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 //===----------------------------------------------------------------------===//
16 //===----------------------------------------------------------------------===//
17 // SSE specific DAG Nodes.
18 //===----------------------------------------------------------------------===//
20 def X86loadp : SDNode<"X86ISD::LOAD_PACK", SDTLoad, [SDNPHasChain]>;
21 def X86loadu : SDNode<"X86ISD::LOAD_UA", SDTLoad, [SDNPHasChain]>;
22 def X86fand : SDNode<"X86ISD::FAND", SDTFPBinOp,
23 [SDNPCommutative, SDNPAssociative]>;
24 def X86fxor : SDNode<"X86ISD::FXOR", SDTFPBinOp,
25 [SDNPCommutative, SDNPAssociative]>;
26 def X86comi : SDNode<"X86ISD::COMI", SDTX86CmpTest,
27 [SDNPHasChain, SDNPOutFlag]>;
28 def X86ucomi : SDNode<"X86ISD::UCOMI", SDTX86CmpTest,
29 [SDNPHasChain, SDNPOutFlag]>;
30 def X86s2vec : SDNode<"X86ISD::S2VEC", SDTypeProfile<1, 1, []>, []>;
31 def X86pextrw : SDNode<"X86ISD::PEXTRW", SDTypeProfile<1, 2, []>, []>;
32 def X86pinsrw : SDNode<"X86ISD::PINSRW", SDTypeProfile<1, 3, []>, []>;
34 //===----------------------------------------------------------------------===//
35 // SSE pattern fragments
36 //===----------------------------------------------------------------------===//
38 def X86loadpf32 : PatFrag<(ops node:$ptr), (f32 (X86loadp node:$ptr))>;
39 def X86loadpf64 : PatFrag<(ops node:$ptr), (f64 (X86loadp node:$ptr))>;
41 def loadv4f32 : PatFrag<(ops node:$ptr), (v4f32 (load node:$ptr))>;
42 def loadv2f64 : PatFrag<(ops node:$ptr), (v2f64 (load node:$ptr))>;
43 def loadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (load node:$ptr))>;
45 def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
46 def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
47 def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
48 def bc_v8i16 : PatFrag<(ops node:$in), (v8i16 (bitconvert node:$in))>;
49 def bc_v4i32 : PatFrag<(ops node:$in), (v4i32 (bitconvert node:$in))>;
50 def bc_v2i64 : PatFrag<(ops node:$in), (v2i64 (bitconvert node:$in))>;
52 def fp32imm0 : PatLeaf<(f32 fpimm), [{
53 return N->isExactlyValue(+0.0);
56 def PSxLDQ_imm : SDNodeXForm<imm, [{
57 // Transformation function: imm >> 3
58 return getI32Imm(N->getValue() >> 3);
61 // SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to PSHUF*,
63 def SHUFFLE_get_shuf_imm : SDNodeXForm<build_vector, [{
64 return getI8Imm(X86::getShuffleSHUFImmediate(N));
67 // SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to
69 def SHUFFLE_get_pshufhw_imm : SDNodeXForm<build_vector, [{
70 return getI8Imm(X86::getShufflePSHUFHWImmediate(N));
73 // SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to
75 def SHUFFLE_get_pshuflw_imm : SDNodeXForm<build_vector, [{
76 return getI8Imm(X86::getShufflePSHUFLWImmediate(N));
79 def SSE_splat_mask : PatLeaf<(build_vector), [{
80 return X86::isSplatMask(N);
81 }], SHUFFLE_get_shuf_imm>;
83 def SSE_splat_v2_mask : PatLeaf<(build_vector), [{
84 return X86::isSplatMask(N);
87 def MOVHLPS_shuffle_mask : PatLeaf<(build_vector), [{
88 return X86::isMOVHLPSMask(N);
91 def MOVHP_shuffle_mask : PatLeaf<(build_vector), [{
92 return X86::isMOVHPMask(N);
95 def MOVLP_shuffle_mask : PatLeaf<(build_vector), [{
96 return X86::isMOVLPMask(N);
99 def MOVL_shuffle_mask : PatLeaf<(build_vector), [{
100 return X86::isMOVLMask(N);
103 def MOVSHDUP_shuffle_mask : PatLeaf<(build_vector), [{
104 return X86::isMOVSHDUPMask(N);
107 def MOVSLDUP_shuffle_mask : PatLeaf<(build_vector), [{
108 return X86::isMOVSLDUPMask(N);
111 def UNPCKL_shuffle_mask : PatLeaf<(build_vector), [{
112 return X86::isUNPCKLMask(N);
115 def UNPCKH_shuffle_mask : PatLeaf<(build_vector), [{
116 return X86::isUNPCKHMask(N);
119 def UNPCKL_v_undef_shuffle_mask : PatLeaf<(build_vector), [{
120 return X86::isUNPCKL_v_undef_Mask(N);
123 def PSHUFD_shuffle_mask : PatLeaf<(build_vector), [{
124 return X86::isPSHUFDMask(N);
125 }], SHUFFLE_get_shuf_imm>;
127 def PSHUFHW_shuffle_mask : PatLeaf<(build_vector), [{
128 return X86::isPSHUFHWMask(N);
129 }], SHUFFLE_get_pshufhw_imm>;
131 def PSHUFLW_shuffle_mask : PatLeaf<(build_vector), [{
132 return X86::isPSHUFLWMask(N);
133 }], SHUFFLE_get_pshuflw_imm>;
135 def SHUFP_unary_shuffle_mask : PatLeaf<(build_vector), [{
136 return X86::isPSHUFDMask(N);
137 }], SHUFFLE_get_shuf_imm>;
139 def SHUFP_shuffle_mask : PatLeaf<(build_vector), [{
140 return X86::isSHUFPMask(N);
141 }], SHUFFLE_get_shuf_imm>;
143 def PSHUFD_binary_shuffle_mask : PatLeaf<(build_vector), [{
144 return X86::isSHUFPMask(N);
145 }], SHUFFLE_get_shuf_imm>;
147 //===----------------------------------------------------------------------===//
148 // SSE scalar FP Instructions
149 //===----------------------------------------------------------------------===//
151 // Instruction templates
152 // SSI - SSE1 instructions with XS prefix.
153 // SDI - SSE2 instructions with XD prefix.
154 // PSI - SSE1 instructions with TB prefix.
155 // PDI - SSE2 instructions with TB and OpSize prefixes.
156 // PSIi8 - SSE1 instructions with ImmT == Imm8 and TB prefix.
157 // PDIi8 - SSE2 instructions with ImmT == Imm8 and TB and OpSize prefixes.
158 // S3I - SSE3 instructions with TB and OpSize prefixes.
159 // S3SI - SSE3 instructions with XS prefix.
160 // S3DI - SSE3 instructions with XD prefix.
161 class SSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
162 : I<o, F, ops, asm, pattern>, XS, Requires<[HasSSE1]>;
163 class SDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
164 : I<o, F, ops, asm, pattern>, XD, Requires<[HasSSE2]>;
165 class PSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
166 : I<o, F, ops, asm, pattern>, TB, Requires<[HasSSE1]>;
167 class PDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
168 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
169 class PSIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
170 : Ii8<o, F, ops, asm, pattern>, TB, Requires<[HasSSE1]>;
171 class PDIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
172 : Ii8<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
174 class S3SI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
175 : I<o, F, ops, asm, pattern>, XS, Requires<[HasSSE3]>;
176 class S3DI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
177 : I<o, F, ops, asm, pattern>, XD, Requires<[HasSSE3]>;
178 class S3I<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
179 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE3]>;
181 //===----------------------------------------------------------------------===//
182 // Helpers for defining instructions that directly correspond to intrinsics.
184 multiclass SS_IntUnary<bits<8> o, string OpcodeStr, Intrinsic IntId> {
185 def r : SSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
186 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
187 [(set VR128:$dst, (v4f32 (IntId VR128:$src)))]>;
188 def m : SSI<o, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
189 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
190 [(set VR128:$dst, (v4f32 (IntId (load addr:$src))))]>;
193 multiclass SD_IntUnary<bits<8> o, string OpcodeStr, Intrinsic IntId> {
194 def r : SDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
195 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
196 [(set VR128:$dst, (v2f64 (IntId VR128:$src)))]>;
197 def m : SDI<o, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
198 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src"),
199 [(set VR128:$dst, (v2f64 (IntId (load addr:$src))))]>;
202 class PS_Intr<bits<8> o, string OpcodeStr, Intrinsic IntId>
203 : PSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
204 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
205 [(set VR128:$dst, (IntId VR128:$src))]>;
206 class PS_Intm<bits<8> o, string OpcodeStr, Intrinsic IntId>
207 : PSI<o, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
208 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
209 [(set VR128:$dst, (IntId (load addr:$src)))]>;
210 class PD_Intr<bits<8> o, string OpcodeStr, Intrinsic IntId>
211 : PDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src),
212 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
213 [(set VR128:$dst, (IntId VR128:$src))]>;
214 class PD_Intm<bits<8> o, string OpcodeStr, Intrinsic IntId>
215 : PDI<o, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
216 !strconcat(OpcodeStr, " {$src, $dst|$dst, $src}"),
217 [(set VR128:$dst, (IntId (load addr:$src)))]>;
219 class PS_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
220 : PSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
221 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
222 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
223 class PS_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
224 : PSI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
225 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
226 [(set VR128:$dst, (IntId VR128:$src1, (load addr:$src2)))]>;
227 class PD_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
228 : PDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
229 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
230 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
231 class PD_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
232 : PDI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
233 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
234 [(set VR128:$dst, (IntId VR128:$src1, (load addr:$src2)))]>;
236 // Some 'special' instructions
237 def IMPLICIT_DEF_FR32 : I<0, Pseudo, (ops FR32:$dst),
238 "#IMPLICIT_DEF $dst",
239 [(set FR32:$dst, (undef))]>, Requires<[HasSSE2]>;
240 def IMPLICIT_DEF_FR64 : I<0, Pseudo, (ops FR64:$dst),
241 "#IMPLICIT_DEF $dst",
242 [(set FR64:$dst, (undef))]>, Requires<[HasSSE2]>;
244 // CMOV* - Used to implement the SSE SELECT DAG operation. Expanded by the
245 // scheduler into a branch sequence.
246 let usesCustomDAGSchedInserter = 1 in { // Expanded by the scheduler.
247 def CMOV_FR32 : I<0, Pseudo,
248 (ops FR32:$dst, FR32:$t, FR32:$f, i8imm:$cond),
249 "#CMOV_FR32 PSEUDO!",
250 [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond))]>;
251 def CMOV_FR64 : I<0, Pseudo,
252 (ops FR64:$dst, FR64:$t, FR64:$f, i8imm:$cond),
253 "#CMOV_FR64 PSEUDO!",
254 [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond))]>;
255 def CMOV_V4F32 : I<0, Pseudo,
256 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
257 "#CMOV_V4F32 PSEUDO!",
259 (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
260 def CMOV_V2F64 : I<0, Pseudo,
261 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
262 "#CMOV_V2F64 PSEUDO!",
264 (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
265 def CMOV_V2I64 : I<0, Pseudo,
266 (ops VR128:$dst, VR128:$t, VR128:$f, i8imm:$cond),
267 "#CMOV_V2I64 PSEUDO!",
269 (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond)))]>;
273 def MOVSSrr : SSI<0x10, MRMSrcReg, (ops FR32:$dst, FR32:$src),
274 "movss {$src, $dst|$dst, $src}", []>;
275 def MOVSSrm : SSI<0x10, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
276 "movss {$src, $dst|$dst, $src}",
277 [(set FR32:$dst, (loadf32 addr:$src))]>;
278 def MOVSDrr : SDI<0x10, MRMSrcReg, (ops FR64:$dst, FR64:$src),
279 "movsd {$src, $dst|$dst, $src}", []>;
280 def MOVSDrm : SDI<0x10, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
281 "movsd {$src, $dst|$dst, $src}",
282 [(set FR64:$dst, (loadf64 addr:$src))]>;
284 def MOVSSmr : SSI<0x11, MRMDestMem, (ops f32mem:$dst, FR32:$src),
285 "movss {$src, $dst|$dst, $src}",
286 [(store FR32:$src, addr:$dst)]>;
287 def MOVSDmr : SDI<0x11, MRMDestMem, (ops f64mem:$dst, FR64:$src),
288 "movsd {$src, $dst|$dst, $src}",
289 [(store FR64:$src, addr:$dst)]>;
291 let isTwoAddress = 1 in {
293 /// scalar_sse12_fp_binop_rm - Scalar SSE binops come in four basic forms:
294 /// 1. f32 vs f64 - These come in SSE1/SSE2 forms for float/doubles.
295 /// 2. rr vs rm - They include a reg+reg form and a ref+mem form.
297 /// In addition, scalar SSE ops have an intrinsic form. This form is unlike the
298 /// normal form, in that they take an entire vector (instead of a scalar) and
299 /// leave the top elements undefined. This adds another two variants of the
300 /// above permutations, giving us 8 forms for 'instruction'.
302 multiclass scalar_sse12_fp_binop_rm<bits<8> opc, string OpcodeStr,
303 SDNode OpNode, Intrinsic F32Int,
304 Intrinsic F64Int, bit Commutable = 0> {
305 // Scalar operation, reg+reg.
306 def SSrr : SSI<opc, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
307 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
308 [(set FR32:$dst, (OpNode FR32:$src1, FR32:$src2))]> {
309 let isCommutable = Commutable;
311 def SDrr : SDI<opc, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
312 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
313 [(set FR64:$dst, (OpNode FR64:$src1, FR64:$src2))]> {
314 let isCommutable = Commutable;
316 // Scalar operation, reg+mem.
317 def SSrm : SSI<opc, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
318 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
319 [(set FR32:$dst, (OpNode FR32:$src1, (loadf32 addr:$src2)))]>;
320 def SDrm : SDI<opc, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
321 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
322 [(set FR64:$dst, (OpNode FR64:$src1, (loadf64 addr:$src2)))]>;
324 // Vector intrinsic operation, reg+reg.
325 def SSrr_Int : SSI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
326 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
327 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2))]> {
328 let isCommutable = Commutable;
330 def SDrr_Int : SDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
331 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
332 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2))]> {
333 let isCommutable = Commutable;
335 // Vector intrinsic operation, reg+mem.
336 def SSrm_Int : SSI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
337 !strconcat(OpcodeStr, "ss {$src2, $dst|$dst, $src2"),
338 [(set VR128:$dst, (F32Int VR128:$src1,
339 (load addr:$src2)))]>;
340 def SDrm_Int : SDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
341 !strconcat(OpcodeStr, "sd {$src2, $dst|$dst, $src2"),
342 [(set VR128:$dst, (F64Int VR128:$src1,
343 (load addr:$src2)))]>;
347 // Arithmetic instructions
349 defm ADD : scalar_sse12_fp_binop_rm<0x58, "add", fadd,
350 int_x86_sse_add_ss, int_x86_sse2_add_sd, 1>;
351 defm MUL : scalar_sse12_fp_binop_rm<0x59, "mul", fmul,
352 int_x86_sse_mul_ss, int_x86_sse2_mul_sd, 1>;
353 defm SUB : scalar_sse12_fp_binop_rm<0x5C, "sub", fsub,
354 int_x86_sse_sub_ss, int_x86_sse2_sub_sd>;
355 defm DIV : scalar_sse12_fp_binop_rm<0x5E, "div", fdiv,
356 int_x86_sse_div_ss, int_x86_sse2_div_sd>;
359 def SQRTSSr : SSI<0x51, MRMSrcReg, (ops FR32:$dst, FR32:$src),
360 "sqrtss {$src, $dst|$dst, $src}",
361 [(set FR32:$dst, (fsqrt FR32:$src))]>;
362 def SQRTSSm : SSI<0x51, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
363 "sqrtss {$src, $dst|$dst, $src}",
364 [(set FR32:$dst, (fsqrt (loadf32 addr:$src)))]>;
365 def SQRTSDr : SDI<0x51, MRMSrcReg, (ops FR64:$dst, FR64:$src),
366 "sqrtsd {$src, $dst|$dst, $src}",
367 [(set FR64:$dst, (fsqrt FR64:$src))]>;
368 def SQRTSDm : SDI<0x51, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
369 "sqrtsd {$src, $dst|$dst, $src}",
370 [(set FR64:$dst, (fsqrt (loadf64 addr:$src)))]>;
372 class SS_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
373 : SSI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
374 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
375 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, VR128:$src2)))]>;
376 class SS_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
377 : SSI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
378 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
379 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, (load addr:$src2))))]>;
380 class SD_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
381 : SDI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
382 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
383 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, VR128:$src2)))]>;
384 class SD_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
385 : SDI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
386 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
387 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, (load addr:$src2))))]>;
390 // Aliases to match intrinsics which expect XMM operand(s).
392 defm SQRTSS_Int : SS_IntUnary<0x51, "sqrtss" , int_x86_sse_sqrt_ss>;
393 defm SQRTSD_Int : SD_IntUnary<0x51, "sqrtsd" , int_x86_sse2_sqrt_sd>;
394 defm RSQRTSS_Int : SS_IntUnary<0x52, "rsqrtss", int_x86_sse_rsqrt_ss>;
395 defm RCPSS_Int : SS_IntUnary<0x53, "rcpss" , int_x86_sse_rcp_ss>;
397 let isTwoAddress = 1 in {
398 let isCommutable = 1 in {
399 def Int_MAXSSrr : SS_Intrr<0x5F, "maxss", int_x86_sse_max_ss>;
400 def Int_MAXSDrr : SD_Intrr<0x5F, "maxsd", int_x86_sse2_max_sd>;
401 def Int_MINSSrr : SS_Intrr<0x5D, "minss", int_x86_sse_min_ss>;
402 def Int_MINSDrr : SD_Intrr<0x5D, "minsd", int_x86_sse2_min_sd>;
404 def Int_MAXSSrm : SS_Intrm<0x5F, "maxss", int_x86_sse_max_ss>;
405 def Int_MAXSDrm : SD_Intrm<0x5F, "maxsd", int_x86_sse2_max_sd>;
406 def Int_MINSSrm : SS_Intrm<0x5D, "minss", int_x86_sse_min_ss>;
407 def Int_MINSDrm : SD_Intrm<0x5D, "minsd", int_x86_sse2_min_sd>;
410 // Conversion instructions
411 def CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops GR32:$dst, FR32:$src),
412 "cvttss2si {$src, $dst|$dst, $src}",
413 [(set GR32:$dst, (fp_to_sint FR32:$src))]>;
414 def CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
415 "cvttss2si {$src, $dst|$dst, $src}",
416 [(set GR32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
417 def CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops GR32:$dst, FR64:$src),
418 "cvttsd2si {$src, $dst|$dst, $src}",
419 [(set GR32:$dst, (fp_to_sint FR64:$src))]>;
420 def CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops GR32:$dst, f64mem:$src),
421 "cvttsd2si {$src, $dst|$dst, $src}",
422 [(set GR32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
423 def CVTSD2SSrr: SDI<0x5A, MRMSrcReg, (ops FR32:$dst, FR64:$src),
424 "cvtsd2ss {$src, $dst|$dst, $src}",
425 [(set FR32:$dst, (fround FR64:$src))]>;
426 def CVTSD2SSrm: SDI<0x5A, MRMSrcMem, (ops FR32:$dst, f64mem:$src),
427 "cvtsd2ss {$src, $dst|$dst, $src}",
428 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
429 def CVTSI2SSrr: SSI<0x2A, MRMSrcReg, (ops FR32:$dst, GR32:$src),
430 "cvtsi2ss {$src, $dst|$dst, $src}",
431 [(set FR32:$dst, (sint_to_fp GR32:$src))]>;
432 def CVTSI2SSrm: SSI<0x2A, MRMSrcMem, (ops FR32:$dst, i32mem:$src),
433 "cvtsi2ss {$src, $dst|$dst, $src}",
434 [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
435 def CVTSI2SDrr: SDI<0x2A, MRMSrcReg, (ops FR64:$dst, GR32:$src),
436 "cvtsi2sd {$src, $dst|$dst, $src}",
437 [(set FR64:$dst, (sint_to_fp GR32:$src))]>;
438 def CVTSI2SDrm: SDI<0x2A, MRMSrcMem, (ops FR64:$dst, i32mem:$src),
439 "cvtsi2sd {$src, $dst|$dst, $src}",
440 [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
442 // SSE2 instructions with XS prefix
443 def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
444 "cvtss2sd {$src, $dst|$dst, $src}",
445 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
447 def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
448 "cvtss2sd {$src, $dst|$dst, $src}",
449 [(set FR64:$dst, (extload addr:$src, f32))]>, XS,
452 // Match intrinsics which expect XMM operand(s).
453 def Int_CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (ops GR32:$dst, VR128:$src),
454 "cvtss2si {$src, $dst|$dst, $src}",
455 [(set GR32:$dst, (int_x86_sse_cvtss2si VR128:$src))]>;
456 def Int_CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
457 "cvtss2si {$src, $dst|$dst, $src}",
458 [(set GR32:$dst, (int_x86_sse_cvtss2si
459 (load addr:$src)))]>;
460 def Int_CVTSD2SIrr: SDI<0x2D, MRMSrcReg, (ops GR32:$dst, VR128:$src),
461 "cvtsd2si {$src, $dst|$dst, $src}",
462 [(set GR32:$dst, (int_x86_sse2_cvtsd2si VR128:$src))]>;
463 def Int_CVTSD2SIrm: SDI<0x2D, MRMSrcMem, (ops GR32:$dst, f128mem:$src),
464 "cvtsd2si {$src, $dst|$dst, $src}",
465 [(set GR32:$dst, (int_x86_sse2_cvtsd2si
466 (load addr:$src)))]>;
468 // Aliases for intrinsics
469 def Int_CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops GR32:$dst, VR128:$src),
470 "cvttss2si {$src, $dst|$dst, $src}",
471 [(set GR32:$dst, (int_x86_sse_cvttss2si VR128:$src))]>;
472 def Int_CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops GR32:$dst, f32mem:$src),
473 "cvttss2si {$src, $dst|$dst, $src}",
474 [(set GR32:$dst, (int_x86_sse_cvttss2si(load addr:$src)))]>;
475 def Int_CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops GR32:$dst, VR128:$src),
476 "cvttsd2si {$src, $dst|$dst, $src}",
477 [(set GR32:$dst, (int_x86_sse2_cvttsd2si VR128:$src))]>;
478 def Int_CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops GR32:$dst, f128mem:$src),
479 "cvttsd2si {$src, $dst|$dst, $src}",
480 [(set GR32:$dst, (int_x86_sse2_cvttsd2si
481 (load addr:$src)))]>;
483 let isTwoAddress = 1 in {
484 def Int_CVTSI2SSrr: SSI<0x2A, MRMSrcReg,
485 (ops VR128:$dst, VR128:$src1, GR32:$src2),
486 "cvtsi2ss {$src2, $dst|$dst, $src2}",
487 [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
489 def Int_CVTSI2SSrm: SSI<0x2A, MRMSrcMem,
490 (ops VR128:$dst, VR128:$src1, i32mem:$src2),
491 "cvtsi2ss {$src2, $dst|$dst, $src2}",
492 [(set VR128:$dst, (int_x86_sse_cvtsi2ss VR128:$src1,
493 (loadi32 addr:$src2)))]>;
496 // Comparison instructions
497 let isTwoAddress = 1 in {
498 def CMPSSrr : SSI<0xC2, MRMSrcReg,
499 (ops FR32:$dst, FR32:$src1, FR32:$src, SSECC:$cc),
500 "cmp${cc}ss {$src, $dst|$dst, $src}",
502 def CMPSSrm : SSI<0xC2, MRMSrcMem,
503 (ops FR32:$dst, FR32:$src1, f32mem:$src, SSECC:$cc),
504 "cmp${cc}ss {$src, $dst|$dst, $src}", []>;
505 def CMPSDrr : SDI<0xC2, MRMSrcReg,
506 (ops FR64:$dst, FR64:$src1, FR64:$src, SSECC:$cc),
507 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
508 def CMPSDrm : SDI<0xC2, MRMSrcMem,
509 (ops FR64:$dst, FR64:$src1, f64mem:$src, SSECC:$cc),
510 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
513 def UCOMISSrr: PSI<0x2E, MRMSrcReg, (ops FR32:$src1, FR32:$src2),
514 "ucomiss {$src2, $src1|$src1, $src2}",
515 [(X86cmp FR32:$src1, FR32:$src2)]>;
516 def UCOMISSrm: PSI<0x2E, MRMSrcMem, (ops FR32:$src1, f32mem:$src2),
517 "ucomiss {$src2, $src1|$src1, $src2}",
518 [(X86cmp FR32:$src1, (loadf32 addr:$src2))]>;
519 def UCOMISDrr: PDI<0x2E, MRMSrcReg, (ops FR64:$src1, FR64:$src2),
520 "ucomisd {$src2, $src1|$src1, $src2}",
521 [(X86cmp FR64:$src1, FR64:$src2)]>;
522 def UCOMISDrm: PDI<0x2E, MRMSrcMem, (ops FR64:$src1, f64mem:$src2),
523 "ucomisd {$src2, $src1|$src1, $src2}",
524 [(X86cmp FR64:$src1, (loadf64 addr:$src2))]>;
526 // Aliases to match intrinsics which expect XMM operand(s).
527 let isTwoAddress = 1 in {
528 def Int_CMPSSrr : SSI<0xC2, MRMSrcReg,
529 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
530 "cmp${cc}ss {$src, $dst|$dst, $src}",
531 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
532 VR128:$src, imm:$cc))]>;
533 def Int_CMPSSrm : SSI<0xC2, MRMSrcMem,
534 (ops VR128:$dst, VR128:$src1, f32mem:$src, SSECC:$cc),
535 "cmp${cc}ss {$src, $dst|$dst, $src}",
536 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
537 (load addr:$src), imm:$cc))]>;
538 def Int_CMPSDrr : SDI<0xC2, MRMSrcReg,
539 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
540 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
541 def Int_CMPSDrm : SDI<0xC2, MRMSrcMem,
542 (ops VR128:$dst, VR128:$src1, f64mem:$src, SSECC:$cc),
543 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
546 def Int_UCOMISSrr: PSI<0x2E, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
547 "ucomiss {$src2, $src1|$src1, $src2}",
548 [(X86ucomi (v4f32 VR128:$src1), VR128:$src2)]>;
549 def Int_UCOMISSrm: PSI<0x2E, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
550 "ucomiss {$src2, $src1|$src1, $src2}",
551 [(X86ucomi (v4f32 VR128:$src1), (load addr:$src2))]>;
552 def Int_UCOMISDrr: PDI<0x2E, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
553 "ucomisd {$src2, $src1|$src1, $src2}",
554 [(X86ucomi (v2f64 VR128:$src1), (v2f64 VR128:$src2))]>;
555 def Int_UCOMISDrm: PDI<0x2E, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
556 "ucomisd {$src2, $src1|$src1, $src2}",
557 [(X86ucomi (v2f64 VR128:$src1), (load addr:$src2))]>;
559 def Int_COMISSrr: PSI<0x2F, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
560 "comiss {$src2, $src1|$src1, $src2}",
561 [(X86comi (v4f32 VR128:$src1), VR128:$src2)]>;
562 def Int_COMISSrm: PSI<0x2F, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
563 "comiss {$src2, $src1|$src1, $src2}",
564 [(X86comi (v4f32 VR128:$src1), (load addr:$src2))]>;
565 def Int_COMISDrr: PDI<0x2F, MRMSrcReg, (ops VR128:$src1, VR128:$src2),
566 "comisd {$src2, $src1|$src1, $src2}",
567 [(X86comi (v2f64 VR128:$src1), (v2f64 VR128:$src2))]>;
568 def Int_COMISDrm: PDI<0x2F, MRMSrcMem, (ops VR128:$src1, f128mem:$src2),
569 "comisd {$src2, $src1|$src1, $src2}",
570 [(X86comi (v2f64 VR128:$src1), (load addr:$src2))]>;
572 // Aliases of packed instructions for scalar use. These all have names that
575 // Alias instructions that map fld0 to pxor for sse.
576 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
577 def FsFLD0SS : I<0xEF, MRMInitReg, (ops FR32:$dst),
578 "pxor $dst, $dst", [(set FR32:$dst, fp32imm0)]>,
579 Requires<[HasSSE1]>, TB, OpSize;
580 def FsFLD0SD : I<0xEF, MRMInitReg, (ops FR64:$dst),
581 "pxor $dst, $dst", [(set FR64:$dst, fp64imm0)]>,
582 Requires<[HasSSE2]>, TB, OpSize;
584 // Alias instructions to do FR32 / FR64 reg-to-reg copy using movaps / movapd.
585 // Upper bits are disregarded.
586 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (ops FR32:$dst, FR32:$src),
587 "movaps {$src, $dst|$dst, $src}", []>;
588 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (ops FR64:$dst, FR64:$src),
589 "movapd {$src, $dst|$dst, $src}", []>;
591 // Alias instructions to load FR32 / FR64 from f128mem using movaps / movapd.
592 // Upper bits are disregarded.
593 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (ops FR32:$dst, f128mem:$src),
594 "movaps {$src, $dst|$dst, $src}",
595 [(set FR32:$dst, (X86loadpf32 addr:$src))]>;
596 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (ops FR64:$dst, f128mem:$src),
597 "movapd {$src, $dst|$dst, $src}",
598 [(set FR64:$dst, (X86loadpf64 addr:$src))]>;
600 // Alias bitwise logical operations using SSE logical ops on packed FP values.
601 let isTwoAddress = 1 in {
602 let isCommutable = 1 in {
603 def FsANDPSrr : PSI<0x54, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
604 "andps {$src2, $dst|$dst, $src2}",
605 [(set FR32:$dst, (X86fand FR32:$src1, FR32:$src2))]>;
606 def FsANDPDrr : PDI<0x54, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
607 "andpd {$src2, $dst|$dst, $src2}",
608 [(set FR64:$dst, (X86fand FR64:$src1, FR64:$src2))]>;
609 def FsORPSrr : PSI<0x56, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
610 "orps {$src2, $dst|$dst, $src2}", []>;
611 def FsORPDrr : PDI<0x56, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
612 "orpd {$src2, $dst|$dst, $src2}", []>;
613 def FsXORPSrr : PSI<0x57, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
614 "xorps {$src2, $dst|$dst, $src2}",
615 [(set FR32:$dst, (X86fxor FR32:$src1, FR32:$src2))]>;
616 def FsXORPDrr : PDI<0x57, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
617 "xorpd {$src2, $dst|$dst, $src2}",
618 [(set FR64:$dst, (X86fxor FR64:$src1, FR64:$src2))]>;
620 def FsANDPSrm : PSI<0x54, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
621 "andps {$src2, $dst|$dst, $src2}",
622 [(set FR32:$dst, (X86fand FR32:$src1,
623 (X86loadpf32 addr:$src2)))]>;
624 def FsANDPDrm : PDI<0x54, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
625 "andpd {$src2, $dst|$dst, $src2}",
626 [(set FR64:$dst, (X86fand FR64:$src1,
627 (X86loadpf64 addr:$src2)))]>;
628 def FsORPSrm : PSI<0x56, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
629 "orps {$src2, $dst|$dst, $src2}", []>;
630 def FsORPDrm : PDI<0x56, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
631 "orpd {$src2, $dst|$dst, $src2}", []>;
632 def FsXORPSrm : PSI<0x57, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
633 "xorps {$src2, $dst|$dst, $src2}",
634 [(set FR32:$dst, (X86fxor FR32:$src1,
635 (X86loadpf32 addr:$src2)))]>;
636 def FsXORPDrm : PDI<0x57, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
637 "xorpd {$src2, $dst|$dst, $src2}",
638 [(set FR64:$dst, (X86fxor FR64:$src1,
639 (X86loadpf64 addr:$src2)))]>;
641 def FsANDNPSrr : PSI<0x55, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
642 "andnps {$src2, $dst|$dst, $src2}", []>;
643 def FsANDNPSrm : PSI<0x55, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
644 "andnps {$src2, $dst|$dst, $src2}", []>;
645 def FsANDNPDrr : PDI<0x55, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
646 "andnpd {$src2, $dst|$dst, $src2}", []>;
647 def FsANDNPDrm : PDI<0x55, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
648 "andnpd {$src2, $dst|$dst, $src2}", []>;
651 //===----------------------------------------------------------------------===//
652 // SSE packed FP Instructions
653 //===----------------------------------------------------------------------===//
655 // Some 'special' instructions
656 def IMPLICIT_DEF_VR128 : I<0, Pseudo, (ops VR128:$dst),
657 "#IMPLICIT_DEF $dst",
658 [(set VR128:$dst, (v4f32 (undef)))]>,
662 def MOVAPSrr : PSI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
663 "movaps {$src, $dst|$dst, $src}", []>;
664 def MOVAPSrm : PSI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
665 "movaps {$src, $dst|$dst, $src}",
666 [(set VR128:$dst, (loadv4f32 addr:$src))]>;
667 def MOVAPDrr : PDI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
668 "movapd {$src, $dst|$dst, $src}", []>;
669 def MOVAPDrm : PDI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
670 "movapd {$src, $dst|$dst, $src}",
671 [(set VR128:$dst, (loadv2f64 addr:$src))]>;
673 def MOVAPSmr : PSI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
674 "movaps {$src, $dst|$dst, $src}",
675 [(store (v4f32 VR128:$src), addr:$dst)]>;
676 def MOVAPDmr : PDI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
677 "movapd {$src, $dst|$dst, $src}",
678 [(store (v2f64 VR128:$src), addr:$dst)]>;
680 def MOVUPSrr : PSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
681 "movups {$src, $dst|$dst, $src}", []>;
682 def MOVUPSrm : PSI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
683 "movups {$src, $dst|$dst, $src}",
684 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
685 def MOVUPSmr : PSI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
686 "movups {$src, $dst|$dst, $src}",
687 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
688 def MOVUPDrr : PDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
689 "movupd {$src, $dst|$dst, $src}", []>;
690 def MOVUPDrm : PDI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
691 "movupd {$src, $dst|$dst, $src}",
692 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
693 def MOVUPDmr : PDI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
694 "movupd {$src, $dst|$dst, $src}",
695 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
697 let isTwoAddress = 1 in {
698 let AddedComplexity = 20 in {
699 def MOVLPSrm : PSI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
700 "movlps {$src2, $dst|$dst, $src2}",
702 (v4f32 (vector_shuffle VR128:$src1,
703 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
704 MOVLP_shuffle_mask)))]>;
705 def MOVLPDrm : PDI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
706 "movlpd {$src2, $dst|$dst, $src2}",
708 (v2f64 (vector_shuffle VR128:$src1,
709 (scalar_to_vector (loadf64 addr:$src2)),
710 MOVLP_shuffle_mask)))]>;
711 def MOVHPSrm : PSI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
712 "movhps {$src2, $dst|$dst, $src2}",
714 (v4f32 (vector_shuffle VR128:$src1,
715 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2)))),
716 MOVHP_shuffle_mask)))]>;
717 def MOVHPDrm : PDI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
718 "movhpd {$src2, $dst|$dst, $src2}",
720 (v2f64 (vector_shuffle VR128:$src1,
721 (scalar_to_vector (loadf64 addr:$src2)),
722 MOVHP_shuffle_mask)))]>;
726 def MOVLPSmr : PSI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
727 "movlps {$src, $dst|$dst, $src}",
728 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
729 (iPTR 0))), addr:$dst)]>;
730 def MOVLPDmr : PDI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
731 "movlpd {$src, $dst|$dst, $src}",
732 [(store (f64 (vector_extract (v2f64 VR128:$src),
733 (iPTR 0))), addr:$dst)]>;
735 // v2f64 extract element 1 is always custom lowered to unpack high to low
736 // and extract element 0 so the non-store version isn't too horrible.
737 def MOVHPSmr : PSI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
738 "movhps {$src, $dst|$dst, $src}",
739 [(store (f64 (vector_extract
740 (v2f64 (vector_shuffle
741 (bc_v2f64 (v4f32 VR128:$src)), (undef),
742 UNPCKH_shuffle_mask)), (iPTR 0))),
744 def MOVHPDmr : PDI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
745 "movhpd {$src, $dst|$dst, $src}",
746 [(store (f64 (vector_extract
747 (v2f64 (vector_shuffle VR128:$src, (undef),
748 UNPCKH_shuffle_mask)), (iPTR 0))),
751 let isTwoAddress = 1 in {
752 let AddedComplexity = 20 in {
753 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
754 "movlhps {$src2, $dst|$dst, $src2}",
756 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
757 MOVHP_shuffle_mask)))]>;
759 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
760 "movhlps {$src2, $dst|$dst, $src2}",
762 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
763 MOVHLPS_shuffle_mask)))]>;
767 def MOVSHDUPrr : S3SI<0x16, MRMSrcReg, (ops VR128:$dst, VR128:$src),
768 "movshdup {$src, $dst|$dst, $src}",
769 [(set VR128:$dst, (v4f32 (vector_shuffle
771 MOVSHDUP_shuffle_mask)))]>;
772 def MOVSHDUPrm : S3SI<0x16, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
773 "movshdup {$src, $dst|$dst, $src}",
774 [(set VR128:$dst, (v4f32 (vector_shuffle
775 (loadv4f32 addr:$src), (undef),
776 MOVSHDUP_shuffle_mask)))]>;
778 def MOVSLDUPrr : S3SI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src),
779 "movsldup {$src, $dst|$dst, $src}",
780 [(set VR128:$dst, (v4f32 (vector_shuffle
782 MOVSLDUP_shuffle_mask)))]>;
783 def MOVSLDUPrm : S3SI<0x12, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
784 "movsldup {$src, $dst|$dst, $src}",
785 [(set VR128:$dst, (v4f32 (vector_shuffle
786 (loadv4f32 addr:$src), (undef),
787 MOVSLDUP_shuffle_mask)))]>;
789 def MOVDDUPrr : S3DI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src),
790 "movddup {$src, $dst|$dst, $src}",
791 [(set VR128:$dst, (v2f64 (vector_shuffle
793 SSE_splat_v2_mask)))]>;
794 def MOVDDUPrm : S3DI<0x12, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
795 "movddup {$src, $dst|$dst, $src}",
796 [(set VR128:$dst, (v2f64 (vector_shuffle
797 (scalar_to_vector (loadf64 addr:$src)),
799 SSE_splat_v2_mask)))]>;
801 // SSE2 instructions without OpSize prefix
802 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
803 "cvtdq2ps {$src, $dst|$dst, $src}",
804 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
805 TB, Requires<[HasSSE2]>;
806 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
807 "cvtdq2ps {$src, $dst|$dst, $src}",
808 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
809 (bitconvert (loadv2i64 addr:$src))))]>,
810 TB, Requires<[HasSSE2]>;
812 // SSE2 instructions with XS prefix
813 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
814 "cvtdq2pd {$src, $dst|$dst, $src}",
815 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
816 XS, Requires<[HasSSE2]>;
817 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
818 "cvtdq2pd {$src, $dst|$dst, $src}",
819 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
820 (bitconvert (loadv2i64 addr:$src))))]>,
821 XS, Requires<[HasSSE2]>;
823 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
824 "cvtps2dq {$src, $dst|$dst, $src}",
825 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
826 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
827 "cvtps2dq {$src, $dst|$dst, $src}",
828 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
829 (load addr:$src)))]>;
830 // SSE2 packed instructions with XS prefix
831 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
832 "cvttps2dq {$src, $dst|$dst, $src}",
833 [(set VR128:$dst, (int_x86_sse2_cvttps2dq VR128:$src))]>,
834 XS, Requires<[HasSSE2]>;
835 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
836 "cvttps2dq {$src, $dst|$dst, $src}",
837 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
838 (load addr:$src)))]>,
839 XS, Requires<[HasSSE2]>;
841 // SSE2 packed instructions with XD prefix
842 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
843 "cvtpd2dq {$src, $dst|$dst, $src}",
844 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
845 XD, Requires<[HasSSE2]>;
846 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
847 "cvtpd2dq {$src, $dst|$dst, $src}",
848 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
849 (load addr:$src)))]>,
850 XD, Requires<[HasSSE2]>;
851 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
852 "cvttpd2dq {$src, $dst|$dst, $src}",
853 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
854 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
855 "cvttpd2dq {$src, $dst|$dst, $src}",
856 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
857 (load addr:$src)))]>;
859 // SSE2 instructions without OpSize prefix
860 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
861 "cvtps2pd {$src, $dst|$dst, $src}",
862 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
863 TB, Requires<[HasSSE2]>;
864 def Int_CVTPS2PDrm : I<0x5A, MRMSrcReg, (ops VR128:$dst, f64mem:$src),
865 "cvtps2pd {$src, $dst|$dst, $src}",
866 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
867 (load addr:$src)))]>,
868 TB, Requires<[HasSSE2]>;
870 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
871 "cvtpd2ps {$src, $dst|$dst, $src}",
872 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
873 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, f128mem:$src),
874 "cvtpd2ps {$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
876 (load addr:$src)))]>;
878 // Match intrinsics which expect XMM operand(s).
879 // Aliases for intrinsics
880 let isTwoAddress = 1 in {
881 def Int_CVTSI2SDrr: SDI<0x2A, MRMSrcReg,
882 (ops VR128:$dst, VR128:$src1, GR32:$src2),
883 "cvtsi2sd {$src2, $dst|$dst, $src2}",
884 [(set VR128:$dst, (int_x86_sse2_cvtsi2sd VR128:$src1,
886 def Int_CVTSI2SDrm: SDI<0x2A, MRMSrcMem,
887 (ops VR128:$dst, VR128:$src1, i32mem:$src2),
888 "cvtsi2sd {$src2, $dst|$dst, $src2}",
889 [(set VR128:$dst, (int_x86_sse2_cvtsi2sd VR128:$src1,
890 (loadi32 addr:$src2)))]>;
891 def Int_CVTSD2SSrr: SDI<0x5A, MRMSrcReg,
892 (ops VR128:$dst, VR128:$src1, VR128:$src2),
893 "cvtsd2ss {$src2, $dst|$dst, $src2}",
894 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
896 def Int_CVTSD2SSrm: SDI<0x5A, MRMSrcMem,
897 (ops VR128:$dst, VR128:$src1, f64mem:$src2),
898 "cvtsd2ss {$src2, $dst|$dst, $src2}",
899 [(set VR128:$dst, (int_x86_sse2_cvtsd2ss VR128:$src1,
900 (load addr:$src2)))]>;
901 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
902 (ops VR128:$dst, VR128:$src1, VR128:$src2),
903 "cvtss2sd {$src2, $dst|$dst, $src2}",
904 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
907 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
908 (ops VR128:$dst, VR128:$src1, f32mem:$src2),
909 "cvtss2sd {$src2, $dst|$dst, $src2}",
910 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
911 (load addr:$src2)))]>, XS,
916 let isTwoAddress = 1 in {
917 let isCommutable = 1 in {
918 def ADDPSrr : PSI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
919 "addps {$src2, $dst|$dst, $src2}",
920 [(set VR128:$dst, (v4f32 (fadd VR128:$src1, VR128:$src2)))]>;
921 def ADDPDrr : PDI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
922 "addpd {$src2, $dst|$dst, $src2}",
923 [(set VR128:$dst, (v2f64 (fadd VR128:$src1, VR128:$src2)))]>;
924 def MULPSrr : PSI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
925 "mulps {$src2, $dst|$dst, $src2}",
926 [(set VR128:$dst, (v4f32 (fmul VR128:$src1, VR128:$src2)))]>;
927 def MULPDrr : PDI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
928 "mulpd {$src2, $dst|$dst, $src2}",
929 [(set VR128:$dst, (v2f64 (fmul VR128:$src1, VR128:$src2)))]>;
932 def ADDPSrm : PSI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
933 "addps {$src2, $dst|$dst, $src2}",
934 [(set VR128:$dst, (v4f32 (fadd VR128:$src1,
935 (load addr:$src2))))]>;
936 def ADDPDrm : PDI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
937 "addpd {$src2, $dst|$dst, $src2}",
938 [(set VR128:$dst, (v2f64 (fadd VR128:$src1,
939 (load addr:$src2))))]>;
940 def MULPSrm : PSI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
941 "mulps {$src2, $dst|$dst, $src2}",
942 [(set VR128:$dst, (v4f32 (fmul VR128:$src1,
943 (load addr:$src2))))]>;
944 def MULPDrm : PDI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
945 "mulpd {$src2, $dst|$dst, $src2}",
946 [(set VR128:$dst, (v2f64 (fmul VR128:$src1,
947 (load addr:$src2))))]>;
949 def DIVPSrr : PSI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
950 "divps {$src2, $dst|$dst, $src2}",
951 [(set VR128:$dst, (v4f32 (fdiv VR128:$src1, VR128:$src2)))]>;
952 def DIVPSrm : PSI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
953 "divps {$src2, $dst|$dst, $src2}",
954 [(set VR128:$dst, (v4f32 (fdiv VR128:$src1,
955 (load addr:$src2))))]>;
956 def DIVPDrr : PDI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
957 "divpd {$src2, $dst|$dst, $src2}",
958 [(set VR128:$dst, (v2f64 (fdiv VR128:$src1, VR128:$src2)))]>;
959 def DIVPDrm : PDI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
960 "divpd {$src2, $dst|$dst, $src2}",
961 [(set VR128:$dst, (v2f64 (fdiv VR128:$src1,
962 (load addr:$src2))))]>;
964 def SUBPSrr : PSI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
965 "subps {$src2, $dst|$dst, $src2}",
966 [(set VR128:$dst, (v4f32 (fsub VR128:$src1, VR128:$src2)))]>;
967 def SUBPSrm : PSI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
968 "subps {$src2, $dst|$dst, $src2}",
969 [(set VR128:$dst, (v4f32 (fsub VR128:$src1,
970 (load addr:$src2))))]>;
971 def SUBPDrr : PDI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
972 "subpd {$src2, $dst|$dst, $src2}",
973 [(set VR128:$dst, (v2f64 (fsub VR128:$src1, VR128:$src2)))]>;
974 def SUBPDrm : PDI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
975 "subpd {$src2, $dst|$dst, $src2}",
976 [(set VR128:$dst, (v2f64 (fsub VR128:$src1,
977 (load addr:$src2))))]>;
979 def ADDSUBPSrr : S3DI<0xD0, MRMSrcReg,
980 (ops VR128:$dst, VR128:$src1, VR128:$src2),
981 "addsubps {$src2, $dst|$dst, $src2}",
982 [(set VR128:$dst, (int_x86_sse3_addsub_ps VR128:$src1,
984 def ADDSUBPSrm : S3DI<0xD0, MRMSrcMem,
985 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
986 "addsubps {$src2, $dst|$dst, $src2}",
987 [(set VR128:$dst, (int_x86_sse3_addsub_ps VR128:$src1,
988 (load addr:$src2)))]>;
989 def ADDSUBPDrr : S3I<0xD0, MRMSrcReg,
990 (ops VR128:$dst, VR128:$src1, VR128:$src2),
991 "addsubpd {$src2, $dst|$dst, $src2}",
992 [(set VR128:$dst, (int_x86_sse3_addsub_pd VR128:$src1,
994 def ADDSUBPDrm : S3I<0xD0, MRMSrcMem,
995 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
996 "addsubpd {$src2, $dst|$dst, $src2}",
997 [(set VR128:$dst, (int_x86_sse3_addsub_pd VR128:$src1,
998 (load addr:$src2)))]>;
1001 def SQRTPSr : PS_Intr<0x51, "sqrtps", int_x86_sse_sqrt_ps>;
1002 def SQRTPSm : PS_Intm<0x51, "sqrtps", int_x86_sse_sqrt_ps>;
1003 def SQRTPDr : PD_Intr<0x51, "sqrtpd", int_x86_sse2_sqrt_pd>;
1004 def SQRTPDm : PD_Intm<0x51, "sqrtpd", int_x86_sse2_sqrt_pd>;
1006 def RSQRTPSr : PS_Intr<0x52, "rsqrtps", int_x86_sse_rsqrt_ps>;
1007 def RSQRTPSm : PS_Intm<0x52, "rsqrtps", int_x86_sse_rsqrt_ps>;
1008 def RCPPSr : PS_Intr<0x53, "rcpps", int_x86_sse_rcp_ps>;
1009 def RCPPSm : PS_Intm<0x53, "rcpps", int_x86_sse_rcp_ps>;
1011 let isTwoAddress = 1 in {
1012 let isCommutable = 1 in {
1013 def MAXPSrr : PS_Intrr<0x5F, "maxps", int_x86_sse_max_ps>;
1014 def MAXPDrr : PD_Intrr<0x5F, "maxpd", int_x86_sse2_max_pd>;
1015 def MINPSrr : PS_Intrr<0x5D, "minps", int_x86_sse_min_ps>;
1016 def MINPDrr : PD_Intrr<0x5D, "minpd", int_x86_sse2_min_pd>;
1018 def MAXPSrm : PS_Intrm<0x5F, "maxps", int_x86_sse_max_ps>;
1019 def MAXPDrm : PD_Intrm<0x5F, "maxpd", int_x86_sse2_max_pd>;
1020 def MINPSrm : PS_Intrm<0x5D, "minps", int_x86_sse_min_ps>;
1021 def MINPDrm : PD_Intrm<0x5D, "minpd", int_x86_sse2_min_pd>;
1025 let isTwoAddress = 1 in {
1026 let isCommutable = 1 in {
1027 def ANDPSrr : PSI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1028 "andps {$src2, $dst|$dst, $src2}",
1029 [(set VR128:$dst, (v2i64 (and VR128:$src1, VR128:$src2)))]>;
1030 def ANDPDrr : PDI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1031 "andpd {$src2, $dst|$dst, $src2}",
1033 (and (bc_v2i64 (v2f64 VR128:$src1)),
1034 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1035 def ORPSrr : PSI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1036 "orps {$src2, $dst|$dst, $src2}",
1037 [(set VR128:$dst, (v2i64 (or VR128:$src1, VR128:$src2)))]>;
1038 def ORPDrr : PDI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1039 "orpd {$src2, $dst|$dst, $src2}",
1041 (or (bc_v2i64 (v2f64 VR128:$src1)),
1042 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1043 def XORPSrr : PSI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1044 "xorps {$src2, $dst|$dst, $src2}",
1045 [(set VR128:$dst, (v2i64 (xor VR128:$src1, VR128:$src2)))]>;
1046 def XORPDrr : PDI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1047 "xorpd {$src2, $dst|$dst, $src2}",
1049 (xor (bc_v2i64 (v2f64 VR128:$src1)),
1050 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1052 def ANDPSrm : PSI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1053 "andps {$src2, $dst|$dst, $src2}",
1054 [(set VR128:$dst, (and VR128:$src1,
1055 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1056 def ANDPDrm : PDI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1057 "andpd {$src2, $dst|$dst, $src2}",
1059 (and (bc_v2i64 (v2f64 VR128:$src1)),
1060 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1061 def ORPSrm : PSI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1062 "orps {$src2, $dst|$dst, $src2}",
1063 [(set VR128:$dst, (or VR128:$src1,
1064 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1065 def ORPDrm : PDI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1066 "orpd {$src2, $dst|$dst, $src2}",
1068 (or (bc_v2i64 (v2f64 VR128:$src1)),
1069 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1070 def XORPSrm : PSI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1071 "xorps {$src2, $dst|$dst, $src2}",
1072 [(set VR128:$dst, (xor VR128:$src1,
1073 (bc_v2i64 (loadv4f32 addr:$src2))))]>;
1074 def XORPDrm : PDI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1075 "xorpd {$src2, $dst|$dst, $src2}",
1077 (xor (bc_v2i64 (v2f64 VR128:$src1)),
1078 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1079 def ANDNPSrr : PSI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1080 "andnps {$src2, $dst|$dst, $src2}",
1081 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1082 (bc_v2i64 (v4i32 immAllOnesV))),
1084 def ANDNPSrm : PSI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
1085 "andnps {$src2, $dst|$dst, $src2}",
1086 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1087 (bc_v2i64 (v4i32 immAllOnesV))),
1088 (bc_v2i64 (loadv4f32 addr:$src2)))))]>;
1089 def ANDNPDrr : PDI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1090 "andnpd {$src2, $dst|$dst, $src2}",
1092 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1093 (bc_v2i64 (v2f64 VR128:$src2))))]>;
1094 def ANDNPDrm : PDI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
1095 "andnpd {$src2, $dst|$dst, $src2}",
1097 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1098 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
1101 let isTwoAddress = 1 in {
1102 def CMPPSrri : PSIi8<0xC2, MRMSrcReg,
1103 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
1104 "cmp${cc}ps {$src, $dst|$dst, $src}",
1105 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
1106 VR128:$src, imm:$cc))]>;
1107 def CMPPSrmi : PSIi8<0xC2, MRMSrcMem,
1108 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
1109 "cmp${cc}ps {$src, $dst|$dst, $src}",
1110 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
1111 (load addr:$src), imm:$cc))]>;
1112 def CMPPDrri : PDIi8<0xC2, MRMSrcReg,
1113 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
1114 "cmp${cc}pd {$src, $dst|$dst, $src}",
1115 [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
1116 VR128:$src, imm:$cc))]>;
1117 def CMPPDrmi : PDIi8<0xC2, MRMSrcMem,
1118 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
1119 "cmp${cc}pd {$src, $dst|$dst, $src}",
1120 [(set VR128:$dst, (int_x86_sse2_cmp_pd VR128:$src1,
1121 (load addr:$src), imm:$cc))]>;
1124 // Shuffle and unpack instructions
1125 let isTwoAddress = 1 in {
1126 let isConvertibleToThreeAddress = 1 in // Convert to pshufd
1127 def SHUFPSrri : PSIi8<0xC6, MRMSrcReg,
1128 (ops VR128:$dst, VR128:$src1, VR128:$src2, i32i8imm:$src3),
1129 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1130 [(set VR128:$dst, (v4f32 (vector_shuffle
1131 VR128:$src1, VR128:$src2,
1132 SHUFP_shuffle_mask:$src3)))]>;
1133 def SHUFPSrmi : PSIi8<0xC6, MRMSrcMem,
1134 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i32i8imm:$src3),
1135 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1136 [(set VR128:$dst, (v4f32 (vector_shuffle
1137 VR128:$src1, (load addr:$src2),
1138 SHUFP_shuffle_mask:$src3)))]>;
1139 def SHUFPDrri : PDIi8<0xC6, MRMSrcReg,
1140 (ops VR128:$dst, VR128:$src1, VR128:$src2, i8imm:$src3),
1141 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1142 [(set VR128:$dst, (v2f64 (vector_shuffle
1143 VR128:$src1, VR128:$src2,
1144 SHUFP_shuffle_mask:$src3)))]>;
1145 def SHUFPDrmi : PDIi8<0xC6, MRMSrcMem,
1146 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i8imm:$src3),
1147 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1148 [(set VR128:$dst, (v2f64 (vector_shuffle
1149 VR128:$src1, (load addr:$src2),
1150 SHUFP_shuffle_mask:$src3)))]>;
1152 let AddedComplexity = 10 in {
1153 def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
1154 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1155 "unpckhps {$src2, $dst|$dst, $src2}",
1156 [(set VR128:$dst, (v4f32 (vector_shuffle
1157 VR128:$src1, VR128:$src2,
1158 UNPCKH_shuffle_mask)))]>;
1159 def UNPCKHPSrm : PSI<0x15, MRMSrcMem,
1160 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1161 "unpckhps {$src2, $dst|$dst, $src2}",
1162 [(set VR128:$dst, (v4f32 (vector_shuffle
1163 VR128:$src1, (load addr:$src2),
1164 UNPCKH_shuffle_mask)))]>;
1165 def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
1166 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1167 "unpckhpd {$src2, $dst|$dst, $src2}",
1168 [(set VR128:$dst, (v2f64 (vector_shuffle
1169 VR128:$src1, VR128:$src2,
1170 UNPCKH_shuffle_mask)))]>;
1171 def UNPCKHPDrm : PDI<0x15, MRMSrcMem,
1172 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1173 "unpckhpd {$src2, $dst|$dst, $src2}",
1174 [(set VR128:$dst, (v2f64 (vector_shuffle
1175 VR128:$src1, (load addr:$src2),
1176 UNPCKH_shuffle_mask)))]>;
1178 def UNPCKLPSrr : PSI<0x14, MRMSrcReg,
1179 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1180 "unpcklps {$src2, $dst|$dst, $src2}",
1181 [(set VR128:$dst, (v4f32 (vector_shuffle
1182 VR128:$src1, VR128:$src2,
1183 UNPCKL_shuffle_mask)))]>;
1184 def UNPCKLPSrm : PSI<0x14, MRMSrcMem,
1185 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1186 "unpcklps {$src2, $dst|$dst, $src2}",
1187 [(set VR128:$dst, (v4f32 (vector_shuffle
1188 VR128:$src1, (load addr:$src2),
1189 UNPCKL_shuffle_mask)))]>;
1190 def UNPCKLPDrr : PDI<0x14, MRMSrcReg,
1191 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1192 "unpcklpd {$src2, $dst|$dst, $src2}",
1193 [(set VR128:$dst, (v2f64 (vector_shuffle
1194 VR128:$src1, VR128:$src2,
1195 UNPCKL_shuffle_mask)))]>;
1196 def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
1197 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1198 "unpcklpd {$src2, $dst|$dst, $src2}",
1199 [(set VR128:$dst, (v2f64 (vector_shuffle
1200 VR128:$src1, (load addr:$src2),
1201 UNPCKL_shuffle_mask)))]>;
1202 } // AddedComplexity
1207 class S3D_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
1208 : S3DI<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1209 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1210 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, VR128:$src2)))]>;
1211 class S3D_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
1212 : S3DI<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1213 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1214 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, (load addr:$src2))))]>;
1215 class S3_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId>
1216 : S3I<o, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1217 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1218 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, VR128:$src2)))]>;
1219 class S3_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId>
1220 : S3I<o, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1221 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2}"),
1222 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, (load addr:$src2))))]>;
1224 let isTwoAddress = 1 in {
1225 def HADDPSrr : S3D_Intrr<0x7C, "haddps", int_x86_sse3_hadd_ps>;
1226 def HADDPSrm : S3D_Intrm<0x7C, "haddps", int_x86_sse3_hadd_ps>;
1227 def HADDPDrr : S3_Intrr <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
1228 def HADDPDrm : S3_Intrm <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
1229 def HSUBPSrr : S3D_Intrr<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
1230 def HSUBPSrm : S3D_Intrm<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
1231 def HSUBPDrr : S3_Intrr <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
1232 def HSUBPDrm : S3_Intrm <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
1235 //===----------------------------------------------------------------------===//
1236 // SSE integer instructions
1237 //===----------------------------------------------------------------------===//
1239 // Move Instructions
1240 def MOVDQArr : PDI<0x6F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
1241 "movdqa {$src, $dst|$dst, $src}", []>;
1242 def MOVDQArm : PDI<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1243 "movdqa {$src, $dst|$dst, $src}",
1244 [(set VR128:$dst, (loadv2i64 addr:$src))]>;
1245 def MOVDQAmr : PDI<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1246 "movdqa {$src, $dst|$dst, $src}",
1247 [(store (v2i64 VR128:$src), addr:$dst)]>;
1248 def MOVDQUrm : I<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1249 "movdqu {$src, $dst|$dst, $src}",
1250 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
1251 XS, Requires<[HasSSE2]>;
1252 def MOVDQUmr : I<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1253 "movdqu {$src, $dst|$dst, $src}",
1254 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
1255 XS, Requires<[HasSSE2]>;
1256 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1257 "lddqu {$src, $dst|$dst, $src}",
1258 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
1261 let isTwoAddress = 1 in {
1262 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
1263 bit Commutable = 0> {
1264 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1265 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1266 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]> {
1267 let isCommutable = Commutable;
1269 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1270 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1271 [(set VR128:$dst, (IntId VR128:$src1,
1272 (bitconvert (loadv2i64 addr:$src2))))]>;
1276 let isTwoAddress = 1 in {
1277 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
1278 string OpcodeStr, Intrinsic IntId> {
1279 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1280 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1281 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
1282 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1283 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1284 [(set VR128:$dst, (IntId VR128:$src1,
1285 (bitconvert (loadv2i64 addr:$src2))))]>;
1286 def ri : PDIi8<opc2, ImmForm, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1287 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1288 [(set VR128:$dst, (IntId VR128:$src1,
1289 (scalar_to_vector (i32 imm:$src2))))]>;
1294 let isTwoAddress = 1 in {
1295 /// PDI_binop_rm - Simple SSE2 binary operator.
1296 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
1297 ValueType OpVT, bit Commutable = 0> {
1298 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1299 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1300 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]> {
1301 let isCommutable = Commutable;
1303 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1304 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1305 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
1306 (bitconvert (loadv2i64 addr:$src2)))))]>;
1309 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
1311 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
1312 /// to collapse (bitconvert VT to VT) into its operand.
1314 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
1315 bit Commutable = 0> {
1316 def rr : PDI<opc, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1317 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1318 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]> {
1319 let isCommutable = Commutable;
1321 def rm : PDI<opc, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1322 !strconcat(OpcodeStr, " {$src2, $dst|$dst, $src2"),
1323 [(set VR128:$dst, (OpNode VR128:$src1,(loadv2i64 addr:$src2)))]>;
1328 // 128-bit Integer Arithmetic
1330 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
1331 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
1332 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
1333 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
1335 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
1336 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
1337 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
1338 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
1340 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
1341 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
1342 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
1343 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
1345 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
1346 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
1347 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
1348 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
1350 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
1352 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
1353 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w , 1>;
1354 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
1356 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
1358 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
1359 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
1362 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
1363 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
1364 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
1365 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
1366 defm PSADBW : PDI_binop_rm_int<0xE0, "psadbw", int_x86_sse2_psad_bw, 1>;
1369 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw", int_x86_sse2_psll_w>;
1370 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld", int_x86_sse2_psll_d>;
1371 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq", int_x86_sse2_psll_q>;
1373 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw", int_x86_sse2_psrl_w>;
1374 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld", int_x86_sse2_psrl_d>;
1375 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq", int_x86_sse2_psrl_q>;
1377 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw", int_x86_sse2_psra_w>;
1378 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad", int_x86_sse2_psra_d>;
1379 // PSRAQ doesn't exist in SSE[1-3].
1382 // 128-bit logical shifts.
1383 let isTwoAddress = 1 in {
1384 def PSLLDQri : PDIi8<0x73, MRM7r, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1385 "pslldq {$src2, $dst|$dst, $src2}", []>;
1386 def PSRLDQri : PDIi8<0x73, MRM3r, (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1387 "psrldq {$src2, $dst|$dst, $src2}", []>;
1388 // PSRADQri doesn't exist in SSE[1-3].
1391 let Predicates = [HasSSE2] in {
1392 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
1393 (v2i64 (PSLLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
1394 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
1395 (v2i64 (PSRLDQri VR128:$src1, (PSxLDQ_imm imm:$src2)))>;
1399 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
1400 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or , 1>;
1401 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
1403 let isTwoAddress = 1 in {
1404 def PANDNrr : PDI<0xDF, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1405 "pandn {$src2, $dst|$dst, $src2}",
1406 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1409 def PANDNrm : PDI<0xDF, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1410 "pandn {$src2, $dst|$dst, $src2}",
1411 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1412 (load addr:$src2))))]>;
1415 // SSE2 Integer comparison
1416 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b>;
1417 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w>;
1418 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d>;
1419 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
1420 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
1421 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
1423 // Pack instructions
1424 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
1425 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
1426 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
1428 // Shuffle and unpack instructions
1429 def PSHUFDri : PDIi8<0x70, MRMSrcReg,
1430 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1431 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1432 [(set VR128:$dst, (v4i32 (vector_shuffle
1433 VR128:$src1, (undef),
1434 PSHUFD_shuffle_mask:$src2)))]>;
1435 def PSHUFDmi : PDIi8<0x70, MRMSrcMem,
1436 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1437 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1438 [(set VR128:$dst, (v4i32 (vector_shuffle
1439 (bc_v4i32(loadv2i64 addr:$src1)),
1441 PSHUFD_shuffle_mask:$src2)))]>;
1443 // SSE2 with ImmT == Imm8 and XS prefix.
1444 def PSHUFHWri : Ii8<0x70, MRMSrcReg,
1445 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1446 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1447 [(set VR128:$dst, (v8i16 (vector_shuffle
1448 VR128:$src1, (undef),
1449 PSHUFHW_shuffle_mask:$src2)))]>,
1450 XS, Requires<[HasSSE2]>;
1451 def PSHUFHWmi : Ii8<0x70, MRMSrcMem,
1452 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1453 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1454 [(set VR128:$dst, (v8i16 (vector_shuffle
1455 (bc_v8i16 (loadv2i64 addr:$src1)),
1457 PSHUFHW_shuffle_mask:$src2)))]>,
1458 XS, Requires<[HasSSE2]>;
1460 // SSE2 with ImmT == Imm8 and XD prefix.
1461 def PSHUFLWri : Ii8<0x70, MRMSrcReg,
1462 (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1463 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1464 [(set VR128:$dst, (v8i16 (vector_shuffle
1465 VR128:$src1, (undef),
1466 PSHUFLW_shuffle_mask:$src2)))]>,
1467 XD, Requires<[HasSSE2]>;
1468 def PSHUFLWmi : Ii8<0x70, MRMSrcMem,
1469 (ops VR128:$dst, i128mem:$src1, i32i8imm:$src2),
1470 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1471 [(set VR128:$dst, (v8i16 (vector_shuffle
1472 (bc_v8i16 (loadv2i64 addr:$src1)),
1474 PSHUFLW_shuffle_mask:$src2)))]>,
1475 XD, Requires<[HasSSE2]>;
1477 let isTwoAddress = 1 in {
1478 def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
1479 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1480 "punpcklbw {$src2, $dst|$dst, $src2}",
1482 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1483 UNPCKL_shuffle_mask)))]>;
1484 def PUNPCKLBWrm : PDI<0x60, MRMSrcMem,
1485 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1486 "punpcklbw {$src2, $dst|$dst, $src2}",
1488 (v16i8 (vector_shuffle VR128:$src1,
1489 (bc_v16i8 (loadv2i64 addr:$src2)),
1490 UNPCKL_shuffle_mask)))]>;
1491 def PUNPCKLWDrr : PDI<0x61, MRMSrcReg,
1492 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1493 "punpcklwd {$src2, $dst|$dst, $src2}",
1495 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1496 UNPCKL_shuffle_mask)))]>;
1497 def PUNPCKLWDrm : PDI<0x61, MRMSrcMem,
1498 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1499 "punpcklwd {$src2, $dst|$dst, $src2}",
1501 (v8i16 (vector_shuffle VR128:$src1,
1502 (bc_v8i16 (loadv2i64 addr:$src2)),
1503 UNPCKL_shuffle_mask)))]>;
1504 def PUNPCKLDQrr : PDI<0x62, MRMSrcReg,
1505 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1506 "punpckldq {$src2, $dst|$dst, $src2}",
1508 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1509 UNPCKL_shuffle_mask)))]>;
1510 def PUNPCKLDQrm : PDI<0x62, MRMSrcMem,
1511 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1512 "punpckldq {$src2, $dst|$dst, $src2}",
1514 (v4i32 (vector_shuffle VR128:$src1,
1515 (bc_v4i32 (loadv2i64 addr:$src2)),
1516 UNPCKL_shuffle_mask)))]>;
1517 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
1518 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1519 "punpcklqdq {$src2, $dst|$dst, $src2}",
1521 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1522 UNPCKL_shuffle_mask)))]>;
1523 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
1524 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1525 "punpcklqdq {$src2, $dst|$dst, $src2}",
1527 (v2i64 (vector_shuffle VR128:$src1,
1528 (loadv2i64 addr:$src2),
1529 UNPCKL_shuffle_mask)))]>;
1531 def PUNPCKHBWrr : PDI<0x68, MRMSrcReg,
1532 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1533 "punpckhbw {$src2, $dst|$dst, $src2}",
1535 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1536 UNPCKH_shuffle_mask)))]>;
1537 def PUNPCKHBWrm : PDI<0x68, MRMSrcMem,
1538 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1539 "punpckhbw {$src2, $dst|$dst, $src2}",
1541 (v16i8 (vector_shuffle VR128:$src1,
1542 (bc_v16i8 (loadv2i64 addr:$src2)),
1543 UNPCKH_shuffle_mask)))]>;
1544 def PUNPCKHWDrr : PDI<0x69, MRMSrcReg,
1545 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1546 "punpckhwd {$src2, $dst|$dst, $src2}",
1548 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1549 UNPCKH_shuffle_mask)))]>;
1550 def PUNPCKHWDrm : PDI<0x69, MRMSrcMem,
1551 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1552 "punpckhwd {$src2, $dst|$dst, $src2}",
1554 (v8i16 (vector_shuffle VR128:$src1,
1555 (bc_v8i16 (loadv2i64 addr:$src2)),
1556 UNPCKH_shuffle_mask)))]>;
1557 def PUNPCKHDQrr : PDI<0x6A, MRMSrcReg,
1558 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1559 "punpckhdq {$src2, $dst|$dst, $src2}",
1561 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1562 UNPCKH_shuffle_mask)))]>;
1563 def PUNPCKHDQrm : PDI<0x6A, MRMSrcMem,
1564 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1565 "punpckhdq {$src2, $dst|$dst, $src2}",
1567 (v4i32 (vector_shuffle VR128:$src1,
1568 (bc_v4i32 (loadv2i64 addr:$src2)),
1569 UNPCKH_shuffle_mask)))]>;
1570 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
1571 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1572 "punpckhqdq {$src2, $dst|$dst, $src2}",
1574 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1575 UNPCKH_shuffle_mask)))]>;
1576 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
1577 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1578 "punpckhqdq {$src2, $dst|$dst, $src2}",
1580 (v2i64 (vector_shuffle VR128:$src1,
1581 (loadv2i64 addr:$src2),
1582 UNPCKH_shuffle_mask)))]>;
1586 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
1587 (ops GR32:$dst, VR128:$src1, i32i8imm:$src2),
1588 "pextrw {$src2, $src1, $dst|$dst, $src1, $src2}",
1589 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
1590 (i32 imm:$src2)))]>;
1591 let isTwoAddress = 1 in {
1592 def PINSRWrri : PDIi8<0xC4, MRMSrcReg,
1593 (ops VR128:$dst, VR128:$src1, GR32:$src2, i32i8imm:$src3),
1594 "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
1595 [(set VR128:$dst, (v8i16 (X86pinsrw (v8i16 VR128:$src1),
1596 GR32:$src2, (iPTR imm:$src3))))]>;
1597 def PINSRWrmi : PDIi8<0xC4, MRMSrcMem,
1598 (ops VR128:$dst, VR128:$src1, i16mem:$src2, i32i8imm:$src3),
1599 "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
1601 (v8i16 (X86pinsrw (v8i16 VR128:$src1),
1602 (i32 (anyext (loadi16 addr:$src2))),
1603 (iPTR imm:$src3))))]>;
1606 //===----------------------------------------------------------------------===//
1607 // Miscellaneous Instructions
1608 //===----------------------------------------------------------------------===//
1611 def MOVMSKPSrr : PSI<0x50, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1612 "movmskps {$src, $dst|$dst, $src}",
1613 [(set GR32:$dst, (int_x86_sse_movmsk_ps VR128:$src))]>;
1614 def MOVMSKPDrr : PSI<0x50, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1615 "movmskpd {$src, $dst|$dst, $src}",
1616 [(set GR32:$dst, (int_x86_sse2_movmsk_pd VR128:$src))]>;
1618 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (ops GR32:$dst, VR128:$src),
1619 "pmovmskb {$src, $dst|$dst, $src}",
1620 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
1622 // Conditional store
1623 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (ops VR128:$src, VR128:$mask),
1624 "maskmovdqu {$mask, $src|$src, $mask}",
1625 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>,
1628 // Prefetching loads.
1629 // TODO: no intrinsics for these?
1630 def PREFETCHT0 : PSI<0x18, MRM1m, (ops i8mem:$src), "prefetcht0 $src", []>;
1631 def PREFETCHT1 : PSI<0x18, MRM2m, (ops i8mem:$src), "prefetcht1 $src", []>;
1632 def PREFETCHT2 : PSI<0x18, MRM3m, (ops i8mem:$src), "prefetcht2 $src", []>;
1633 def PREFETCHTNTA : PSI<0x18, MRM0m, (ops i8mem:$src), "prefetchtnta $src", []>;
1635 // Non-temporal stores
1636 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1637 "movntps {$src, $dst|$dst, $src}",
1638 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
1639 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1640 "movntpd {$src, $dst|$dst, $src}",
1641 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
1642 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (ops f128mem:$dst, VR128:$src),
1643 "movntdq {$src, $dst|$dst, $src}",
1644 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
1645 def MOVNTImr : I<0xC3, MRMDestMem, (ops i32mem:$dst, GR32:$src),
1646 "movnti {$src, $dst|$dst, $src}",
1647 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
1648 TB, Requires<[HasSSE2]>;
1651 def CLFLUSH : I<0xAE, MRM7m, (ops i8mem:$src),
1652 "clflush $src", [(int_x86_sse2_clflush addr:$src)]>,
1653 TB, Requires<[HasSSE2]>;
1655 // Load, store, and memory fence
1656 def SFENCE : PSI<0xAE, MRM7m, (ops), "sfence", [(int_x86_sse_sfence)]>;
1657 def LFENCE : I<0xAE, MRM5m, (ops),
1658 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
1659 def MFENCE : I<0xAE, MRM6m, (ops),
1660 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
1663 def LDMXCSR : I<0xAE, MRM5m, (ops i32mem:$src),
1665 [(int_x86_sse_ldmxcsr addr:$src)]>, TB, Requires<[HasSSE1]>;
1666 def STMXCSR : I<0xAE, MRM3m, (ops i32mem:$dst),
1668 [(int_x86_sse_stmxcsr addr:$dst)]>, TB, Requires<[HasSSE1]>;
1670 // Thread synchronization
1671 def MONITOR : I<0xC8, RawFrm, (ops), "monitor",
1672 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
1673 def MWAIT : I<0xC9, RawFrm, (ops), "mwait",
1674 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
1676 //===----------------------------------------------------------------------===//
1677 // Alias Instructions
1678 //===----------------------------------------------------------------------===//
1680 // Alias instructions that map zero vector to pxor / xorp* for sse.
1681 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
1682 def V_SET0 : PSI<0x57, MRMInitReg, (ops VR128:$dst),
1684 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
1686 def V_SETALLONES : PDI<0x76, MRMInitReg, (ops VR128:$dst),
1687 "pcmpeqd $dst, $dst",
1688 [(set VR128:$dst, (v2f64 immAllOnesV))]>;
1690 // FR32 / FR64 to 128-bit vector conversion.
1691 def MOVSS2PSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, FR32:$src),
1692 "movss {$src, $dst|$dst, $src}",
1694 (v4f32 (scalar_to_vector FR32:$src)))]>;
1695 def MOVSS2PSrm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
1696 "movss {$src, $dst|$dst, $src}",
1698 (v4f32 (scalar_to_vector (loadf32 addr:$src))))]>;
1699 def MOVSD2PDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, FR64:$src),
1700 "movsd {$src, $dst|$dst, $src}",
1702 (v2f64 (scalar_to_vector FR64:$src)))]>;
1703 def MOVSD2PDrm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
1704 "movsd {$src, $dst|$dst, $src}",
1706 (v2f64 (scalar_to_vector (loadf64 addr:$src))))]>;
1708 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, GR32:$src),
1709 "movd {$src, $dst|$dst, $src}",
1711 (v4i32 (scalar_to_vector GR32:$src)))]>;
1712 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1713 "movd {$src, $dst|$dst, $src}",
1715 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
1716 // SSE2 instructions with XS prefix
1717 def MOVQI2PQIrr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR64:$src),
1718 "movq {$src, $dst|$dst, $src}",
1720 (v2i64 (scalar_to_vector VR64:$src)))]>, XS,
1721 Requires<[HasSSE2]>;
1722 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
1723 "movq {$src, $dst|$dst, $src}",
1725 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
1726 Requires<[HasSSE2]>;
1727 // FIXME: may not be able to eliminate this movss with coalescing the src and
1728 // dest register classes are different. We really want to write this pattern
1730 // def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
1731 // (f32 FR32:$src)>;
1732 def MOVPS2SSrr : SSI<0x10, MRMSrcReg, (ops FR32:$dst, VR128:$src),
1733 "movss {$src, $dst|$dst, $src}",
1734 [(set FR32:$dst, (vector_extract (v4f32 VR128:$src),
1736 def MOVPS2SSmr : SSI<0x11, MRMDestMem, (ops f32mem:$dst, VR128:$src),
1737 "movss {$src, $dst|$dst, $src}",
1738 [(store (f32 (vector_extract (v4f32 VR128:$src),
1739 (iPTR 0))), addr:$dst)]>;
1740 def MOVPD2SDrr : SDI<0x10, MRMSrcReg, (ops FR64:$dst, VR128:$src),
1741 "movsd {$src, $dst|$dst, $src}",
1742 [(set FR64:$dst, (vector_extract (v2f64 VR128:$src),
1744 def MOVPD2SDmr : SDI<0x11, MRMDestMem, (ops f64mem:$dst, VR128:$src),
1745 "movsd {$src, $dst|$dst, $src}",
1746 [(store (f64 (vector_extract (v2f64 VR128:$src),
1747 (iPTR 0))), addr:$dst)]>;
1748 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (ops GR32:$dst, VR128:$src),
1749 "movd {$src, $dst|$dst, $src}",
1750 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
1752 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (ops i32mem:$dst, VR128:$src),
1753 "movd {$src, $dst|$dst, $src}",
1754 [(store (i32 (vector_extract (v4i32 VR128:$src),
1755 (iPTR 0))), addr:$dst)]>;
1757 // Move to lower bits of a VR128, leaving upper bits alone.
1758 // Three operand (but two address) aliases.
1759 let isTwoAddress = 1 in {
1760 def MOVLSS2PSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR32:$src2),
1761 "movss {$src2, $dst|$dst, $src2}", []>;
1762 def MOVLSD2PDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR64:$src2),
1763 "movsd {$src2, $dst|$dst, $src2}", []>;
1765 let AddedComplexity = 20 in {
1766 def MOVLPSrr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1767 "movss {$src2, $dst|$dst, $src2}",
1769 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
1770 MOVL_shuffle_mask)))]>;
1771 def MOVLPDrr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1772 "movsd {$src2, $dst|$dst, $src2}",
1774 (v2f64 (vector_shuffle VR128:$src1, VR128:$src2,
1775 MOVL_shuffle_mask)))]>;
1779 // Store / copy lower 64-bits of a XMM register.
1780 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (ops i64mem:$dst, VR128:$src),
1781 "movq {$src, $dst|$dst, $src}",
1782 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
1784 // Move to lower bits of a VR128 and zeroing upper bits.
1785 // Loading from memory automatically zeroing upper bits.
1786 let AddedComplexity = 20 in {
1787 def MOVZSS2PSrm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
1788 "movss {$src, $dst|$dst, $src}",
1789 [(set VR128:$dst, (v4f32 (vector_shuffle immAllZerosV,
1790 (v4f32 (scalar_to_vector (loadf32 addr:$src))),
1791 MOVL_shuffle_mask)))]>;
1792 def MOVZSD2PDrm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
1793 "movsd {$src, $dst|$dst, $src}",
1794 [(set VR128:$dst, (v2f64 (vector_shuffle immAllZerosV,
1795 (v2f64 (scalar_to_vector (loadf64 addr:$src))),
1796 MOVL_shuffle_mask)))]>;
1797 // movd / movq to XMM register zero-extends
1798 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, GR32:$src),
1799 "movd {$src, $dst|$dst, $src}",
1800 [(set VR128:$dst, (v4i32 (vector_shuffle immAllZerosV,
1801 (v4i32 (scalar_to_vector GR32:$src)),
1802 MOVL_shuffle_mask)))]>;
1803 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1804 "movd {$src, $dst|$dst, $src}",
1805 [(set VR128:$dst, (v4i32 (vector_shuffle immAllZerosV,
1806 (v4i32 (scalar_to_vector (loadi32 addr:$src))),
1807 MOVL_shuffle_mask)))]>;
1808 // Moving from XMM to XMM but still clear upper 64 bits.
1809 def MOVZQI2PQIrr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR128:$src),
1810 "movq {$src, $dst|$dst, $src}",
1811 [(set VR128:$dst, (int_x86_sse2_movl_dq VR128:$src))]>,
1812 XS, Requires<[HasSSE2]>;
1813 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
1814 "movq {$src, $dst|$dst, $src}",
1815 [(set VR128:$dst, (int_x86_sse2_movl_dq
1816 (bitconvert (loadv2i64 addr:$src))))]>,
1817 XS, Requires<[HasSSE2]>;
1820 //===----------------------------------------------------------------------===//
1821 // Non-Instruction Patterns
1822 //===----------------------------------------------------------------------===//
1824 // 128-bit vector undef's.
1825 def : Pat<(v2f64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1826 def : Pat<(v16i8 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1827 def : Pat<(v8i16 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1828 def : Pat<(v4i32 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1829 def : Pat<(v2i64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1831 // 128-bit vector all zero's.
1832 def : Pat<(v16i8 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1833 def : Pat<(v8i16 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1834 def : Pat<(v4i32 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1835 def : Pat<(v2i64 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1836 def : Pat<(v2f64 immAllZerosV), (V_SET0)>, Requires<[HasSSE2]>;
1838 // 128-bit vector all one's.
1839 def : Pat<(v16i8 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1840 def : Pat<(v8i16 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1841 def : Pat<(v4i32 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1842 def : Pat<(v2i64 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE2]>;
1843 def : Pat<(v4f32 immAllOnesV), (V_SETALLONES)>, Requires<[HasSSE1]>;
1845 // Store 128-bit integer vector values.
1846 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1847 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1848 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1849 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1850 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1851 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1853 // Scalar to v8i16 / v16i8. The source may be a GR32, but only the lower 8 or
1855 def : Pat<(v8i16 (X86s2vec GR32:$src)), (MOVDI2PDIrr GR32:$src)>,
1856 Requires<[HasSSE2]>;
1857 def : Pat<(v16i8 (X86s2vec GR32:$src)), (MOVDI2PDIrr GR32:$src)>,
1858 Requires<[HasSSE2]>;
1861 let Predicates = [HasSSE2] in {
1862 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
1863 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
1864 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
1865 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
1866 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
1867 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
1868 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
1869 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
1870 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
1871 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
1872 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
1873 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
1874 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
1875 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
1876 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
1877 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
1878 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
1879 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
1880 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
1881 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
1882 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
1883 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
1884 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
1885 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
1886 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
1887 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
1888 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
1889 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
1890 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
1891 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
1894 // Move scalar to XMM zero-extended
1895 // movd to XMM register zero-extends
1896 let AddedComplexity = 20 in {
1897 def : Pat<(v8i16 (vector_shuffle immAllZerosV,
1898 (v8i16 (X86s2vec GR32:$src)), MOVL_shuffle_mask)),
1899 (MOVZDI2PDIrr GR32:$src)>, Requires<[HasSSE2]>;
1900 def : Pat<(v16i8 (vector_shuffle immAllZerosV,
1901 (v16i8 (X86s2vec GR32:$src)), MOVL_shuffle_mask)),
1902 (MOVZDI2PDIrr GR32:$src)>, Requires<[HasSSE2]>;
1903 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
1904 def : Pat<(v2f64 (vector_shuffle immAllZerosV,
1905 (v2f64 (scalar_to_vector FR64:$src)), MOVL_shuffle_mask)),
1906 (MOVLSD2PDrr (V_SET0), FR64:$src)>, Requires<[HasSSE2]>;
1907 def : Pat<(v4f32 (vector_shuffle immAllZerosV,
1908 (v4f32 (scalar_to_vector FR32:$src)), MOVL_shuffle_mask)),
1909 (MOVLSS2PSrr (V_SET0), FR32:$src)>, Requires<[HasSSE2]>;
1912 // Splat v2f64 / v2i64
1913 let AddedComplexity = 10 in {
1914 def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
1915 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1916 def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), SSE_splat_v2_mask:$sm),
1917 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1921 def : Pat<(vector_shuffle (v4f32 VR128:$src), (undef), SSE_splat_mask:$sm),
1922 (SHUFPSrri VR128:$src, VR128:$src, SSE_splat_mask:$sm)>,
1923 Requires<[HasSSE1]>;
1925 // Special unary SHUFPSrri case.
1926 // FIXME: when we want non two-address code, then we should use PSHUFD?
1927 def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef),
1928 SHUFP_unary_shuffle_mask:$sm),
1929 (SHUFPSrri VR128:$src1, VR128:$src1, SHUFP_unary_shuffle_mask:$sm)>,
1930 Requires<[HasSSE1]>;
1931 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
1932 def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef),
1933 SHUFP_unary_shuffle_mask:$sm),
1934 (PSHUFDmi addr:$src1, SHUFP_unary_shuffle_mask:$sm)>,
1935 Requires<[HasSSE2]>;
1936 // Special binary v4i32 shuffle cases with SHUFPS.
1937 def : Pat<(vector_shuffle (v4i32 VR128:$src1), (v4i32 VR128:$src2),
1938 PSHUFD_binary_shuffle_mask:$sm),
1939 (SHUFPSrri VR128:$src1, VR128:$src2, PSHUFD_binary_shuffle_mask:$sm)>,
1940 Requires<[HasSSE2]>;
1941 def : Pat<(vector_shuffle (v4i32 VR128:$src1),
1942 (bc_v4i32 (loadv2i64 addr:$src2)), PSHUFD_binary_shuffle_mask:$sm),
1943 (SHUFPSrmi VR128:$src1, addr:$src2, PSHUFD_binary_shuffle_mask:$sm)>,
1944 Requires<[HasSSE2]>;
1946 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
1947 let AddedComplexity = 10 in {
1948 def : Pat<(v4f32 (vector_shuffle VR128:$src, (undef),
1949 UNPCKL_v_undef_shuffle_mask)),
1950 (UNPCKLPSrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1951 def : Pat<(v16i8 (vector_shuffle VR128:$src, (undef),
1952 UNPCKL_v_undef_shuffle_mask)),
1953 (PUNPCKLBWrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1954 def : Pat<(v8i16 (vector_shuffle VR128:$src, (undef),
1955 UNPCKL_v_undef_shuffle_mask)),
1956 (PUNPCKLWDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
1957 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1958 UNPCKL_v_undef_shuffle_mask)),
1959 (PUNPCKLDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE1]>;
1962 let AddedComplexity = 20 in {
1963 // vector_shuffle v1, <undef> <1, 1, 3, 3>
1964 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1965 MOVSHDUP_shuffle_mask)),
1966 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
1967 def : Pat<(v4i32 (vector_shuffle (bc_v4i32 (loadv2i64 addr:$src)), (undef),
1968 MOVSHDUP_shuffle_mask)),
1969 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
1971 // vector_shuffle v1, <undef> <0, 0, 2, 2>
1972 def : Pat<(v4i32 (vector_shuffle VR128:$src, (undef),
1973 MOVSLDUP_shuffle_mask)),
1974 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
1975 def : Pat<(v4i32 (vector_shuffle (bc_v4i32 (loadv2i64 addr:$src)), (undef),
1976 MOVSLDUP_shuffle_mask)),
1977 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
1980 let AddedComplexity = 20 in {
1981 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1982 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1983 MOVHP_shuffle_mask)),
1984 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1986 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1987 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1988 MOVHLPS_shuffle_mask)),
1989 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1991 // vector_shuffle v1, undef <2, 3, ?, ?> using MOVHLPS
1992 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (undef),
1993 UNPCKH_shuffle_mask)),
1994 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1995 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (undef),
1996 UNPCKH_shuffle_mask)),
1997 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1999 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
2000 // vector_shuffle v1, (load v2) <0, 1, 4, 5> using MOVHPS
2001 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (loadv4f32 addr:$src2),
2002 MOVLP_shuffle_mask)),
2003 (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
2004 def : Pat<(v2f64 (vector_shuffle VR128:$src1, (loadv2f64 addr:$src2),
2005 MOVLP_shuffle_mask)),
2006 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2007 def : Pat<(v4f32 (vector_shuffle VR128:$src1, (loadv4f32 addr:$src2),
2008 MOVHP_shuffle_mask)),
2009 (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
2010 def : Pat<(v2f64 (vector_shuffle VR128:$src1, (loadv2f64 addr:$src2),
2011 MOVHP_shuffle_mask)),
2012 (MOVHPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2014 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)),
2015 MOVLP_shuffle_mask)),
2016 (MOVLPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2017 def : Pat<(v2i64 (vector_shuffle VR128:$src1, (loadv2i64 addr:$src2),
2018 MOVLP_shuffle_mask)),
2019 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2020 def : Pat<(v4i32 (vector_shuffle VR128:$src1, (bc_v4i32 (loadv2i64 addr:$src2)),
2021 MOVHP_shuffle_mask)),
2022 (MOVHPSrm VR128:$src1, addr:$src2)>, Requires<[HasSSE1]>;
2023 def : Pat<(v2i64 (vector_shuffle VR128:$src1, (loadv2i64 addr:$src2),
2024 MOVLP_shuffle_mask)),
2025 (MOVLPDrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2027 // Setting the lowest element in the vector.
2028 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
2029 MOVL_shuffle_mask)),
2030 (MOVLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2031 def : Pat<(v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
2032 MOVL_shuffle_mask)),
2033 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2035 // vector_shuffle v1, v2 <4, 5, 2, 3> using MOVLPDrr (movsd)
2036 def : Pat<(v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
2037 MOVLP_shuffle_mask)),
2038 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2039 def : Pat<(v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
2040 MOVLP_shuffle_mask)),
2041 (MOVLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2043 // Set lowest element and zero upper elements.
2044 def : Pat<(bc_v2i64 (vector_shuffle immAllZerosV,
2045 (v2f64 (scalar_to_vector (loadf64 addr:$src))),
2046 MOVL_shuffle_mask)),
2047 (MOVZQI2PQIrm addr:$src)>, Requires<[HasSSE2]>;
2050 // FIXME: Temporary workaround since 2-wide shuffle is broken.
2051 def : Pat<(int_x86_sse2_movs_d VR128:$src1, VR128:$src2),
2052 (v2f64 (MOVLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2053 def : Pat<(int_x86_sse2_loadh_pd VR128:$src1, addr:$src2),
2054 (v2f64 (MOVHPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2055 def : Pat<(int_x86_sse2_loadl_pd VR128:$src1, addr:$src2),
2056 (v2f64 (MOVLPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2057 def : Pat<(int_x86_sse2_shuf_pd VR128:$src1, VR128:$src2, imm:$src3),
2058 (v2f64 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$src3))>,
2059 Requires<[HasSSE2]>;
2060 def : Pat<(int_x86_sse2_shuf_pd VR128:$src1, (load addr:$src2), imm:$src3),
2061 (v2f64 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$src3))>,
2062 Requires<[HasSSE2]>;
2063 def : Pat<(int_x86_sse2_unpckh_pd VR128:$src1, VR128:$src2),
2064 (v2f64 (UNPCKHPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2065 def : Pat<(int_x86_sse2_unpckh_pd VR128:$src1, (load addr:$src2)),
2066 (v2f64 (UNPCKHPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2067 def : Pat<(int_x86_sse2_unpckl_pd VR128:$src1, VR128:$src2),
2068 (v2f64 (UNPCKLPDrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2069 def : Pat<(int_x86_sse2_unpckl_pd VR128:$src1, (load addr:$src2)),
2070 (v2f64 (UNPCKLPDrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2071 def : Pat<(int_x86_sse2_punpckh_qdq VR128:$src1, VR128:$src2),
2072 (v2i64 (PUNPCKHQDQrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2073 def : Pat<(int_x86_sse2_punpckh_qdq VR128:$src1, (load addr:$src2)),
2074 (v2i64 (PUNPCKHQDQrm VR128:$src1, addr:$src2))>, Requires<[HasSSE2]>;
2075 def : Pat<(int_x86_sse2_punpckl_qdq VR128:$src1, VR128:$src2),
2076 (v2i64 (PUNPCKLQDQrr VR128:$src1, VR128:$src2))>, Requires<[HasSSE2]>;
2077 def : Pat<(int_x86_sse2_punpckl_qdq VR128:$src1, (load addr:$src2)),
2078 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2080 // Some special case pandn patterns.
2081 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
2083 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2084 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
2086 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2087 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
2089 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
2091 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
2092 (load addr:$src2))),
2093 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2094 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
2095 (load addr:$src2))),
2096 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2097 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
2098 (load addr:$src2))),
2099 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
2102 def : Pat<(v4f32 (X86loadu addr:$src)), (MOVUPSrm addr:$src)>,
2103 Requires<[HasSSE1]>;