bit Commutable = 0> {
942 // Packed operation, reg+reg.
943 def PSrr : PSI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
944 !strconcat(OpcodeStr, "ps {$src2, $dst|$dst, $src2"),
945 [(set VR128:$dst, (v4f32 (OpNode VR128:$src1, VR128:$src2)))]> {
946 let isCommutable = Commutable;
948 def PDrr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
949 !strconcat(OpcodeStr, "pd {$src2, $dst|$dst, $src2"),
950 [(set VR128:$dst, (v2f64 (OpNode VR128:$src1, VR128:$src2)))]> {
951 let isCommutable = Commutable;
953 // Packed operation, reg+mem.
954 def PSrm : PSI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
955 !strconcat(OpcodeStr, "ps {$src2, $dst|$dst, $src2"),
956 [(set VR128:$dst, (OpNode VR128:$src1, (loadv4f32 addr:$src2)))]>;
957 def PDrm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
958 !strconcat(OpcodeStr, "pd {$src2, $dst|$dst, $src2"),
959 [(set VR128:$dst, (OpNode VR128:$src1, (loadv2f64 addr:$src2)))]>;
963 defm ADD : packed_sse12_fp_binop_rm<0x58, "add", fadd, 1>;
964 defm MUL : packed_sse12_fp_binop_rm<0x59, "mul", fmul, 1>;
965 defm DIV : packed_sse12_fp_binop_rm<0x5E, "div", fdiv>;
966 defm SUB : packed_sse12_fp_binop_rm<0x5C, "sub", fsub>;
969 let isTwoAddress = 1 in {
970 def ADDSUBPSrr : S3DI<0xD0, MRMSrcReg,
971 (ops VR128:$dst, VR128:$src1, VR128:$src2),
972 "addsubps {$src2, $dst|$dst, $src2}",
973 [(set VR128:$dst, (int_x86_sse3_addsub_ps VR128:$src1,
975 def ADDSUBPSrm : S3DI<0xD0, MRMSrcMem,
976 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
977 "addsubps {$src2, $dst|$dst, $src2}",
978 [(set VR128:$dst, (int_x86_sse3_addsub_ps VR128:$src1,
979 (load addr:$src2)))]>;
980 def ADDSUBPDrr : S3I<0xD0, MRMSrcReg,
981 (ops VR128:$dst, VR128:$src1, VR128:$src2),
982 "addsubpd {$src2, $dst|$dst, $src2}",
983 [(set VR128:$dst, (int_x86_sse3_addsub_pd VR128:$src1,
985 def ADDSUBPDrm : S3I<0xD0, MRMSrcMem,
986 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
987 "addsubpd {$src2, $dst|$dst, $src2}",
988 [(set VR128:$dst, (int_x86_sse3_addsub_pd VR128:$src1,
989 (load addr:$src2)))]>;
992 def SQRTPSr : PS_Intr<0x51, "sqrtps", int_x86_sse_sqrt_ps>;
993 def SQRTPSm : PS_Intm<0x51, "sqrtps", int_x86_sse_sqrt_ps>;
994 def SQRTPDr : PD_Intr<0x51, "sqrtpd", int_x86_sse2_sqrt_pd>;
995 def SQRTPDm : PD_Intm<0x51, "sqrtpd", int_x86_sse2_sqrt_pd>;
997 def RSQRTPSr : PS_Intr<0x52, "rsqrtps", int_x86_sse_rsqrt_ps>;
998 def RSQRTPSm : PS_Intm<0x52, "rsqrtps", int_x86_sse_rsqrt_ps>;
999 def RCPPSr : PS_Intr<0x53, "rcpps", int_x86_sse_rcp_ps>;
1000 def RCPPSm : PS_Intm<0x53, "rcpps", int_x86_sse_rcp_ps>;
1002 let isTwoAddress = 1 in {
1003 let isCommutable = 1 in {
1004 def MAXPSrr : PS_Intrr<0x5F, "maxps", int_x86_sse_max_ps>;
1005 def MAXPDrr : PD_Intrr<0x5F, "maxpd", int_x86_sse2_max_pd>;
1006 def MINPSrr : PS_Intrr<0x5D, "minps", int_x86_sse_min_ps>;
1007 def MINPDrr : PD_Intrr<0x5D, "minpd", int_x86_sse2_min_pd>;
1009 def MAXPSrm : PS_Intrm<0x5F, "maxps", int_x86_sse_max_ps>;
1010 def MAXPDrm : PD_Intrm<0x5F, "maxpd", int_x86_sse2_max_pd>;
1011 def MINPSrm : PS_Intrm<0x5D, "minps", int_x86_sse_min_ps>;
1012 def MINPDrm : PD_Intrm<0x5D, "minpd", int_x86_sse2_min_pd>;
1016 let isTwoAddress = 1 in {
1017 let isCommutable = 1 in {
1018 def ANDPSrr : PSI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1019 "andps {$src2, $dst|$dst, $src2}",
1020 [(set VR128:$dst, (v2i64 (and VR128:$src1, VR128:$src2)))]>;
1021 def ANDPDrr : PDI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1022 "andpd {$src2, $dst|$dst, $src2}",
1024 (and (bc_v2i64 (v2f64 VR128:$src1)),
1025 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1026 def ORPSrr : PSI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1027 "orps {$src2, $dst|$dst, $src2}",
1028 [(set VR128:$dst, (v2i64 (or VR128:$src1, VR128:$src2)))]>;
1029 def ORPDrr : PDI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1030 "orpd {$src2, $dst|$dst, $src2}",
1032 (or (bc_v2i64 (v2f64 VR128:$src1)),
1033 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1034 def XORPSrr : PSI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1035 "xorps {$src2, $dst|$dst, $src2}",
1036 [(set VR128:$dst, (v2i64 (xor VR128:$src1, VR128:$src2)))]>;
1037 def XORPDrr : PDI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1038 "xorpd {$src2, $dst|$dst, $src2}",
1040 (xor (bc_v2i64 (v2f64 VR128:$src1)),
1041 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1043 def ANDPSrm : PSI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1044 "andps {$src2, $dst|$dst, $src2}",
1045 [(set VR128:$dst, (and VR128:$src1,
1046 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1047 def ANDPDrm : PDI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1048 "andpd {$src2, $dst|$dst, $src2}",
1050 (and (bc_v2i64 (v2f64 VR128:$src1)),
1051 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1052 def ORPSrm : PSI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1053 "orps {$src2, $dst|$dst, $src2}",
1054 [(set VR128:$dst, (or VR128:$src1,
1055 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1056 def ORPDrm : PDI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1057 "orpd {$src2, $dst|$dst, $src2}",
1059 (or (bc_v2i64 (v2f64 VR128:$src1)),
1060 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1061 def XORPSrm : PSI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1062 "xorps {$src2, $dst|$dst, $src2}",
1063 [(set VR128:$dst, (xor VR128:$src1,
1064 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1065 def XORPDrm : PDI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1066 "xorpd {$src2, $dst|$dst, $src2}",
1068 (xor (bc_v2i64 (v2f64 VR128:$src1)),
1069 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1070 def ANDNPSrr : PSI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1071 "andnps {$src2, $dst|$dst, $src2}",
1072 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1073 (bc_v2i64 (v4i32 immAllOnesV))),
1075 def ANDNPSrm : PSI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
1076 "andnps {$src2, $dst|$dst, $src2}",
1077 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1078 (bc_v2i64 (v4i32 immAllOnesV))),
1079 (bc_v2i64 (loadv4f32 addr:$src2)))))]>;
1080 def ANDNPDrr : PDI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1081 "andnpd {$src2, $dst|$dst, $src2}",
1083 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1084 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1085 def ANDNPDrm : PDI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
1086 "andnpd {$src2, $dst|$dst, $src2}",
1088 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1089 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1092 let isTwoAddress = 1 in {
1093 def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
1094 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
1095 "cmp${cc}ps {$src, $dst|$dst, $src}",
1096 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
1097 VR128:$src, imm:$cc))]>;
1098 def CMPPSrmi : PSIi8<0xC2, MRMSrcMem,
1099 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
1100 "cmp${cc}ps {$src, $dst|$dst, $src}",
1101 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
1102 (load addr:$src), imm:$cc))]>;
1103 def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
1104 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
1105 "cmp${cc}pd {$src, $dst|$dst, $src}",
1106 [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
1107 VR128:$src, imm:$cc))]>;
1108 def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
1109 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
1110 "cmp${cc}pd {$src, $dst|$dst, $src}",
1111 [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
1112 (load addr:$src), imm:$cc))]>;
1115 // Shuffle and unpack instructions
1116 let isTwoAddress = 1 in {
1117 let isConvertibleToThreeAddress = 1 in // Convert to pshufd
1118 def SHUFPSrri : PSIi8<0xC6, MRMSrcReg,
1119 (ops VR128:$dst, VR128:$src1, VR128:$src2, i32i8imm:$src3),
1120 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1121 [(set VR128:$dst, (v4f32 (vector_shuffle
1122 VR128:$src1, VR128:$src2,
1123 SHUFP_shuffle_mask:$src3)))]>;
1124 def SHUFPSrmi : PSIi8<0xC6, MRMSrcMem,
1125 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i32i8imm:$src3),
1126 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1127 [(set VR128:$dst, (v4f32 (vector_shuffle
1128 VR128:$src1, (load addr:$src2),
1129 SHUFP_shuffle_mask:$src3)))]>;
1130 def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
1131 (ops VR128:$dst, VR128:$src1, VR128:$src2, i8imm:$src3),
1132 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1133 [(set VR128:$dst, (v2f64 (vector_shuffle
1134 VR128:$src1, VR128:$src2,
1135 SHUFP_shuffle_mask:$src3)))]>;
1136 def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
1137 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i8imm:$src3),
1138 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1139 [(set VR128:$dst, (v2f64 (vector_shuffle
1140 VR128:$src1, (load addr:$src2),
1141 SHUFP_shuffle_mask:$src3)))]>;
1143 let AddedComplexity = 10 in {
1144 def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
1145 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1146 "unpckhps {$src2, $dst|$dst, $src2}",
1147 [(set VR128:$dst, (v4f32 (vector_shuffle
1148 VR128:$src1, VR128:$src2,
1149 UNPCKH_shuffle_mask)))]>;
1150 def UNPCKHPSrm : PSI<0x15, MRMSrcMem,
1151 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1152 "unpckhps {$src2, $dst|$dst, $src2}",
1153 [(set VR128:$dst, (v4f32 (vector_shuffle
1154 VR128:$src1, (load addr:$src2),
1155 UNPCKH_shuffle_mask)))]>;
1156 def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
1157 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1158 "unpckhpd {$src2, $dst|$dst, $src2}",
1159 [(set VR128:$dst, (v2f64 (vector_shuffle
1160 VR128:$src1, VR128:$src2,
1161 UNPCKH_shuffle_mask)))]>;
1162 def UNPCKHPDrm : PDI<0x15, MRMSrcMem,
1163 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1164 "unpckhpd {$src2, $dst|$dst, $src2}",
1165 [(set VR128:$dst, (v2f64 (vector_shuffle
1166 VR128:$src1, (load addr:$src2),
1167 UNPCKH_shuffle_mask)))]>;
1169 def UNPCKLPSrr : PSI<0x14, MRMSrcReg,
1170 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1171 "unpcklps {$src2, $dst|$dst, $src2}",
1172 [(set VR128:$dst, (v4f32 (vector_shuffle
1173 VR128:$src1, VR128:$src2,
1174 UNPCKL_shuffle_mask)))]>;
1175 def UNPCKLPSrm : PSI<0x14, MRMSrcMem,
1176 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1177 "unpcklps {$src2, $dst|$dst, $src2}",
1178 [(set VR128:$dst, (v4f32 (vector_shuffle
1179 VR128:$src1, (load addr:$src2),
1180 UNPCKL_shuffle_mask)))]>;
1181 def UNPCKLPDrr : PDI<0x14, MRMSrcReg,
1182 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1183 "unpcklpd {$src2, $dst|$dst, $src2}",
1184 [(set VR128:$dst, (v2f64 (vector_shuffle
1185 VR128:$src1, VR128:$src2,
1186 UNPCKL_shuffle_mask)))]>;
1187 def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
1188 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1189 "unpcklpd {$src2, $dst|$dst, $src2}",
1190 [(set VR128:$dst, (v2f64 (vector_shuffle
1191 VR128:$src1, (load addr:$src2),
1192 UNPCKL_shuffle_mask)))]>;
1193 } // AddedComplexity
1198 class S3D_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
1199 : S3DI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1200 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1201 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, VR128:$src2)))]>;
1202 class S3D_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
1203 : S3DI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1204 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1205 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, (load addr:$src2))))]>;
1206 class S3_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
1207 : S3I<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1208 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1209 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, VR128:$src2)))]>;
1210 class S3_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
1211 : S3I<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1212 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1213 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, (load addr:$src2))))]>;
1215 let isTwoAddress = 1 in {
1216 def HADDPSrr : S3D_Intrr<0x7C, "haddps", int_x86_sse3_hadd_ps>;
1217 def HADDPSrm : S3D_Intrm<0x7C, "haddps", int_x86_sse3_hadd_ps>;
1218 def HADDPDrr : S3_Intrr <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
1219 def HADDPDrm : S3_Intrm <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
1220 def HSUBPSrr : S3D_Intrr<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
1221 def HSUBPSrm : S3D_Intrm<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
1222 def HSUBPDrr : S3_Intrr <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
1223 def HSUBPDrm : S3_Intrm <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
1226 //===----------------------------------------------------------------------===//
1227 // SSE integer instructions
1228 //===----------------------------------------------------------------------===//
1230 // Move Instructions
1231 def MOVDQArr : PDI<0x6F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
1232 "movdqa {$src, $dst|$dst, $src}", []>;
1233 def MOVDQArm : PDI<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1234 "movdqa {$src, $dst|$dst, $src}",
1235 [(set VR128:$dst, (loadv2i64 addr:$src))]>;
1236 def MOVDQAmr : PDI<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1237 "movdqa {$src, $dst|$dst, $src}",
1238 [(store (v2i64 VR128:$src), addr:$dst)]>;
1239 def MOVDQUrm : I<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1240 "movdqu {$src, $dst|$dst, $src}",
1241 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
1242 XS, Requires<[HasSSE2]>;
1243 def MOVDQUmr : I<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1244 "movdqu {$src, $dst|$dst, $src}",
1245 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
1246 XS, Requires<[HasSSE2]>;
1247 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1248 "lddqu {$src, $dst|$dst, $src}",
1249 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
1252 let isTwoAddress = 1 in {
1253 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
1254 bit Commutable = 0> {
1255 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1256 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1257 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]> {
1258 let isCommutable = Commutable;
1260 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1261 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1262 [(set VR128:$dst, (IntId VR128:$src1,
1263 (bitconvert (loadv2i64 addr:$src2))))]>;
1267 let isTwoAddress = 1 in {
1268 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
1269 string OpcodeStr, Intrinsic IntId> {
1270 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1271 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1272 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
1273 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1274 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1275 [(set VR128:$dst, (IntId VR128:$src1,
1276 (bitconvert (loadv2i64 addr:$src2))))]>;
1277 def ri : PDIi8<opc2, ImmForm, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1278 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1279 [(set VR128:$dst, (IntId VR128:$src1,
1280 (scalar_to_vector (i32 imm:$src2))))]>;
1285 let isTwoAddress = 1 in {
1286 /// PDI_binop_rm - Simple SSE2 binary operator.
1287 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
1288 ValueType OpVT, bit Commutable = 0> {
1289 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1290 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1291 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]> {
1292 let isCommutable = Commutable;
1294 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1295 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1296 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
1297 (bitconvert (loadv2i64 addr:$src2)))))]>;
1300 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
1302 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
1303 /// to collapse (bitconvert VT to VT) into its operand.
1305 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
1306 bit Commutable = 0> {
1307 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1308 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1309 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]> {
1310 let isCommutable = Commutable;
1312 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1313 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1314 [(set VR128:$dst, (OpNode VR128:$src1,(loadv2i64 addr:$src2)))]>;
1319 // 128-bit Integer Arithmetic
1321 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
1322 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
1323 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
1324 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
1326 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
1327 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
1328 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
1329 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
1331 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
1332 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
1333 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
1334 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
1336 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
1337 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
1338 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
1339 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
1341 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
1343 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
1344 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w , 1>;
1345 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
1347 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
1349 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
1350 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
1353 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
1354 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
1355 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
1356 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
1357 defm PSADBW : PDI_binop_rm_int<0xE0, "psadbw", int_x86_sse2_psad_bw, 1>;
1360 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw", int_x86_sse2_psll_w>;
1361 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld", int_x86_sse2_psll_d>;
1362 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq", int_x86_sse2_psll_q>;
1364 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw", int_x86_sse2_psrl_w>;
1365 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld", int_x86_sse2_psrl_d>;
1366 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq", int_x86_sse2_psrl_q>;
1368 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw", int_x86_sse2_psra_w>;
1369 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad", int_x86_sse2_psra_d>;
1370 // PSRAQ doesn't exist in SSE[1-3].
1373 // 128-bit logical shifts.
1374 let isTwoAddress = 1 in {
1375 def PSLLDQri : PDIi8<0x73, MRM7r, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1376 "pslldq {$src2, $dst|$dst, $src2}", []>;
1377 def PSRLDQri : PDIi8<0x73, MRM3r, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1378 "psrldq {$src2, $dst|$dst, $src2}", []>;
1379 // PSRADQri doesn't exist in SSE[1-3].
1382 let Predicates = [HasSSE2] in {
1383 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
1384 (v2i64 (PSLLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
1385 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
1386 (v2i64 (PSRLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
1390 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
1391 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or , 1>;
1392 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
1394 let isTwoAddress = 1 in {
1395 def PANDNrr : PDI<0xDF, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1396 "pandn {$src2, $dst|$dst, $src2}",
1397 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1400 def PANDNrm : PDI<0xDF, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1401 "pandn {$src2, $dst|$dst, $src2}",
1402 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1403 (load addr:$src2))))]>;
1406 // SSE2 Integer comparison
1407 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b>;
1408 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w>;
1409 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d>;
1410 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
1411 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
1412 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
1414 // Pack instructions
1415 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
1416 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
1417 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
1419 // Shuffle and unpack instructions
1420 def PSHUFDri : PDIi8<0x70, MRMSrcReg,
1421 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1422 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1423 [(set VR128:$dst, (v4i32 (vector_shuffle
1424 VR128:$src1, (undef),
1425 PSHUFD_shuffle_mask:$src2)))]>;
1426 def PSHUFDmi : PDIi8<0x70, MRMSrcMem,
1427 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1428 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1429 [(set VR128:$dst, (v4i32 (vector_shuffle
1430 (bc_v4i32(loadv2i64 addr:$src1)),
1432 PSHUFD_shuffle_mask:$src2)))]>;
1434 // SSE2 with ImmT == Imm8 and XS prefix.
1435 def PSHUFHWri : Ii8<0x70, MRMSrcReg,
1436 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1437 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1438 [(set VR128:$dst, (v8i16 (vector_shuffle
1439 VR128:$src1, (undef),
1440 PSHUFHW_shuffle_mask:$src2)))]>,
1441 XS, Requires<[HasSSE2]>;
1442 def PSHUFHWmi : Ii8<0x70, MRMSrcMem,
1443 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1444 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1445 [(set VR128:$dst, (v8i16 (vector_shuffle
1446 (bc_v8i16 (loadv2i64 addr:$src1)),
1448 PSHUFHW_shuffle_mask:$src2)))]>,
1449 XS, Requires<[HasSSE2]>;
1451 // SSE2 with ImmT == Imm8 and XD prefix.
1452 def PSHUFLWri : Ii8<0x70, MRMSrcReg,
1453 (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1454 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1455 [(set VR128:$dst, (v8i16 (vector_shuffle
1456 VR128:$src1, (undef),
1457 PSHUFLW_shuffle_mask:$src2)))]>,
1458 XD, Requires<[HasSSE2]>;
1459 def PSHUFLWmi : Ii8<0x70, MRMSrcMem,
1460 (ops VR128:$dst, i128mem:$src1, i32i8imm:$src2),
1461 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1462 [(set VR128:$dst, (v8i16 (vector_shuffle
1463 (bc_v8i16 (loadv2i64 addr:$src1)),
1465 PSHUFLW_shuffle_mask:$src2)))]>,
1466 XD, Requires<[HasSSE2]>;
1468 let isTwoAddress = 1 in {
1469 def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
1470 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1471 "punpcklbw {$src2, $dst|$dst, $src2}",
1473 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1474 UNPCKL_shuffle_mask)))]>;
1475 def PUNPCKLBWrm : PDI<0x60, MRMSrcMem,
1476 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1477 "punpcklbw {$src2, $dst|$dst, $src2}",
1479 (v16i8 (vector_shuffle VR128:$src1,
1480 (bc_v16i8 (loadv2i64 addr:$src2)),
1481 UNPCKL_shuffle_mask)))]>;
1482 def PUNPCKLWDrr : PDI<0x61, MRMSrcReg,
1483 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1484 "punpcklwd {$src2, $dst|$dst, $src2}",
1486 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1487 UNPCKL_shuffle_mask)))]>;
1488 def PUNPCKLWDrm : PDI<0x61, MRMSrcMem,
1489 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1490 "punpcklwd {$src2, $dst|$dst, $src2}",
1492 (v8i16 (vector_shuffle VR128:$src1,
1493 (bc_v8i16 (loadv2i64 addr:$src2)),
1494 UNPCKL_shuffle_mask)))]>;
1495 def PUNPCKLDQrr : PDI<0x62, MRMSrcReg,
1496 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1497 "punpckldq {$src2, $dst|$dst, $src2}",
1499 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1500 UNPCKL_shuffle_mask)))]>;
1501 def PUNPCKLDQrm : PDI<0x62, MRMSrcMem,
1502 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1503 "punpckldq {$src2, $dst|$dst, $src2}",
1505 (v4i32 (vector_shuffle VR128:$src1,
1506 (bc_v4i32 (loadv2i64 addr:$src2)),
1507 UNPCKL_shuffle_mask)))]>;
1508 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
1509 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1510 "punpcklqdq {$src2, $dst|$dst, $src2}",
1512 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1513 UNPCKL_shuffle_mask)))]>;
1514 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
1515 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1516 "punpcklqdq {$src2, $dst|$dst, $src2}",
1518 (v2i64 (vector_shuffle VR128:$src1,
1519 (loadv2i64 addr:$src2),
1520 UNPCKL_shuffle_mask)))]>;
1522 def PUNPCKHBWrr : PDI<0x68, MRMSrcReg,
1523 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1524 "punpckhbw {$src2, $dst|$dst, $src2}",
1526 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1527 UNPCKH_shuffle_mask)))]>;
1528 def PUNPCKHBWrm : PDI<0x68, MRMSrcMem,
1529 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1530 "punpckhbw {$src2, $dst|$dst, $src2}",
1532 (v16i8 (vector_shuffle VR128:$src1,
1533 (bc_v16i8 (loadv2i64 addr:$src2)),
1534 UNPCKH_shuffle_mask)))]>;
1535 def PUNPCKHWDrr : PDI<0x69, MRMSrcReg,
1536 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1537 "punpckhwd {$src2, $dst|$dst, $src2}",
1539 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1540 UNPCKH_shuffle_mask)))]>;
1541 def PUNPCKHWDrm : PDI<0x69, MRMSrcMem,
1542 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1543 "punpckhwd {$src2, $dst|$dst, $src2}",
1545 (v8i16 (vector_shuffle VR128:$src1,
1546 (bc_v8i16 (loadv2i64 addr:$src2)),
1547 UNPCKH_shuffle_mask)))]>;
1548 def PUNPCKHDQrr : PDI<0x6A, MRMSrcReg,
1549 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1550 "punpckhdq {$src2, $dst|$dst, $src2}",
1552 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1553 UNPCKH_shuffle_mask)))]>;
1554 def PUNPCKHDQrm : PDI<0x6A, MRMSrcMem,
1555 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1556 "punpckhdq {$src2, $dst|$dst, $src2}",
1558 (v4i32 (vector_shuffle VR128:$src1,
1559 (bc_v4i32 (loadv2i64 addr:$src2)),
1560 UNPCKH_shuffle_mask)))]>;
1561 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
1562 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1563 "punpckhqdq {$src2, $dst|$dst, $src2}",
1565 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1566 UNPCKH_shuffle_mask)))]>;
1567 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
1568 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1569 "punpckhqdq {$src2, $dst|$dst, $src2}",
1571 (v2i64 (vector_shuffle VR128:$src1,
1572 (loadv2i64 addr:$src2),
1573 UNPCKH_shuffle_mask)))]>;
1577 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
1578 (ops GR32:$dst, VR128:$src1, i32i8imm:$src2),
1579 "pextrw {$src2, $src1, $dst|$dst, $src1, $src2}",
1580 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
1581 (i32 imm:$src2)))]>;
1582 let isTwoAddress = 1 in {
1583 def PINSRWrri : PDIi8<0xC4, MRMSrcReg,
1584 (ops VR128:$dst, VR128:$src1, GR32:$src2, i32i8imm:$src3),
1585 "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
1586 [(set VR128:$dst, (v8i16 (X86pinsrw (v8i16 VR128:$src1),
1587 GR32:$src2, (iPTR imm:$src3))))]>;
1588 def PINSRWrmi : PDIi8<0xC4, MRMSrcMem,
1589 (ops VR128:$dst, VR128:$src1, i16mem:$src2, i32i8imm:$src3),
1590 "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
1592 (v8i16 (X86pinsrw (v8i16 VR128:$src1),
1593 (i32 (anyext (loadi16 addr:$src2))),
1594 (iPTR imm:$src3))))]>;
1597 //===----------------------------------------------------------------------===//
1598 // Miscellaneous Instructions
1599 //===----------------------------------------------------------------------===//
1602 def MOVMSKPSrr : PSI<0x50, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1603 "movmskps {$src, $dst|$dst, $src}",
1604 [(set GR32:$dst, (int_x86_sse_movmsk_ps VR128:$src))]>;
1605 def MOVMSKPDrr : PSI<0x50, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1606 "movmskpd {$src, $dst|$dst, $src}",
1607 [(set GR32:$dst, (int_x86_sse2_movmsk_pd VR128:$src))]>;
1609 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1610 "pmovmskb {$src, $dst|$dst, $src}",
1611 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
1613 // Conditional store
1614 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (ops VR128:$src, VR128:$mask),
1615 "maskmovdqu {$mask, $src|$src, $mask}",
1616 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>,
1619 // Prefetching loads.
1620 // TODO: no intrinsics for these?
1621 def PREFETCHT0 : PSI<0x18, MRM1m, (ops i8mem:$src), "prefetcht0 $src", []>;
1622 def PREFETCHT1 : PSI<0x18, MRM2m, (ops i8mem:$src), "prefetcht1 $src", []>;
1623 def PREFETCHT2 : PSI<0x18, MRM3m, (ops i8mem:$src), "prefetcht2 $src", []>;
1624 def PREFETCHTNTA : PSI<0x18, MRM0m, (ops i8mem:$src), "prefetchtnta $src", []>;
1626 // Non-temporal stores
1627 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1628 "movntps {$src, $dst|$dst, $src}",
1629 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
1630 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1631 "movntpd {$src, $dst|$dst, $src}",
1632 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
1633 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (ops f128mem:$dst, VR128:$src),
1634 "movntdq {$src, $dst|$dst, $src}",
1635 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
1636 def MOVNTImr : I<0xC3, MRMDestMem, (ops i32mem:$dst, GR32:$src),
1637 "movnti {$src, $dst|$dst, $src}",
1638 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
1639 TB, Requires<[HasSSE2]>;
1642 def CLFLUSH : I<0xAE, MRM7m, (ops i8mem:$src),
1643 "clflush $src", [(int_x86_sse2_clflush addr:$src)]>,
1644 TB, Requires<[HasSSE2]>;
1646 // Load, store, and memory fence
1647 def SFENCE : PSI<0xAE, MRM7m, (ops), "sfence", [(int_x86_sse_sfence)]>;
1648 def LFENCE : I<0xAE, MRM5m, (ops),
1649 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
1650 def MFENCE : I<0xAE, MRM6m, (ops),
1651 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
1654 def LDMXCSR : I<0xAE, MRM5m, (ops i32mem:$src),
1656 [(int_x86_sse_ldmxcsr addr:$src)]>, TB, Requires<[HasSSE1]>;
1657 def STMXCSR : I<0xAE, MRM3m, (ops i32mem:$dst),
1659 [(int_x86_sse_stmxcsr addr:$dst)]>, TB, Requires<[HasSSE1]>;
1661 // Thread synchronization
1662 def MONITOR : I<0xC8, RawFrm, (ops), "monitor",
1663 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
1664 def MWAIT : I<0xC9, RawFrm, (ops), "mwait",
1665 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
1667 //===----------------------------------------------------------------------===//
1668 // Alias Instructions
1669 //===----------------------------------------------------------------------===//
1671 // Alias instructions that map zero vector to pxor / xorp* for sse.
1672 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
1673 def V_SET0 : PSI<0x57, MRMInitReg, (ops VR128:$dst),
1675 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
1677 def V_SETALLONES : PDI<0x76, MRMInitReg, (ops VR128:$dst),
1678 "pcmpeqd $dst, $dst",
1679 [(set VR128:$dst, (v2f64 immAllOnesV))]>;
1681 // FR32 / FR64 to 128-bit vector conversion.
1682 def MOVSS2PSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, FR32:$src),
1683 "movss {$src, $dst|$dst, $src}",
1685 (v4f32 (scalar_to_vector FR32:$src)))]>;
1686 def MOVSS2PSrm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
1687 "movss {$src, $dst|$dst, $src}",
1689 (v4f32 (scalar_to_vector (loadf32 addr:$src))))]>;
1690 def MOVSD2PDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, FR64:$src),
1691 "movsd {$src, $dst|$dst, $src}",
1693 (v2f64 (scalar_to_vector FR64:$src)))]>;
1694 def MOVSD2PDrm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
1695 "movsd {$src, $dst|$dst, $src}",
1697 (v2f64 (scalar_to_vector (loadf64 addr:$src))))]>;
1699 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, GR32:$src),
1700 "movd {$src, $dst|$dst, $src}",
1702 (v4i32 (scalar_to_vector GR32:$src)))]>;
1703 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1704 "movd {$src, $dst|$dst, $src}",
1706 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
1707 // SSE2 instructions with XS prefix
1708 def MOVQI2PQIrr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR64:$src),
1709 "movq {$src, $dst|$dst, $src}",
1711 (v2i64 (scalar_to_vector VR64:$src)))]>, XS,
1712 Requires<[HasSSE2]>;
1713 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
1714 "movq {$src, $dst|$dst, $src}",
1716 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
1717 Requires<[HasSSE2]>;
1718 // FIXME: may not be able to eliminate this movss with coalescing the src and
1719 // dest register classes are different. We really want to write this pattern
1721 // def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
1722 // (f32 FR32:$src)>;
1723 def MOVPS2SSrr : SSI<0x10, MRMSrcReg, (ops FR32:$dst, VR128:$src),
1724 "movss {$src, $dst|$dst, $src}",
1725 [(set FR32:$dst, (vector_extract (v4f32 VR128:$src),
1727 def MOVPS2SSmr : SSI<0x11, MRMDestMem, (ops f32mem:$dst, VR128:$src),
1728 "movss {$src, $dst|$dst, $src}",
1729 [(store (f32 (vector_extract (v4f32 VR128:$src),
1730 (iPTR 0))), addr:$dst)]>;
1731 def MOVPD2SDrr : SDI<0x10, MRMSrcReg, (ops FR64:$dst, VR128:$src),
1732 "movsd {$src, $dst|$dst, $src}",
1733 [(set FR64:$dst, (vector_extract (v2f64 VR128:$src),
1735 def MOVPD2SDmr : SDI<0x11, MRMDestMem, (ops f64mem:$dst, VR128:$src),
1736 "movsd {$src, $dst|$dst, $src}",
1737 [(store (f64 (vector_extract (v2f64 VR128:$src),
1738 (iPTR 0))), addr:$dst)]>;
1739 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (ops GR32:$dst, VR128:$src),
1740 "movd {$src, $dst|$dst, $src}",
1741 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
1743 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (ops i32mem:$dst, VR128:$src),
1744 "movd {$src, $dst|$dst, $src}",
1745 [(store (i32 (vector_extract (v4i32 VR128:$src),
1746 (iPTR 0))), addr:$dst)]>;
1748 // Move to lower bits of a VR128, leaving upper bits alone.
1749 // Three operand (but two address) aliases.
1750 let isTwoAddress = 1 in {
1751 def MOVLSS2PSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR32:$src2),
1752 "movss {$src2, $dst|$dst, $src2}", []>;
1753 def MOVLSD2PDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR64:$src2),
1754 "movsd {$src2, $dst|$dst, $src2}", []>;
1756 let AddedComplexity = 20 in {
1757 def MOVLPSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1758 "movss {$src2, $dst|$dst, $src2}",
1760 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
1761 MOVL_shuffle_mask)))]>;
1762 def MOVLPDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1763 "movsd {$src2, $dst|$dst, $src2}",
1765 (v2f64 (vector_shuffle VR128:$src1, VR128:$src2,
1766 MOVL_shuffle_mask)))]>;
1770 // Store / copy lower 64-bits of a XMM register.
1771 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (ops i64mem:$dst, VR128:$src),
1772 "movq {$src, $dst|$dst, $src}",
1773 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
1775 // Move to lower bits of a VR128 and zeroing upper bits.
1776 // Loading from memory automatically zeroing upper bits.
1777 let AddedComplexity = 20 in {
1778 def MOVZSS2PSrm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
1779 "movss {$src, $dst|$dst, $src}",
1780 [(set VR128:$dst, (v4f32 (vector_shuffle immAllZerosV,
1781 (v4f32 (scalar_to_vector (loadf32 addr:$src))),
1782 MOVL_shuffle_mask)))]>;
1783 def MOVZSD2PDrm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
1784 "movsd {$src, $dst|$dst, $src}",
1785 [(set VR128:$dst, (v2f64 (vector_shuffle immAllZerosV,
1786 (v2f64 (scalar_to_vector (loadf64 addr:$src))),
1787 MOVL_shuffle_mask)))]>;
1788 // movd / movq to XMM register zero-extends
1789 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, GR32:$src),
1790 "movd {$src, $dst|$dst, $src}",
1791 [(set VR128:$dst, (v4i32 (vector_shuffle immAllZerosV,
1792 (v4i32 (scalar_to_vector GR32:$src)),
1793 MOVL_shuffle_mask)))]>;
1794 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1795 "movd {$src, $dst|$dst, $src}",
1796 [(set VR128:$dst, (v4i32 (vector_shuffle immAllZerosV,
1797 (v4i32 (scalar_to_vector (loadi32 addr:$src))),
1798 MOVL_shuffle_mask)))]>;
1799 // Moving from XMM to XMM but still clear upper 64 bits.
1800 def MOVZQI2PQIrr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR128:$src),
1801 "movq {$src, $dst|$dst, $src}",
1802 [(set VR128:$dst, (int_x86_sse2_movl_dq VR128:$src))]>,
1803 XS, Requires<[HasSSE2]>;
1804 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
1805 "movq {$src, $dst|$dst, $src}",
1806 [(set VR128:$dst, (int_x86_sse2_movl_dq
1807 (bitconvert (loadv2i64 addr:$src))))]>,
1808 XS, Requires<[HasSSE2]>;
1811 //===----------------------------------------------------------------------===//
1812 // Non-Instruction Patterns
1813 //===----------------------------------------------------------------------===//
1815 // 128-bit vector undef's.
1816 def : Pat<(v2f64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1817 def : Pat<(v16i8 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1818 def : Pat<(v8i16 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1819 def : Pat<(v4i32 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1820 def : Pat<(v2i64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1822 // 128-bit vector all zero's.
1823 def : Pat<(v16i8 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1824 def : Pat<(v8i16 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1825 def : Pat<(v4i32 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1826 def : Pat<(v2i64 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1827 def : Pat<(v2f64 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1829 // 128-bit vector all one's.
1830 def : Pat<(v16i8 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1831 def : Pat<(v8i16 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1832 def : Pat<(v4i32 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1833 def : Pat<(v2i64 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1834 def : Pat<(v4f32 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE1]>;
1836 // Store 128-bit integer vector values.
1837 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1838 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1839 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1840 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1841 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1842 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1844 // Scalar to v8i16 / v16i8. The source may be a GR32, but only the lower 8 or
1846 def : Pat<(v8i16 (X86s2vec GR32:$src)), (MOVDI2PDIrr GR32:$src)>,
1847 Requires<[HasSSE2]>;
1848 def : Pat<(v16i8 (X86s2vec GR32:$src)), (MOVDI2PDIrr GR32:$src)>,
1849 Requires<[HasSSE2]>;
1852 let Predicates = [HasSSE2] in {
1853 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
1854 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
1855 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
1856 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
1857 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
1858 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
1859 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
1860 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
1861 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
1862 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
1863 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
1864 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
1865 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
1866 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
1867 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
1868 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
1869 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
1870 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
1871 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
1872 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
1873 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
1874 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
1875 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
1876 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
1877 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
1878 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
1879 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
1880 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
1881 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
1882 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
1885 // Move scalar to XMM zero-extended
1886 // movd to XMM register zero-extends
1887 let AddedComplexity = 20 in {
1888 def : Pat<(v8i16 (vector_shuffle immAllZerosV,
1889 (v8i16 (X86s2vec GR32:$src)), MOVL_shuffle_mask)),
1890 (MOVZDI2PDIrr GR32:$src)>, Requires<[HasSSE2]>;
1891 def : Pat<(v16i8 (vector_shuffle immAllZerosV,
1892 (v16i8 (X86s2vec GR32:$src)), MOVL_shuffle_mask)),
1893 (MOVZDI2PDIrr GR32:$src)>, Requires<[HasSSE2]>;
1894 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
1895 def : Pat<(v2f64 (vector_shuffle immAllZerosV,
1896 (v2f64 (scalar_to_vector FR64:$src)), MOVL_shuffle_mask)),
1897 (MOVLSD2PDrr (V_SET0), FR64:$src)>, Requires<[HasSSE2]>;
1898 def : Pat<(v4f32 (vector_shuffle immAllZerosV,
1899 (v4f32 (scalar_to_vector FR32:$src)), MOVL_shuffle_mask)),
1900 (MOVLSS2PSrr (V_SET0), FR32:$src)>, Requires<[HasSSE2]>;
1903 // Splat v2f64 / v2i64
1904 let AddedComplexity = 10 in {
1905 def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
1906 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1907 def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
1908 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1912 def : Pat<(vector_shuffle (v4f32 VR128:$src), (undef), SSE_splat_mask:$sm),
1913 (SHUFPSrri VR128:$src, VR128:$src, SSE_splat_mask:$sm)>,
1914 Requires<[HasSSE1]>;
1916 // Special unary SHUFPSrri case.
1917 // FIXME: when we want non two-address code, then we should use PSHUFD?
1918 def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef),
1919 SHUFP_unary_shuffle_mask:$sm),
1920 (SHUFPSrri VR128:$src1, VR128:$src1, SHUFP_unary_shuffle_mask:$sm)>,
1921 Requires<[HasSSE1]>;
1922 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
1923 def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef),
1924 SHUFP_unary_shuffle_mask:$sm),
1925 (PSHUFDmi addr:$src1, SHUFP_unary_shuffle_mask:$sm)>,
1926 Requires<[HasSSE2]>;
1927 // Special binary v4i32 shuffle cases with SHUFPS.
1928 def : Pat<(vector_shuffle (v4i32 VR128:$src1), (v4i32 VR128:$src2),
1929 PSHUFD_binary_shuffle_mask:$sm),
1930 (SHUFPSrri VR128:$src1, VR128:$src2, PSHUFD_binary_shuffle_mask:$sm)>,
1931 Requires<[HasSSE2]>;
1932 def : Pat<(vector_shuffle (v4i32 VR128:$src1),
1933 (bc_v4i32 (loadv2i64 addr:$src2)), PSHUFD_binary_shuffle_mask:$sm),
1934 (SHUFPSrmi VR128:$src1, addr:$src2, PSHUFD_binary_shuffle_mask:$sm)>,
1935 Requires<[HasSSE2]>;
1937 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
1938 let AddedComplexity = 10 in {
1939 def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef),
1940 UNPCKL_v_undef_shuffle_mask)),
1941 (UNPCKLPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1942 def : Pat<(v16i8 (vector_shuffle VR128:$src, (undef),
1943 UNPCKL_v_undef_shuffle_mask)),
1944 (PUNPCKLBWrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1945 def : Pat<(v8i16 (vector_shuffle VR128:$src, (undef),
1946 UNPCKL_v_undef_shuffle_mask)),
1947 (PUNPCKLWDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1948 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1949 UNPCKL_v_undef_shuffle_mask)),
1950 (PUNPCKLDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
1953 let AddedComplexity = 20 in {
1954 // vector_shuffle v1, <undef> <1, 1, 3, 3>
1955 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1956 MOVSHDUP_shuffle_mask)),
1957 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
1958 def : Pat<(v4i32 (vector_shuffle (bc_v4i32 (loadv2i64 addr:$src)), (undef),
1959 MOVSHDUP_shuffle_mask)),
1960 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
1962 // vector_shuffle v1, <undef> <0, 0, 2, 2>
1963 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1964 MOVSLDUP_shuffle_mask)),
1965 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
1966 def : Pat<(v4i32 (vector_shuffle (bc_v4i32 (loadv2i64 addr:$src)), (undef),
1967 MOVSLDUP_shuffle_mask)),
1968 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
1971 let AddedComplexity = 20 in {
1972 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1973 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1974 MOVHP_shuffle_mask)),
1975 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1977 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1978 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1979 MOVHLPS_shuffle_mask)),
1980 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1982 // vector_shuffle v1, undef <2, 3, ?, ?> using MOVHLPS
1983 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (undef),
1984 UNPCKH_shuffle_mask)),
1985 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1986 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (undef),
1987 UNPCKH_shuffle_mask)),
1988 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1990 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1991 // vector_shuffle v1, (load v2) <0, 1, 4, 5> using MOVHPS
1992 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (loadv4f32 addr:$src2),
1993 MOVLP_shuffle_mask)),
1994 (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
1995 def : Pat<(v2f64 (vector_shuffle VR128:$src1, (loadv2f64 addr:$src2),
1996 MOVLP_shuffle_mask)),
1997 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
1998 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (loadv4f32 addr:$src2),
1999 MOVHP_shuffle_mask)),
2000 (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
2001 def : Pat<(v2f64 (vector_shuffle VR128:$src1, (loadv2f64 addr:$src2),
2002 MOVHP_shuffle_mask)),
2003 (MOVHPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2005 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)),
2006 MOVLP_shuffle_mask)),
2007 (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2008 def : Pat<(v2i64 (vector_shuffle VR128:$src1, (loadv2i64 addr:$src2),
2009 MOVLP_shuffle_mask)),
2010 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2011 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)),
2012 MOVHP_shuffle_mask)),
2013 (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
2014 def : Pat<(v2i64 (vector_shuffle VR128:$src1, (loadv2i64 addr:$src2),
2015 MOVLP_shuffle_mask)),
2016 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2018 // Setting the lowest element in the vector.
2019 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
2020 MOVL_shuffle_mask)),
2021 (MOVLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2022 def : Pat<(v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
2023 MOVL_shuffle_mask)),
2024 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2026 // vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
2027 def : Pat<(v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
2028 MOVLP_shuffle_mask)),
2029 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2030 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
2031 MOVLP_shuffle_mask)),
2032 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2034 // Set lowest element and zero upper elements.
2035 def : Pat<(bc_v2i64 (vector_shuffle immAllZerosV,
2036 (v2f64 (scalar_to_vector (loadf64 addr:$src))),
2037 MOVL_shuffle_mask)),
2038 (MOVZQI2PQIrm addr:$src)>, Requires<[HasSSE2]>;
2041 // FIXME: Temporary workaround since 2-wide shuffle is broken.
2042 def : Pat<(int_x86_sse2_movs_d VR128:$src1, VR128:$src2),
2043 (v2f64 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2044 def : Pat<(int_x86_sse2_loadh_pd VR128:$src1, addr:$src2),
2045 (v2f64 (MOVHPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2046 def : Pat<(int_x86_sse2_loadl_pd VR128:$src1, addr:$src2),
2047 (v2f64 (MOVLPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2048 def : Pat<(int_x86_sse2_shuf_pd VR128:$src1, VR128:$src2, imm:$src3),
2049 (v2f64 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$src3))>,
2050 Requires<[HasSSE2]>;
2051 def : Pat<(int_x86_sse2_shuf_pd VR128:$src1, (load addr:$src2), imm:$src3),
2052 (v2f64 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$src3))>,
2053 Requires<[HasSSE2]>;
2054 def : Pat<(int_x86_sse2_unpckh_pd VR128:$src1, VR128:$src2),
2055 (v2f64 (UNPCKHPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2056 def : Pat<(int_x86_sse2_unpckh_pd VR128:$src1, (load addr:$src2)),
2057 (v2f64 (UNPCKHPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2058 def : Pat<(int_x86_sse2_unpckl_pd VR128:$src1, VR128:$src2),
2059 (v2f64 (UNPCKLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2060 def : Pat<(int_x86_sse2_unpckl_pd VR128:$src1, (load addr:$src2)),
2061 (v2f64 (UNPCKLPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2062 def : Pat<(int_x86_sse2_punpckh_qdq VR128:$src1, VR128:$src2),
2063 (v2i64 (PUNPCKHQDQrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2064 def : Pat<(int_x86_sse2_punpckh_qdq VR128:$src1, (load addr:$src2)),
2065 (v2i64 (PUNPCKHQDQrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2066 def : Pat<(int_x86_sse2_punpckl_qdq VR128:$src1, VR128:$src2),
2067 (v2i64 (PUNPCKLQDQrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2068 def : Pat<(int_x86_sse2_punpckl_qdq VR128:$src1, (load addr:$src2)),
2069 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2071 // Some special case pandn patterns.
2072 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
2074 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2075 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
2077 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2078 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
2080 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2082 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
2083 (load addr:$src2))),
2084 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2085 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
2086 (load addr:$src2))),
2087 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2088 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
2089 (load addr:$src2))),
2090 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2093 def : Pat<(v4f32 (X86loadu addr:$src)), (MOVUPSrm addr:$src)>,
2094 Requires<[HasSSE1]>;