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,
22 def X86fand : SDNode<"X86ISD::FAND", SDTFPBinOp,
23 [SDNPCommutative, SDNPAssociative]>;
24 def X86fxor : SDNode<"X86ISD::FXOR", SDTFPBinOp,
25 [SDNPCommutative, SDNPAssociative]>;
26 def X86s2vec : SDNode<"X86ISD::S2VEC",
27 SDTypeProfile<1, 1, []>, []>;
28 def X86zexts2vec : SDNode<"X86ISD::ZEXT_S2VEC",
29 SDTypeProfile<1, 1, []>, []>;
31 def SDTUnpckl : SDTypeProfile<1, 2,
32 [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>]>;
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 loadv16i8 : PatFrag<(ops node:$ptr), (v16i8 (load node:$ptr))>;
44 def loadv8i16 : PatFrag<(ops node:$ptr), (v8i16 (load node:$ptr))>;
45 def loadv4i32 : PatFrag<(ops node:$ptr), (v4i32 (load node:$ptr))>;
46 def loadv2i64 : PatFrag<(ops node:$ptr), (v2i64 (load node:$ptr))>;
48 def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
49 def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
50 def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
51 def bc_v8i16 : PatFrag<(ops node:$in), (v8i16 (bitconvert node:$in))>;
52 def bc_v4i32 : PatFrag<(ops node:$in), (v4i32 (bitconvert node:$in))>;
53 def bc_v2i64 : PatFrag<(ops node:$in), (v2i64 (bitconvert node:$in))>;
55 def fp32imm0 : PatLeaf<(f32 fpimm), [{
56 return N->isExactlyValue(+0.0);
59 // SHUFFLE_get_shuf_imm xform function: convert vector_shuffle mask to PSHUF*,
61 def SHUFFLE_get_shuf_imm : SDNodeXForm<build_vector, [{
62 return getI8Imm(X86::getShuffleSHUFImmediate(N));
65 // SHUFFLE_get_pshufhw_imm xform function: convert vector_shuffle mask to
67 def SHUFFLE_get_pshufhw_imm : SDNodeXForm<build_vector, [{
68 return getI8Imm(X86::getShufflePSHUFHWImmediate(N));
71 // SHUFFLE_get_pshuflw_imm xform function: convert vector_shuffle mask to
73 def SHUFFLE_get_pshuflw_imm : SDNodeXForm<build_vector, [{
74 return getI8Imm(X86::getShufflePSHUFLWImmediate(N));
77 def SSE_splat_mask : PatLeaf<(build_vector), [{
78 return X86::isSplatMask(N);
79 }], SHUFFLE_get_shuf_imm>;
81 def MOVLHPS_shuffle_mask : PatLeaf<(build_vector), [{
82 return X86::isMOVLHPSMask(N);
85 def MOVHLPS_shuffle_mask : PatLeaf<(build_vector), [{
86 return X86::isMOVHLPSMask(N);
89 def UNPCKL_shuffle_mask : PatLeaf<(build_vector), [{
90 return X86::isUNPCKLMask(N);
93 def UNPCKH_shuffle_mask : PatLeaf<(build_vector), [{
94 return X86::isUNPCKHMask(N);
97 def PSHUFD_shuffle_mask : PatLeaf<(build_vector), [{
98 return X86::isPSHUFDMask(N);
99 }], SHUFFLE_get_shuf_imm>;
101 def PSHUFHW_shuffle_mask : PatLeaf<(build_vector), [{
102 return X86::isPSHUFHWMask(N);
103 }], SHUFFLE_get_pshufhw_imm>;
105 def PSHUFLW_shuffle_mask : PatLeaf<(build_vector), [{
106 return X86::isPSHUFLWMask(N);
107 }], SHUFFLE_get_pshuflw_imm>;
109 // Only use PSHUF* for v4f32 if SHUFP does not match.
110 def PSHUFD_fp_shuffle_mask : PatLeaf<(build_vector), [{
111 return !X86::isSHUFPMask(N) &&
112 X86::isPSHUFDMask(N);
113 }], SHUFFLE_get_shuf_imm>;
115 def PSHUFHW_fp_shuffle_mask : PatLeaf<(build_vector), [{
116 return !X86::isSHUFPMask(N) &&
117 X86::isPSHUFHWMask(N);
118 }], SHUFFLE_get_pshufhw_imm>;
120 def PSHUFLW_fp_shuffle_mask : PatLeaf<(build_vector), [{
121 return !X86::isSHUFPMask(N) &&
122 X86::isPSHUFLWMask(N);
123 }], SHUFFLE_get_pshuflw_imm>;
125 def SHUFP_shuffle_mask : PatLeaf<(build_vector), [{
126 return X86::isSHUFPMask(N);
127 }], SHUFFLE_get_shuf_imm>;
129 // Only use SHUFP for v4i32 if PSHUF* do not match.
130 def SHUFP_int_shuffle_mask : PatLeaf<(build_vector), [{
131 return !X86::isPSHUFDMask(N) &&
132 !X86::isPSHUFHWMask(N) &&
133 !X86::isPSHUFLWMask(N) &&
135 }], SHUFFLE_get_shuf_imm>;
137 //===----------------------------------------------------------------------===//
138 // SSE scalar FP Instructions
139 //===----------------------------------------------------------------------===//
141 // Instruction templates
142 // SSI - SSE1 instructions with XS prefix.
143 // SDI - SSE2 instructions with XD prefix.
144 // PSI - SSE1 instructions with TB prefix.
145 // PDI - SSE2 instructions with TB and OpSize prefixes.
146 // PSIi8 - SSE1 instructions with ImmT == Imm8 and TB prefix.
147 // PDIi8 - SSE2 instructions with ImmT == Imm8 and TB and OpSize prefixes.
148 class SSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
149 : I<o, F, ops, asm, pattern>, XS, Requires<[HasSSE1]>;
150 class SDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
151 : I<o, F, ops, asm, pattern>, XD, Requires<[HasSSE2]>;
152 class PSI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
153 : I<o, F, ops, asm, pattern>, TB, Requires<[HasSSE1]>;
154 class PDI<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
155 : I<o, F, ops, asm, pattern>, TB, OpSize, Requires<[HasSSE2]>;
156 class PSIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
157 : X86Inst<o, F, Imm8, ops, asm>, TB, Requires<[HasSSE1]> {
158 let Pattern = pattern;
160 class PDIi8<bits<8> o, Format F, dag ops, string asm, list<dag> pattern>
161 : X86Inst<o, F, Imm8, ops, asm>, TB, OpSize, Requires<[HasSSE2]> {
162 let Pattern = pattern;
165 // Some 'special' instructions
166 def IMPLICIT_DEF_FR32 : I<0, Pseudo, (ops FR32:$dst),
167 "#IMPLICIT_DEF $dst",
168 [(set FR32:$dst, (undef))]>, Requires<[HasSSE2]>;
169 def IMPLICIT_DEF_FR64 : I<0, Pseudo, (ops FR64:$dst),
170 "#IMPLICIT_DEF $dst",
171 [(set FR64:$dst, (undef))]>, Requires<[HasSSE2]>;
173 // CMOV* - Used to implement the SSE SELECT DAG operation. Expanded by the
174 // scheduler into a branch sequence.
175 let usesCustomDAGSchedInserter = 1 in { // Expanded by the scheduler.
176 def CMOV_FR32 : I<0, Pseudo,
177 (ops FR32:$dst, FR32:$t, FR32:$f, i8imm:$cond),
178 "#CMOV_FR32 PSEUDO!",
179 [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond))]>;
180 def CMOV_FR64 : I<0, Pseudo,
181 (ops FR64:$dst, FR64:$t, FR64:$f, i8imm:$cond),
182 "#CMOV_FR64 PSEUDO!",
183 [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond))]>;
187 def MOVSSrr : SSI<0x10, MRMSrcReg, (ops FR32:$dst, FR32:$src),
188 "movss {$src, $dst|$dst, $src}", []>;
189 def MOVSSrm : SSI<0x10, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
190 "movss {$src, $dst|$dst, $src}",
191 [(set FR32:$dst, (loadf32 addr:$src))]>;
192 def MOVSDrr : SDI<0x10, MRMSrcReg, (ops FR64:$dst, FR64:$src),
193 "movsd {$src, $dst|$dst, $src}", []>;
194 def MOVSDrm : SDI<0x10, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
195 "movsd {$src, $dst|$dst, $src}",
196 [(set FR64:$dst, (loadf64 addr:$src))]>;
198 def MOVSSmr : SSI<0x11, MRMDestMem, (ops f32mem:$dst, FR32:$src),
199 "movss {$src, $dst|$dst, $src}",
200 [(store FR32:$src, addr:$dst)]>;
201 def MOVSDmr : SDI<0x11, MRMDestMem, (ops f64mem:$dst, FR64:$src),
202 "movsd {$src, $dst|$dst, $src}",
203 [(store FR64:$src, addr:$dst)]>;
205 // FR32 / FR64 to 128-bit vector conversion.
206 def MOVSS128rr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, FR32:$src),
207 "movss {$src, $dst|$dst, $src}",
209 (v4f32 (scalar_to_vector FR32:$src)))]>;
210 def MOVSS128rm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
211 "movss {$src, $dst|$dst, $src}",
213 (v4f32 (scalar_to_vector (loadf32 addr:$src))))]>;
214 def MOVSD128rr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, FR64:$src),
215 "movsd {$src, $dst|$dst, $src}",
217 (v2f64 (scalar_to_vector FR64:$src)))]>;
218 def MOVSD128rm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
219 "movsd {$src, $dst|$dst, $src}",
221 (v2f64 (scalar_to_vector (loadf64 addr:$src))))]>;
223 // Arithmetic instructions
224 let isTwoAddress = 1 in {
225 let isCommutable = 1 in {
226 def ADDSSrr : SSI<0x58, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
227 "addss {$src2, $dst|$dst, $src2}",
228 [(set FR32:$dst, (fadd FR32:$src1, FR32:$src2))]>;
229 def ADDSDrr : SDI<0x58, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
230 "addsd {$src2, $dst|$dst, $src2}",
231 [(set FR64:$dst, (fadd FR64:$src1, FR64:$src2))]>;
232 def MULSSrr : SSI<0x59, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
233 "mulss {$src2, $dst|$dst, $src2}",
234 [(set FR32:$dst, (fmul FR32:$src1, FR32:$src2))]>;
235 def MULSDrr : SDI<0x59, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
236 "mulsd {$src2, $dst|$dst, $src2}",
237 [(set FR64:$dst, (fmul FR64:$src1, FR64:$src2))]>;
240 def ADDSSrm : SSI<0x58, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
241 "addss {$src2, $dst|$dst, $src2}",
242 [(set FR32:$dst, (fadd FR32:$src1, (loadf32 addr:$src2)))]>;
243 def ADDSDrm : SDI<0x58, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
244 "addsd {$src2, $dst|$dst, $src2}",
245 [(set FR64:$dst, (fadd FR64:$src1, (loadf64 addr:$src2)))]>;
246 def MULSSrm : SSI<0x59, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
247 "mulss {$src2, $dst|$dst, $src2}",
248 [(set FR32:$dst, (fmul FR32:$src1, (loadf32 addr:$src2)))]>;
249 def MULSDrm : SDI<0x59, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
250 "mulsd {$src2, $dst|$dst, $src2}",
251 [(set FR64:$dst, (fmul FR64:$src1, (loadf64 addr:$src2)))]>;
253 def DIVSSrr : SSI<0x5E, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
254 "divss {$src2, $dst|$dst, $src2}",
255 [(set FR32:$dst, (fdiv FR32:$src1, FR32:$src2))]>;
256 def DIVSSrm : SSI<0x5E, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
257 "divss {$src2, $dst|$dst, $src2}",
258 [(set FR32:$dst, (fdiv FR32:$src1, (loadf32 addr:$src2)))]>;
259 def DIVSDrr : SDI<0x5E, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
260 "divsd {$src2, $dst|$dst, $src2}",
261 [(set FR64:$dst, (fdiv FR64:$src1, FR64:$src2))]>;
262 def DIVSDrm : SDI<0x5E, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
263 "divsd {$src2, $dst|$dst, $src2}",
264 [(set FR64:$dst, (fdiv FR64:$src1, (loadf64 addr:$src2)))]>;
266 def SUBSSrr : SSI<0x5C, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
267 "subss {$src2, $dst|$dst, $src2}",
268 [(set FR32:$dst, (fsub FR32:$src1, FR32:$src2))]>;
269 def SUBSSrm : SSI<0x5C, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f32mem:$src2),
270 "subss {$src2, $dst|$dst, $src2}",
271 [(set FR32:$dst, (fsub FR32:$src1, (loadf32 addr:$src2)))]>;
272 def SUBSDrr : SDI<0x5C, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
273 "subsd {$src2, $dst|$dst, $src2}",
274 [(set FR64:$dst, (fsub FR64:$src1, FR64:$src2))]>;
275 def SUBSDrm : SDI<0x5C, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f64mem:$src2),
276 "subsd {$src2, $dst|$dst, $src2}",
277 [(set FR64:$dst, (fsub FR64:$src1, (loadf64 addr:$src2)))]>;
280 def SQRTSSrr : SSI<0x51, MRMSrcReg, (ops FR32:$dst, FR32:$src),
281 "sqrtss {$src, $dst|$dst, $src}",
282 [(set FR32:$dst, (fsqrt FR32:$src))]>;
283 def SQRTSSrm : SSI<0x51, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
284 "sqrtss {$src, $dst|$dst, $src}",
285 [(set FR32:$dst, (fsqrt (loadf32 addr:$src)))]>;
286 def SQRTSDrr : SDI<0x51, MRMSrcReg, (ops FR64:$dst, FR64:$src),
287 "sqrtsd {$src, $dst|$dst, $src}",
288 [(set FR64:$dst, (fsqrt FR64:$src))]>;
289 def SQRTSDrm : SDI<0x51, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
290 "sqrtsd {$src, $dst|$dst, $src}",
291 [(set FR64:$dst, (fsqrt (loadf64 addr:$src)))]>;
293 def RSQRTSSrr : SSI<0x52, MRMSrcReg, (ops FR32:$dst, FR32:$src),
294 "rsqrtss {$src, $dst|$dst, $src}", []>;
295 def RSQRTSSrm : SSI<0x52, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
296 "rsqrtss {$src, $dst|$dst, $src}", []>;
297 def RCPSSrr : SSI<0x53, MRMSrcReg, (ops FR32:$dst, FR32:$src),
298 "rcpss {$src, $dst|$dst, $src}", []>;
299 def RCPSSrm : SSI<0x53, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
300 "rcpss {$src, $dst|$dst, $src}", []>;
302 def MAXSSrr : SSI<0x5F, MRMSrcReg, (ops FR32:$dst, FR32:$src),
303 "maxss {$src, $dst|$dst, $src}", []>;
304 def MAXSSrm : SSI<0x5F, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
305 "maxss {$src, $dst|$dst, $src}", []>;
306 def MAXSDrr : SDI<0x5F, MRMSrcReg, (ops FR64:$dst, FR64:$src),
307 "maxsd {$src, $dst|$dst, $src}", []>;
308 def MAXSDrm : SDI<0x5F, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
309 "maxsd {$src, $dst|$dst, $src}", []>;
310 def MINSSrr : SSI<0x5D, MRMSrcReg, (ops FR32:$dst, FR32:$src),
311 "minss {$src, $dst|$dst, $src}", []>;
312 def MINSSrm : SSI<0x5D, MRMSrcMem, (ops FR32:$dst, f32mem:$src),
313 "minss {$src, $dst|$dst, $src}", []>;
314 def MINSDrr : SDI<0x5D, MRMSrcReg, (ops FR64:$dst, FR64:$src),
315 "minsd {$src, $dst|$dst, $src}", []>;
316 def MINSDrm : SDI<0x5D, MRMSrcMem, (ops FR64:$dst, f64mem:$src),
317 "minsd {$src, $dst|$dst, $src}", []>;
320 // Aliases to match intrinsics which expect XMM operand(s).
321 let isTwoAddress = 1 in {
322 let isCommutable = 1 in {
323 def Int_ADDSSrr : SSI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
325 "addss {$src2, $dst|$dst, $src2}",
326 [(set VR128:$dst, (int_x86_sse_add_ss VR128:$src1,
328 def Int_ADDSDrr : SDI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
330 "addsd {$src2, $dst|$dst, $src2}",
331 [(set VR128:$dst, (int_x86_sse2_add_sd VR128:$src1,
333 def Int_MULSSrr : SSI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
335 "mulss {$src2, $dst|$dst, $src2}",
336 [(set VR128:$dst, (int_x86_sse_mul_ss VR128:$src1,
338 def Int_MULSDrr : SDI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
340 "mulsd {$src2, $dst|$dst, $src2}",
341 [(set VR128:$dst, (int_x86_sse2_mul_sd VR128:$src1,
345 def Int_ADDSSrm : SSI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
347 "addss {$src2, $dst|$dst, $src2}",
348 [(set VR128:$dst, (int_x86_sse_add_ss VR128:$src1,
349 (load addr:$src2)))]>;
350 def Int_ADDSDrm : SDI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
352 "addsd {$src2, $dst|$dst, $src2}",
353 [(set VR128:$dst, (int_x86_sse2_add_sd VR128:$src1,
354 (load addr:$src2)))]>;
355 def Int_MULSSrm : SSI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
357 "mulss {$src2, $dst|$dst, $src2}",
358 [(set VR128:$dst, (int_x86_sse_mul_ss VR128:$src1,
359 (load addr:$src2)))]>;
360 def Int_MULSDrm : SDI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
362 "mulsd {$src2, $dst|$dst, $src2}",
363 [(set VR128:$dst, (int_x86_sse2_mul_sd VR128:$src1,
364 (load addr:$src2)))]>;
366 def Int_DIVSSrr : SSI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
367 "divss {$src2, $dst|$dst, $src2}",
368 [(set VR128:$dst, (int_x86_sse_div_ss VR128:$src1,
370 def Int_DIVSSrm : SSI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
371 "divss {$src2, $dst|$dst, $src2}",
372 [(set VR128:$dst, (int_x86_sse_div_ss VR128:$src1,
373 (load addr:$src2)))]>;
374 def Int_DIVSDrr : SDI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
375 "divsd {$src2, $dst|$dst, $src2}",
376 [(set VR128:$dst, (int_x86_sse2_div_sd VR128:$src1,
378 def Int_DIVSDrm : SDI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
379 "divsd {$src2, $dst|$dst, $src2}",
380 [(set VR128:$dst, (int_x86_sse2_div_sd VR128:$src1,
381 (load addr:$src2)))]>;
383 def Int_SUBSSrr : SSI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
384 "subss {$src2, $dst|$dst, $src2}",
385 [(set VR128:$dst, (int_x86_sse_sub_ss VR128:$src1,
387 def Int_SUBSSrm : SSI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f32mem:$src2),
388 "subss {$src2, $dst|$dst, $src2}",
389 [(set VR128:$dst, (int_x86_sse_sub_ss VR128:$src1,
390 (load addr:$src2)))]>;
391 def Int_SUBSDrr : SDI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
392 "subsd {$src2, $dst|$dst, $src2}",
393 [(set VR128:$dst, (int_x86_sse2_sub_sd VR128:$src1,
395 def Int_SUBSDrm : SDI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
396 "subsd {$src2, $dst|$dst, $src2}",
397 [(set VR128:$dst, (int_x86_sse2_sub_sd VR128:$src1,
398 (load addr:$src2)))]>;
401 def Int_SQRTSSrr : SSI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
402 "sqrtss {$src, $dst|$dst, $src}",
403 [(set VR128:$dst, (int_x86_sse_sqrt_ss VR128:$src))]>;
404 def Int_SQRTSSrm : SSI<0x51, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
405 "sqrtss {$src, $dst|$dst, $src}",
406 [(set VR128:$dst, (int_x86_sse_sqrt_ss
407 (load addr:$src)))]>;
408 def Int_SQRTSDrr : SDI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
409 "sqrtsd {$src, $dst|$dst, $src}",
410 [(set VR128:$dst, (int_x86_sse2_sqrt_sd VR128:$src))]>;
411 def Int_SQRTSDrm : SDI<0x51, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
412 "sqrtsd {$src, $dst|$dst, $src}",
413 [(set VR128:$dst, (int_x86_sse2_sqrt_sd
414 (load addr:$src)))]>;
416 def Int_RSQRTSSrr : SSI<0x52, MRMSrcReg, (ops VR128:$dst, VR128:$src),
417 "rsqrtss {$src, $dst|$dst, $src}",
418 [(set VR128:$dst, (int_x86_sse_rsqrt_ss VR128:$src))]>;
419 def Int_RSQRTSSrm : SSI<0x52, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
420 "rsqrtss {$src, $dst|$dst, $src}",
421 [(set VR128:$dst, (int_x86_sse_rsqrt_ss
422 (load addr:$src)))]>;
423 def Int_RCPSSrr : SSI<0x53, MRMSrcReg, (ops VR128:$dst, VR128:$src),
424 "rcpss {$src, $dst|$dst, $src}",
425 [(set VR128:$dst, (int_x86_sse_rcp_ss VR128:$src))]>;
426 def Int_RCPSSrm : SSI<0x53, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
427 "rcpss {$src, $dst|$dst, $src}",
428 [(set VR128:$dst, (int_x86_sse_rcp_ss
429 (load addr:$src)))]>;
431 let isTwoAddress = 1 in {
432 def Int_MAXSSrr : SSI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
434 "maxss {$src2, $dst|$dst, $src2}",
435 [(set VR128:$dst, (int_x86_sse_max_ss VR128:$src1,
437 def Int_MAXSSrm : SSI<0x5F, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
439 "maxss {$src2, $dst|$dst, $src2}",
440 [(set VR128:$dst, (int_x86_sse_max_ss VR128:$src1,
441 (load addr:$src2)))]>;
442 def Int_MAXSDrr : SDI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
444 "maxsd {$src2, $dst|$dst, $src2}",
445 [(set VR128:$dst, (int_x86_sse2_max_sd VR128:$src1,
447 def Int_MAXSDrm : SDI<0x5F, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
449 "maxsd {$src2, $dst|$dst, $src2}",
450 [(set VR128:$dst, (int_x86_sse2_max_sd VR128:$src1,
451 (load addr:$src2)))]>;
452 def Int_MINSSrr : SSI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
454 "minss {$src2, $dst|$dst, $src2}",
455 [(set VR128:$dst, (int_x86_sse_min_ss VR128:$src1,
457 def Int_MINSSrm : SSI<0x5D, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
459 "minss {$src2, $dst|$dst, $src2}",
460 [(set VR128:$dst, (int_x86_sse_min_ss VR128:$src1,
461 (load addr:$src2)))]>;
462 def Int_MINSDrr : SDI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
464 "minsd {$src2, $dst|$dst, $src2}",
465 [(set VR128:$dst, (int_x86_sse2_min_sd VR128:$src1,
467 def Int_MINSDrm : SDI<0x5D, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
469 "minsd {$src2, $dst|$dst, $src2}",
470 [(set VR128:$dst, (int_x86_sse2_min_sd VR128:$src1,
471 (load addr:$src2)))]>;
474 // Conversion instructions
475 def CVTSS2SIrr: SSI<0x2D, MRMSrcReg, (ops R32:$dst, FR32:$src),
476 "cvtss2si {$src, $dst|$dst, $src}", []>;
477 def CVTSS2SIrm: SSI<0x2D, MRMSrcMem, (ops R32:$dst, f32mem:$src),
478 "cvtss2si {$src, $dst|$dst, $src}", []>;
480 def CVTTSS2SIrr: SSI<0x2C, MRMSrcReg, (ops R32:$dst, FR32:$src),
481 "cvttss2si {$src, $dst|$dst, $src}",
482 [(set R32:$dst, (fp_to_sint FR32:$src))]>;
483 def CVTTSS2SIrm: SSI<0x2C, MRMSrcMem, (ops R32:$dst, f32mem:$src),
484 "cvttss2si {$src, $dst|$dst, $src}",
485 [(set R32:$dst, (fp_to_sint (loadf32 addr:$src)))]>;
486 def CVTTSD2SIrr: SDI<0x2C, MRMSrcReg, (ops R32:$dst, FR64:$src),
487 "cvttsd2si {$src, $dst|$dst, $src}",
488 [(set R32:$dst, (fp_to_sint FR64:$src))]>;
489 def CVTTSD2SIrm: SDI<0x2C, MRMSrcMem, (ops R32:$dst, f64mem:$src),
490 "cvttsd2si {$src, $dst|$dst, $src}",
491 [(set R32:$dst, (fp_to_sint (loadf64 addr:$src)))]>;
492 def CVTSD2SSrr: SDI<0x5A, MRMSrcReg, (ops FR32:$dst, FR64:$src),
493 "cvtsd2ss {$src, $dst|$dst, $src}",
494 [(set FR32:$dst, (fround FR64:$src))]>;
495 def CVTSD2SSrm: SDI<0x5A, MRMSrcMem, (ops FR32:$dst, f64mem:$src),
496 "cvtsd2ss {$src, $dst|$dst, $src}",
497 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>;
498 def CVTSI2SSrr: SSI<0x2A, MRMSrcReg, (ops FR32:$dst, R32:$src),
499 "cvtsi2ss {$src, $dst|$dst, $src}",
500 [(set FR32:$dst, (sint_to_fp R32:$src))]>;
501 def CVTSI2SSrm: SSI<0x2A, MRMSrcMem, (ops FR32:$dst, i32mem:$src),
502 "cvtsi2ss {$src, $dst|$dst, $src}",
503 [(set FR32:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
504 def CVTSI2SDrr: SDI<0x2A, MRMSrcReg, (ops FR64:$dst, R32:$src),
505 "cvtsi2sd {$src, $dst|$dst, $src}",
506 [(set FR64:$dst, (sint_to_fp R32:$src))]>;
507 def CVTSI2SDrm: SDI<0x2A, MRMSrcMem, (ops FR64:$dst, i32mem:$src),
508 "cvtsi2sd {$src, $dst|$dst, $src}",
509 [(set FR64:$dst, (sint_to_fp (loadi32 addr:$src)))]>;
510 // SSE2 instructions with XS prefix
511 def CVTSS2SDrr: I<0x5A, MRMSrcReg, (ops FR64:$dst, FR32:$src),
512 "cvtss2sd {$src, $dst|$dst, $src}",
513 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
515 def CVTSS2SDrm: I<0x5A, MRMSrcMem, (ops FR64:$dst, f32mem:$src),
516 "cvtss2sd {$src, $dst|$dst, $src}",
517 [(set FR64:$dst, (fextend (loadf32 addr:$src)))]>, XS,
520 // Comparison instructions
521 let isTwoAddress = 1 in {
522 def CMPSSrr : SSI<0xC2, MRMSrcReg,
523 (ops FR32:$dst, FR32:$src1, FR32:$src, SSECC:$cc),
524 "cmp${cc}ss {$src, $dst|$dst, $src}",
526 def CMPSSrm : SSI<0xC2, MRMSrcMem,
527 (ops FR32:$dst, FR32:$src1, f32mem:$src, SSECC:$cc),
528 "cmp${cc}ss {$src, $dst|$dst, $src}", []>;
529 def CMPSDrr : SDI<0xC2, MRMSrcReg,
530 (ops FR64:$dst, FR64:$src1, FR64:$src, SSECC:$cc),
531 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
532 def CMPSDrm : SDI<0xC2, MRMSrcMem,
533 (ops FR64:$dst, FR64:$src1, f64mem:$src, SSECC:$cc),
534 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
537 def UCOMISSrr: PSI<0x2E, MRMSrcReg, (ops FR32:$src1, FR32:$src2),
538 "ucomiss {$src2, $src1|$src1, $src2}",
539 [(X86cmp FR32:$src1, FR32:$src2)]>;
540 def UCOMISSrm: PSI<0x2E, MRMSrcMem, (ops FR32:$src1, f32mem:$src2),
541 "ucomiss {$src2, $src1|$src1, $src2}",
542 [(X86cmp FR32:$src1, (loadf32 addr:$src2))]>;
543 def UCOMISDrr: PDI<0x2E, MRMSrcReg, (ops FR64:$src1, FR64:$src2),
544 "ucomisd {$src2, $src1|$src1, $src2}",
545 [(X86cmp FR64:$src1, FR64:$src2)]>;
546 def UCOMISDrm: PDI<0x2E, MRMSrcMem, (ops FR64:$src1, f64mem:$src2),
547 "ucomisd {$src2, $src1|$src1, $src2}",
548 [(X86cmp FR64:$src1, (loadf64 addr:$src2))]>;
550 // Aliases to match intrinsics which expect XMM operand(s).
551 let isTwoAddress = 1 in {
552 def Int_CMPSSrr : SSI<0xC2, MRMSrcReg,
553 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
554 "cmp${cc}ss {$src, $dst|$dst, $src}",
555 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
556 VR128:$src, imm:$cc))]>;
557 def Int_CMPSSrm : SSI<0xC2, MRMSrcMem,
558 (ops VR128:$dst, VR128:$src1, f32mem:$src, SSECC:$cc),
559 "cmp${cc}ss {$src, $dst|$dst, $src}",
560 [(set VR128:$dst, (int_x86_sse_cmp_ss VR128:$src1,
561 (load addr:$src), imm:$cc))]>;
562 def Int_CMPSDrr : SDI<0xC2, MRMSrcReg,
563 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
564 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
565 def Int_CMPSDrm : SDI<0xC2, MRMSrcMem,
566 (ops VR128:$dst, VR128:$src1, f64mem:$src, SSECC:$cc),
567 "cmp${cc}sd {$src, $dst|$dst, $src}", []>;
571 // Aliases of packed instructions for scalar use. These all have names that
574 // Alias instructions that map fld0 to pxor for sse.
575 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
576 def FsFLD0SS : I<0xEF, MRMInitReg, (ops FR32:$dst),
577 "pxor $dst, $dst", [(set FR32:$dst, fp32imm0)]>,
578 Requires<[HasSSE1]>, TB, OpSize;
579 def FsFLD0SD : I<0xEF, MRMInitReg, (ops FR64:$dst),
580 "pxor $dst, $dst", [(set FR64:$dst, fp64imm0)]>,
581 Requires<[HasSSE2]>, TB, OpSize;
583 // Alias instructions to do FR32 / FR64 reg-to-reg copy using movaps / movapd.
584 // Upper bits are disregarded.
585 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (ops FR32:$dst, FR32:$src),
586 "movaps {$src, $dst|$dst, $src}", []>;
587 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (ops FR64:$dst, FR64:$src),
588 "movapd {$src, $dst|$dst, $src}", []>;
590 // Alias instructions to load FR32 / FR64 from f128mem using movaps / movapd.
591 // Upper bits are disregarded.
592 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (ops FR32:$dst, f128mem:$src),
593 "movaps {$src, $dst|$dst, $src}",
594 [(set FR32:$dst, (X86loadpf32 addr:$src))]>;
595 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (ops FR64:$dst, f128mem:$src),
596 "movapd {$src, $dst|$dst, $src}",
597 [(set FR64:$dst, (X86loadpf64 addr:$src))]>;
599 // Alias bitwise logical operations using SSE logical ops on packed FP values.
600 let isTwoAddress = 1 in {
601 let isCommutable = 1 in {
602 def FsANDPSrr : PSI<0x54, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
603 "andps {$src2, $dst|$dst, $src2}",
604 [(set FR32:$dst, (X86fand FR32:$src1, FR32:$src2))]>;
605 def FsANDPDrr : PDI<0x54, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
606 "andpd {$src2, $dst|$dst, $src2}",
607 [(set FR64:$dst, (X86fand FR64:$src1, FR64:$src2))]>;
608 def FsORPSrr : PSI<0x56, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
609 "orps {$src2, $dst|$dst, $src2}", []>;
610 def FsORPDrr : PDI<0x56, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
611 "orpd {$src2, $dst|$dst, $src2}", []>;
612 def FsXORPSrr : PSI<0x57, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
613 "xorps {$src2, $dst|$dst, $src2}",
614 [(set FR32:$dst, (X86fxor FR32:$src1, FR32:$src2))]>;
615 def FsXORPDrr : PDI<0x57, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
616 "xorpd {$src2, $dst|$dst, $src2}",
617 [(set FR64:$dst, (X86fxor FR64:$src1, FR64:$src2))]>;
619 def FsANDPSrm : PSI<0x54, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
620 "andps {$src2, $dst|$dst, $src2}",
621 [(set FR32:$dst, (X86fand FR32:$src1,
622 (X86loadpf32 addr:$src2)))]>;
623 def FsANDPDrm : PDI<0x54, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
624 "andpd {$src2, $dst|$dst, $src2}",
625 [(set FR64:$dst, (X86fand FR64:$src1,
626 (X86loadpf64 addr:$src2)))]>;
627 def FsORPSrm : PSI<0x56, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
628 "orps {$src2, $dst|$dst, $src2}", []>;
629 def FsORPDrm : PDI<0x56, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
630 "orpd {$src2, $dst|$dst, $src2}", []>;
631 def FsXORPSrm : PSI<0x57, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
632 "xorps {$src2, $dst|$dst, $src2}",
633 [(set FR32:$dst, (X86fxor FR32:$src1,
634 (X86loadpf32 addr:$src2)))]>;
635 def FsXORPDrm : PDI<0x57, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
636 "xorpd {$src2, $dst|$dst, $src2}",
637 [(set FR64:$dst, (X86fxor FR64:$src1,
638 (X86loadpf64 addr:$src2)))]>;
640 def FsANDNPSrr : PSI<0x55, MRMSrcReg, (ops FR32:$dst, FR32:$src1, FR32:$src2),
641 "andnps {$src2, $dst|$dst, $src2}", []>;
642 def FsANDNPSrm : PSI<0x55, MRMSrcMem, (ops FR32:$dst, FR32:$src1, f128mem:$src2),
643 "andnps {$src2, $dst|$dst, $src2}", []>;
644 def FsANDNPDrr : PDI<0x55, MRMSrcReg, (ops FR64:$dst, FR64:$src1, FR64:$src2),
645 "andnpd {$src2, $dst|$dst, $src2}", []>;
646 def FsANDNPDrm : PDI<0x55, MRMSrcMem, (ops FR64:$dst, FR64:$src1, f128mem:$src2),
647 "andnpd {$src2, $dst|$dst, $src2}", []>;
650 //===----------------------------------------------------------------------===//
651 // SSE packed FP Instructions
652 //===----------------------------------------------------------------------===//
654 // Some 'special' instructions
655 def IMPLICIT_DEF_VR128 : I<0, Pseudo, (ops VR128:$dst),
656 "#IMPLICIT_DEF $dst",
657 [(set VR128:$dst, (v4f32 (undef)))]>,
661 def MOVAPSrr : PSI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
662 "movaps {$src, $dst|$dst, $src}", []>;
663 def MOVAPSrm : PSI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
664 "movaps {$src, $dst|$dst, $src}",
665 [(set VR128:$dst, (loadv4f32 addr:$src))]>;
666 def MOVAPDrr : PDI<0x28, MRMSrcReg, (ops VR128:$dst, VR128:$src),
667 "movapd {$src, $dst|$dst, $src}", []>;
668 def MOVAPDrm : PDI<0x28, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
669 "movapd {$src, $dst|$dst, $src}",
670 [(set VR128:$dst, (loadv2f64 addr:$src))]>;
672 def MOVAPSmr : PSI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
673 "movaps {$src, $dst|$dst, $src}",
674 [(store (v4f32 VR128:$src), addr:$dst)]>;
675 def MOVAPDmr : PDI<0x29, MRMDestMem, (ops f128mem:$dst, VR128:$src),
676 "movapd {$src, $dst|$dst, $src}",
677 [(store (v2f64 VR128:$src), addr:$dst)]>;
679 def MOVUPSrr : PSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
680 "movups {$src, $dst|$dst, $src}", []>;
681 def MOVUPSrm : PSI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
682 "movups {$src, $dst|$dst, $src}", []>;
683 def MOVUPSmr : PSI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
684 "movups {$src, $dst|$dst, $src}", []>;
685 def MOVUPDrr : PDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src),
686 "movupd {$src, $dst|$dst, $src}", []>;
687 def MOVUPDrm : PDI<0x10, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
688 "movupd {$src, $dst|$dst, $src}", []>;
689 def MOVUPDmr : PDI<0x11, MRMDestMem, (ops f128mem:$dst, VR128:$src),
690 "movupd {$src, $dst|$dst, $src}", []>;
692 let isTwoAddress = 1 in {
693 def MOVLPSrm : PSI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
694 "movlps {$src2, $dst|$dst, $src2}", []>;
695 def MOVLPDrm : PDI<0x12, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
696 "movlpd {$src2, $dst|$dst, $src2}", []>;
697 def MOVHPSrm : PSI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
698 "movhps {$src2, $dst|$dst, $src2}", []>;
699 def MOVHPDrm : PDI<0x16, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f64mem:$src2),
700 "movhpd {$src2, $dst|$dst, $src2}",
702 (v2f64 (vector_shuffle VR128:$src1,
703 (scalar_to_vector (loadf64 addr:$src2)),
704 UNPCKL_shuffle_mask)))]>;
707 def MOVLPSmr : PSI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
708 "movlps {$src, $dst|$dst, $src}", []>;
709 def MOVLPDmr : PDI<0x13, MRMDestMem, (ops f64mem:$dst, VR128:$src),
710 "movlpd {$src, $dst|$dst, $src}", []>;
712 def MOVHPSmr : PSI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
713 "movhps {$src, $dst|$dst, $src}", []>;
714 def MOVHPDmr : PDI<0x17, MRMDestMem, (ops f64mem:$dst, VR128:$src),
715 "movhpd {$src, $dst|$dst, $src}", []>;
717 let isTwoAddress = 1 in {
718 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
719 "movlhps {$src2, $dst|$dst, $src2}",
721 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
722 MOVLHPS_shuffle_mask)))]>;
724 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
725 "movhlps {$src2, $dst|$dst, $src2}",
727 (v4f32 (vector_shuffle VR128:$src1, VR128:$src2,
728 MOVHLPS_shuffle_mask)))]>;
731 // Conversion instructions
732 def CVTPI2PSrr : PSI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src),
733 "cvtpi2ps {$src, $dst|$dst, $src}", []>;
734 def CVTPI2PSrm : PSI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
735 "cvtpi2ps {$src, $dst|$dst, $src}", []>;
736 def CVTPI2PDrr : PDI<0x2A, MRMSrcReg, (ops VR128:$dst, VR64:$src),
737 "cvtpi2pd {$src, $dst|$dst, $src}", []>;
738 def CVTPI2PDrm : PDI<0x2A, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
739 "cvtpi2pd {$src, $dst|$dst, $src}", []>;
741 // SSE2 instructions without OpSize prefix
742 def CVTDQ2PSrr : I<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
743 "cvtdq2ps {$src, $dst|$dst, $src}", []>, TB,
745 def CVTDQ2PSrm : I<0x5B, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
746 "cvtdq2ps {$src, $dst|$dst, $src}", []>, TB,
749 // SSE2 instructions with XS prefix
750 def CVTDQ2PDrr : I<0xE6, MRMSrcReg, (ops VR128:$dst, VR64:$src),
751 "cvtdq2pd {$src, $dst|$dst, $src}", []>,
752 XS, Requires<[HasSSE2]>;
753 def CVTDQ2PDrm : I<0xE6, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
754 "cvtdq2pd {$src, $dst|$dst, $src}", []>,
755 XS, Requires<[HasSSE2]>;
757 def CVTPS2PIrr : PSI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src),
758 "cvtps2pi {$src, $dst|$dst, $src}", []>;
759 def CVTPS2PIrm : PSI<0x2D, MRMSrcMem, (ops VR64:$dst, f64mem:$src),
760 "cvtps2pi {$src, $dst|$dst, $src}", []>;
761 def CVTPD2PIrr : PDI<0x2D, MRMSrcReg, (ops VR64:$dst, VR128:$src),
762 "cvtpd2pi {$src, $dst|$dst, $src}", []>;
763 def CVTPD2PIrm : PDI<0x2D, MRMSrcMem, (ops VR64:$dst, f128mem:$src),
764 "cvtpd2pi {$src, $dst|$dst, $src}", []>;
766 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (ops VR128:$dst, VR128:$src),
767 "cvtps2dq {$src, $dst|$dst, $src}", []>;
768 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
769 "cvtps2dq {$src, $dst|$dst, $src}", []>;
770 // SSE2 packed instructions with XD prefix
771 def CVTPD2DQrr : SDI<0xE6, MRMSrcReg, (ops VR128:$dst, VR128:$src),
772 "cvtpd2dq {$src, $dst|$dst, $src}", []>;
773 def CVTPD2DQrm : SDI<0xE6, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
774 "cvtpd2dq {$src, $dst|$dst, $src}", []>;
776 // SSE2 instructions without OpSize prefix
777 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
778 "cvtps2pd {$src, $dst|$dst, $src}", []>, TB,
780 def CVTPS2PDrm : I<0x5A, MRMSrcReg, (ops VR128:$dst, f64mem:$src),
781 "cvtps2pd {$src, $dst|$dst, $src}", []>, TB,
784 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, VR128:$src),
785 "cvtpd2ps {$src, $dst|$dst, $src}", []>;
786 def CVTPD2PSrm : PDI<0x5A, MRMSrcReg, (ops VR128:$dst, f128mem:$src),
787 "cvtpd2ps {$src, $dst|$dst, $src}", []>;
790 let isTwoAddress = 1 in {
791 let isCommutable = 1 in {
792 def ADDPSrr : PSI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
793 "addps {$src2, $dst|$dst, $src2}",
794 [(set VR128:$dst, (v4f32 (fadd VR128:$src1, VR128:$src2)))]>;
795 def ADDPDrr : PDI<0x58, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
796 "addpd {$src2, $dst|$dst, $src2}",
797 [(set VR128:$dst, (v2f64 (fadd VR128:$src1, VR128:$src2)))]>;
798 def MULPSrr : PSI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
799 "mulps {$src2, $dst|$dst, $src2}",
800 [(set VR128:$dst, (v4f32 (fmul VR128:$src1, VR128:$src2)))]>;
801 def MULPDrr : PDI<0x59, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
802 "mulpd {$src2, $dst|$dst, $src2}",
803 [(set VR128:$dst, (v2f64 (fmul VR128:$src1, VR128:$src2)))]>;
806 def ADDPSrm : PSI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
807 "addps {$src2, $dst|$dst, $src2}",
808 [(set VR128:$dst, (v4f32 (fadd VR128:$src1,
809 (load addr:$src2))))]>;
810 def ADDPDrm : PDI<0x58, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
811 "addpd {$src2, $dst|$dst, $src2}",
812 [(set VR128:$dst, (v2f64 (fadd VR128:$src1,
813 (load addr:$src2))))]>;
814 def MULPSrm : PSI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
815 "mulps {$src2, $dst|$dst, $src2}",
816 [(set VR128:$dst, (v4f32 (fmul VR128:$src1,
817 (load addr:$src2))))]>;
818 def MULPDrm : PDI<0x59, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
819 "mulpd {$src2, $dst|$dst, $src2}",
820 [(set VR128:$dst, (v2f64 (fmul VR128:$src1,
821 (load addr:$src2))))]>;
823 def DIVPSrr : PSI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
824 "divps {$src2, $dst|$dst, $src2}",
825 [(set VR128:$dst, (v4f32 (fdiv VR128:$src1, VR128:$src2)))]>;
826 def DIVPSrm : PSI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
827 "divps {$src2, $dst|$dst, $src2}",
828 [(set VR128:$dst, (v4f32 (fdiv VR128:$src1,
829 (load addr:$src2))))]>;
830 def DIVPDrr : PDI<0x5E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
831 "divpd {$src2, $dst|$dst, $src2}",
832 [(set VR128:$dst, (v2f64 (fdiv VR128:$src1, VR128:$src2)))]>;
833 def DIVPDrm : PDI<0x5E, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
834 "divpd {$src2, $dst|$dst, $src2}",
835 [(set VR128:$dst, (v2f64 (fdiv VR128:$src1,
836 (load addr:$src2))))]>;
838 def SUBPSrr : PSI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
839 "subps {$src2, $dst|$dst, $src2}",
840 [(set VR128:$dst, (v4f32 (fsub VR128:$src1, VR128:$src2)))]>;
841 def SUBPSrm : PSI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
842 "subps {$src2, $dst|$dst, $src2}",
843 [(set VR128:$dst, (v4f32 (fsub VR128:$src1,
844 (load addr:$src2))))]>;
845 def SUBPDrr : PDI<0x5C, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
846 "subpd {$src2, $dst|$dst, $src2}",
847 [(set VR128:$dst, (v2f64 (fsub VR128:$src1, VR128:$src2)))]>;
848 def SUBPDrm : PDI<0x5C, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
849 "subpd {$src2, $dst|$dst, $src2}",
850 [(set VR128:$dst, (v2f64 (fsub VR128:$src1,
851 (load addr:$src2))))]>;
854 def SQRTPSrr : PSI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
855 "sqrtps {$src, $dst|$dst, $src}",
856 [(set VR128:$dst, (v4f32 (fsqrt VR128:$src)))]>;
857 def SQRTPSrm : PSI<0x51, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
858 "sqrtps {$src, $dst|$dst, $src}",
859 [(set VR128:$dst, (v4f32 (fsqrt (load addr:$src))))]>;
860 def SQRTPDrr : PDI<0x51, MRMSrcReg, (ops VR128:$dst, VR128:$src),
861 "sqrtpd {$src, $dst|$dst, $src}",
862 [(set VR128:$dst, (v2f64 (fsqrt VR128:$src)))]>;
863 def SQRTPDrm : PDI<0x51, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
864 "sqrtpd {$src, $dst|$dst, $src}",
865 [(set VR128:$dst, (v2f64 (fsqrt (load addr:$src))))]>;
867 def RSQRTPSrr : PSI<0x52, MRMSrcReg, (ops VR128:$dst, VR128:$src),
868 "rsqrtps {$src, $dst|$dst, $src}", []>;
869 def RSQRTPSrm : PSI<0x52, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
870 "rsqrtps {$src, $dst|$dst, $src}", []>;
871 def RCPPSrr : PSI<0x53, MRMSrcReg, (ops VR128:$dst, VR128:$src),
872 "rcpps {$src, $dst|$dst, $src}", []>;
873 def RCPPSrm : PSI<0x53, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
874 "rcpps {$src, $dst|$dst, $src}", []>;
876 def MAXPSrr : PSI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
877 "maxps {$src, $dst|$dst, $src}", []>;
878 def MAXPSrm : PSI<0x5F, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
879 "maxps {$src, $dst|$dst, $src}", []>;
880 def MAXPDrr : PDI<0x5F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
881 "maxpd {$src, $dst|$dst, $src}", []>;
882 def MAXPDrm : PDI<0x5F, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
883 "maxpd {$src, $dst|$dst, $src}", []>;
884 def MINPSrr : PSI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src),
885 "minps {$src, $dst|$dst, $src}", []>;
886 def MINPSrm : PSI<0x5D, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
887 "minps {$src, $dst|$dst, $src}", []>;
888 def MINPDrr : PDI<0x5D, MRMSrcReg, (ops VR128:$dst, VR128:$src),
889 "minpd {$src, $dst|$dst, $src}", []>;
890 def MINPDrm : PDI<0x5D, MRMSrcMem, (ops VR128:$dst, f128mem:$src),
891 "minpd {$src, $dst|$dst, $src}", []>;
894 let isTwoAddress = 1 in {
895 let isCommutable = 1 in {
896 def ANDPSrr : PSI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
897 "andps {$src2, $dst|$dst, $src2}",
899 (and (bc_v4i32 (v4f32 VR128:$src1)),
900 (bc_v4i32 (v4f32 VR128:$src2))))]>;
901 def ANDPDrr : PDI<0x54, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
902 "andpd {$src2, $dst|$dst, $src2}",
904 (and (bc_v2i64 (v2f64 VR128:$src1)),
905 (bc_v2i64 (v2f64 VR128:$src2))))]>;
906 def ORPSrr : PSI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
907 "orps {$src2, $dst|$dst, $src2}",
909 (or (bc_v4i32 (v4f32 VR128:$src1)),
910 (bc_v4i32 (v4f32 VR128:$src2))))]>;
911 def ORPDrr : PDI<0x56, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
912 "orpd {$src2, $dst|$dst, $src2}",
914 (or (bc_v2i64 (v2f64 VR128:$src1)),
915 (bc_v2i64 (v2f64 VR128:$src2))))]>;
916 def XORPSrr : PSI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
917 "xorps {$src2, $dst|$dst, $src2}",
919 (xor (bc_v4i32 (v4f32 VR128:$src1)),
920 (bc_v4i32 (v4f32 VR128:$src2))))]>;
921 def XORPDrr : PDI<0x57, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
922 "xorpd {$src2, $dst|$dst, $src2}",
924 (xor (bc_v2i64 (v2f64 VR128:$src1)),
925 (bc_v2i64 (v2f64 VR128:$src2))))]>;
927 def ANDPSrm : PSI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
928 "andps {$src2, $dst|$dst, $src2}",
930 (and (bc_v4i32 (v4f32 VR128:$src1)),
931 (bc_v4i32 (loadv4f32 addr:$src2))))]>;
932 def ANDPDrm : PDI<0x54, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
933 "andpd {$src2, $dst|$dst, $src2}",
935 (and (bc_v2i64 (v2f64 VR128:$src1)),
936 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
937 def ORPSrm : PSI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
938 "orps {$src2, $dst|$dst, $src2}",
940 (or (bc_v4i32 (v4f32 VR128:$src1)),
941 (bc_v4i32 (loadv4f32 addr:$src2))))]>;
942 def ORPDrm : PDI<0x56, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
943 "orpd {$src2, $dst|$dst, $src2}",
945 (or (bc_v2i64 (v2f64 VR128:$src1)),
946 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
947 def XORPSrm : PSI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
948 "xorps {$src2, $dst|$dst, $src2}",
950 (xor (bc_v4i32 (v4f32 VR128:$src1)),
951 (bc_v4i32 (loadv4f32 addr:$src2))))]>;
952 def XORPDrm : PDI<0x57, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
953 "xorpd {$src2, $dst|$dst, $src2}",
955 (xor (bc_v2i64 (v2f64 VR128:$src1)),
956 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
957 def ANDNPSrr : PSI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
958 "andnps {$src2, $dst|$dst, $src2}",
960 (and (vnot (bc_v4i32 (v4f32 VR128:$src1))),
961 (bc_v4i32 (v4f32 VR128:$src2))))]>;
962 def ANDNPSrm : PSI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
963 "andnps {$src2, $dst|$dst, $src2}",
965 (and (vnot (bc_v4i32 (v4f32 VR128:$src1))),
966 (bc_v4i32 (loadv4f32 addr:$src2))))]>;
967 def ANDNPDrr : PDI<0x55, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
968 "andnpd {$src2, $dst|$dst, $src2}",
970 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
971 (bc_v2i64 (v2f64 VR128:$src2))))]>;
972 def ANDNPDrm : PDI<0x55, MRMSrcMem, (ops VR128:$dst, VR128:$src1,f128mem:$src2),
973 "andnpd {$src2, $dst|$dst, $src2}",
975 (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
976 (bc_v2i64 (loadv2f64 addr:$src2))))]>;
979 let isTwoAddress = 1 in {
980 def CMPPSrr : PSI<0xC2, MRMSrcReg,
981 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
982 "cmp${cc}ps {$src, $dst|$dst, $src}",
983 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
984 VR128:$src, imm:$cc))]>;
985 def CMPPSrm : PSI<0xC2, MRMSrcMem,
986 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
987 "cmp${cc}ps {$src, $dst|$dst, $src}",
988 [(set VR128:$dst, (int_x86_sse_cmp_ps VR128:$src1,
989 (load addr:$src), imm:$cc))]>;
990 def CMPPDrr : PDI<0xC2, MRMSrcReg,
991 (ops VR128:$dst, VR128:$src1, VR128:$src, SSECC:$cc),
992 "cmp${cc}pd {$src, $dst|$dst, $src}", []>;
993 def CMPPDrm : PDI<0xC2, MRMSrcMem,
994 (ops VR128:$dst, VR128:$src1, f128mem:$src, SSECC:$cc),
995 "cmp${cc}pd {$src, $dst|$dst, $src}", []>;
998 // Shuffle and unpack instructions
999 let isTwoAddress = 1 in {
1000 def SHUFPSrr : PSIi8<0xC6, MRMSrcReg,
1001 (ops VR128:$dst, VR128:$src1, VR128:$src2, i32i8imm:$src3),
1002 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1003 [(set VR128:$dst, (v4f32 (vector_shuffle
1004 VR128:$src1, VR128:$src2,
1005 SHUFP_shuffle_mask:$src3)))]>;
1006 def SHUFPSrm : PSIi8<0xC6, MRMSrcMem,
1007 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i32i8imm:$src3),
1008 "shufps {$src3, $src2, $dst|$dst, $src2, $src3}",
1009 [(set VR128:$dst, (v4f32 (vector_shuffle
1010 VR128:$src1, (load addr:$src2),
1011 SHUFP_shuffle_mask:$src3)))]>;
1012 def SHUFPDrr : PDIi8<0xC6, MRMSrcReg,
1013 (ops VR128:$dst, VR128:$src1, VR128:$src2, i8imm:$src3),
1014 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1015 [(set VR128:$dst, (v2f64 (vector_shuffle
1016 VR128:$src1, VR128:$src2,
1017 SHUFP_shuffle_mask:$src3)))]>;
1018 def SHUFPDrm : PDIi8<0xC6, MRMSrcMem,
1019 (ops VR128:$dst, VR128:$src1, f128mem:$src2, i8imm:$src3),
1020 "shufpd {$src3, $src2, $dst|$dst, $src2, $src3}",
1021 [(set VR128:$dst, (v2f64 (vector_shuffle
1022 VR128:$src1, (load addr:$src2),
1023 SHUFP_shuffle_mask:$src3)))]>;
1025 def UNPCKHPSrr : PSI<0x15, MRMSrcReg,
1026 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1027 "unpckhps {$src2, $dst|$dst, $src2}",
1028 [(set VR128:$dst, (v4f32 (vector_shuffle
1029 VR128:$src1, VR128:$src2,
1030 UNPCKH_shuffle_mask)))]>;
1031 def UNPCKHPSrm : PSI<0x15, MRMSrcMem,
1032 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1033 "unpckhps {$src2, $dst|$dst, $src2}",
1034 [(set VR128:$dst, (v4f32 (vector_shuffle
1035 VR128:$src1, (load addr:$src2),
1036 UNPCKH_shuffle_mask)))]>;
1037 def UNPCKHPDrr : PDI<0x15, MRMSrcReg,
1038 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1039 "unpckhpd {$src2, $dst|$dst, $src2}",
1040 [(set VR128:$dst, (v2f64 (vector_shuffle
1041 VR128:$src1, VR128:$src2,
1042 UNPCKH_shuffle_mask)))]>;
1043 def UNPCKHPDrm : PDI<0x15, MRMSrcMem,
1044 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1045 "unpckhpd {$src2, $dst|$dst, $src2}",
1046 [(set VR128:$dst, (v2f64 (vector_shuffle
1047 VR128:$src1, (load addr:$src2),
1048 UNPCKH_shuffle_mask)))]>;
1050 def UNPCKLPSrr : PSI<0x14, MRMSrcReg,
1051 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1052 "unpcklps {$src2, $dst|$dst, $src2}",
1053 [(set VR128:$dst, (v4f32 (vector_shuffle
1054 VR128:$src1, VR128:$src2,
1055 UNPCKL_shuffle_mask)))]>;
1056 def UNPCKLPSrm : PSI<0x14, MRMSrcMem,
1057 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1058 "unpcklps {$src2, $dst|$dst, $src2}",
1059 [(set VR128:$dst, (v4f32 (vector_shuffle
1060 VR128:$src1, (load addr:$src2),
1061 UNPCKL_shuffle_mask)))]>;
1062 def UNPCKLPDrr : PDI<0x14, MRMSrcReg,
1063 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1064 "unpcklpd {$src2, $dst|$dst, $src2}",
1065 [(set VR128:$dst, (v2f64 (vector_shuffle
1066 VR128:$src1, VR128:$src2,
1067 UNPCKL_shuffle_mask)))]>;
1068 def UNPCKLPDrm : PDI<0x14, MRMSrcMem,
1069 (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1070 "unpcklpd {$src2, $dst|$dst, $src2}",
1071 [(set VR128:$dst, (v2f64 (vector_shuffle
1072 VR128:$src1, (load addr:$src2),
1073 UNPCKL_shuffle_mask)))]>;
1076 //===----------------------------------------------------------------------===//
1077 // SSE integer instructions
1078 //===----------------------------------------------------------------------===//
1080 // Move Instructions
1081 def MOVD128rr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, R32:$src),
1082 "movd {$src, $dst|$dst, $src}",
1084 (v4i32 (scalar_to_vector R32:$src)))]>;
1085 def MOVD128rm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1086 "movd {$src, $dst|$dst, $src}",
1088 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
1090 def MOVD128mr : PDI<0x7E, MRMDestMem, (ops i32mem:$dst, VR128:$src),
1091 "movd {$src, $dst|$dst, $src}", []>;
1093 def MOVDQArr : PDI<0x6F, MRMSrcReg, (ops VR128:$dst, VR128:$src),
1094 "movdqa {$src, $dst|$dst, $src}", []>;
1095 def MOVDQArm : PDI<0x6F, MRMSrcMem, (ops VR128:$dst, i128mem:$src),
1096 "movdqa {$src, $dst|$dst, $src}",
1097 [(set VR128:$dst, (loadv4i32 addr:$src))]>;
1098 def MOVDQAmr : PDI<0x7F, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1099 "movdqa {$src, $dst|$dst, $src}",
1100 [(store (v4i32 VR128:$src), addr:$dst)]>;
1102 // SSE2 instructions with XS prefix
1103 def MOVQ128rr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR64:$src),
1104 "movq {$src, $dst|$dst, $src}",
1106 (v2i64 (scalar_to_vector VR64:$src)))]>, XS,
1107 Requires<[HasSSE2]>;
1108 def MOVQ128rm : I<0x7E, MRMSrcMem, (ops VR128:$dst, i64mem:$src),
1109 "movq {$src, $dst|$dst, $src}", []>, XS,
1110 Requires<[HasSSE2]>;
1111 def MOVQ128mr : PDI<0xD6, MRMSrcMem, (ops i64mem:$dst, VR128:$src),
1112 "movq {$src, $dst|$dst, $src}", []>;
1114 // 128-bit Integer Arithmetic
1115 let isTwoAddress = 1 in {
1116 let isCommutable = 1 in {
1117 def PADDBrr : PDI<0xFC, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1118 "paddb {$src2, $dst|$dst, $src2}",
1119 [(set VR128:$dst, (v16i8 (add VR128:$src1, VR128:$src2)))]>;
1120 def PADDWrr : PDI<0xFD, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1121 "paddw {$src2, $dst|$dst, $src2}",
1122 [(set VR128:$dst, (v8i16 (add VR128:$src1, VR128:$src2)))]>;
1123 def PADDDrr : PDI<0xFE, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1124 "paddd {$src2, $dst|$dst, $src2}",
1125 [(set VR128:$dst, (v4i32 (add VR128:$src1, VR128:$src2)))]>;
1127 def PADDQrr : PDI<0xD4, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1128 "paddq {$src2, $dst|$dst, $src2}",
1129 [(set VR128:$dst, (v2i64 (add VR128:$src1, VR128:$src2)))]>;
1131 def PADDBrm : PDI<0xFC, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1132 "paddb {$src2, $dst|$dst, $src2}",
1133 [(set VR128:$dst, (v16i8 (add VR128:$src1,
1134 (load addr:$src2))))]>;
1135 def PADDWrm : PDI<0xFD, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1136 "paddw {$src2, $dst|$dst, $src2}",
1137 [(set VR128:$dst, (v8i16 (add VR128:$src1,
1138 (load addr:$src2))))]>;
1139 def PADDDrm : PDI<0xFE, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1140 "paddd {$src2, $dst|$dst, $src2}",
1141 [(set VR128:$dst, (v4i32 (add VR128:$src1,
1142 (load addr:$src2))))]>;
1143 def PADDQrm : PDI<0xD4, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1144 "paddd {$src2, $dst|$dst, $src2}",
1145 [(set VR128:$dst, (v2i64 (add VR128:$src1,
1146 (load addr:$src2))))]>;
1148 def PSUBBrr : PDI<0xF8, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1149 "psubb {$src2, $dst|$dst, $src2}",
1150 [(set VR128:$dst, (v16i8 (sub VR128:$src1, VR128:$src2)))]>;
1151 def PSUBWrr : PDI<0xF9, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1152 "psubw {$src2, $dst|$dst, $src2}",
1153 [(set VR128:$dst, (v8i16 (sub VR128:$src1, VR128:$src2)))]>;
1154 def PSUBDrr : PDI<0xFA, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1155 "psubd {$src2, $dst|$dst, $src2}",
1156 [(set VR128:$dst, (v4i32 (sub VR128:$src1, VR128:$src2)))]>;
1157 def PSUBQrr : PDI<0xFB, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1158 "psubq {$src2, $dst|$dst, $src2}",
1159 [(set VR128:$dst, (v2i64 (sub VR128:$src1, VR128:$src2)))]>;
1161 def PSUBBrm : PDI<0xF8, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1162 "psubb {$src2, $dst|$dst, $src2}",
1163 [(set VR128:$dst, (v16i8 (sub VR128:$src1,
1164 (load addr:$src2))))]>;
1165 def PSUBWrm : PDI<0xF9, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1166 "psubw {$src2, $dst|$dst, $src2}",
1167 [(set VR128:$dst, (v8i16 (sub VR128:$src1,
1168 (load addr:$src2))))]>;
1169 def PSUBDrm : PDI<0xFA, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1170 "psubd {$src2, $dst|$dst, $src2}",
1171 [(set VR128:$dst, (v4i32 (sub VR128:$src1,
1172 (load addr:$src2))))]>;
1173 def PSUBQrm : PDI<0xFB, MRMSrcMem, (ops VR128:$dst, VR128:$src1, f128mem:$src2),
1174 "psubd {$src2, $dst|$dst, $src2}",
1175 [(set VR128:$dst, (v2i64 (sub VR128:$src1,
1176 (load addr:$src2))))]>;
1180 let isTwoAddress = 1 in {
1181 let isCommutable = 1 in {
1182 def PANDrr : PDI<0xDB, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1183 "pand {$src2, $dst|$dst, $src2}",
1184 [(set VR128:$dst, (v2i64 (and VR128:$src1, VR128:$src2)))]>;
1186 def PANDrm : PDI<0xDB, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1187 "pand {$src2, $dst|$dst, $src2}",
1188 [(set VR128:$dst, (v2i64 (and VR128:$src1,
1189 (load addr:$src2))))]>;
1190 def PORrr : PDI<0xDB, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1191 "por {$src2, $dst|$dst, $src2}",
1192 [(set VR128:$dst, (v2i64 (or VR128:$src1, VR128:$src2)))]>;
1194 def PORrm : PDI<0xDB, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1195 "por {$src2, $dst|$dst, $src2}",
1196 [(set VR128:$dst, (v2i64 (or VR128:$src1,
1197 (load addr:$src2))))]>;
1198 def PXORrr : PDI<0xEF, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1199 "pxor {$src2, $dst|$dst, $src2}",
1200 [(set VR128:$dst, (v2i64 (xor VR128:$src1, VR128:$src2)))]>;
1202 def PXORrm : PDI<0xEF, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1203 "pxor {$src2, $dst|$dst, $src2}",
1204 [(set VR128:$dst, (v2i64 (xor VR128:$src1,
1205 (load addr:$src2))))]>;
1208 def PANDNrr : PDI<0xDF, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR128:$src2),
1209 "pandn {$src2, $dst|$dst, $src2}",
1210 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1213 def PANDNrm : PDI<0xDF, MRMSrcMem, (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1214 "pandn {$src2, $dst|$dst, $src2}",
1215 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
1216 (load addr:$src2))))]>;
1219 // Pack instructions
1220 let isTwoAddress = 1 in {
1221 def PACKSSWBrr : PDI<0x63, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
1223 "packsswb {$src2, $dst|$dst, $src2}",
1224 [(set VR128:$dst, (v8i16 (int_x86_sse2_packsswb_128
1227 def PACKSSWBrm : PDI<0x63, MRMSrcMem, (ops VR128:$dst, VR128:$src1,
1229 "packsswb {$src2, $dst|$dst, $src2}",
1230 [(set VR128:$dst, (v8i16 (int_x86_sse2_packsswb_128
1232 (bc_v8i16 (loadv2f64 addr:$src2)))))]>;
1233 def PACKSSDWrr : PDI<0x6B, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
1235 "packssdw {$src2, $dst|$dst, $src2}",
1236 [(set VR128:$dst, (v4i32 (int_x86_sse2_packssdw_128
1239 def PACKSSDWrm : PDI<0x6B, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
1241 "packssdw {$src2, $dst|$dst, $src2}",
1242 [(set VR128:$dst, (v4i32 (int_x86_sse2_packssdw_128
1244 (bc_v4i32 (loadv2i64 addr:$src2)))))]>;
1245 def PACKUSWBrr : PDI<0x67, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
1247 "packuswb {$src2, $dst|$dst, $src2}",
1248 [(set VR128:$dst, (v8i16 (int_x86_sse2_packuswb_128
1251 def PACKUSWBrm : PDI<0x67, MRMSrcReg, (ops VR128:$dst, VR128:$src1,
1253 "packuswb {$src2, $dst|$dst, $src2}",
1254 [(set VR128:$dst, (v8i16 (int_x86_sse2_packuswb_128
1256 (bc_v8i16 (loadv2i64 addr:$src2)))))]>;
1259 // Shuffle and unpack instructions
1260 def PSHUFWrr : PSIi8<0x70, MRMDestReg,
1261 (ops VR64:$dst, VR64:$src1, i8imm:$src2),
1262 "pshufw {$src2, $src1, $dst|$dst, $src1, $src2}", []>;
1263 def PSHUFWrm : PSIi8<0x70, MRMSrcMem,
1264 (ops VR64:$dst, i64mem:$src1, i8imm:$src2),
1265 "pshufw {$src2, $src1, $dst|$dst, $src1, $src2}", []>;
1267 def PSHUFDrr : PDIi8<0x70, MRMDestReg,
1268 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1269 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1270 [(set VR128:$dst, (v4i32 (vector_shuffle
1271 VR128:$src1, (undef),
1272 PSHUFD_shuffle_mask:$src2)))]>;
1273 def PSHUFDrm : PDIi8<0x70, MRMSrcMem,
1274 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1275 "pshufd {$src2, $src1, $dst|$dst, $src1, $src2}",
1276 [(set VR128:$dst, (v4i32 (vector_shuffle
1277 (load addr:$src1), (undef),
1278 PSHUFD_shuffle_mask:$src2)))]>;
1280 // SSE2 with ImmT == Imm8 and XS prefix.
1281 def PSHUFHWrr : Ii8<0x70, MRMDestReg,
1282 (ops VR128:$dst, VR128:$src1, i8imm:$src2),
1283 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1284 [(set VR128:$dst, (v8i16 (vector_shuffle
1285 VR128:$src1, (undef),
1286 PSHUFHW_shuffle_mask:$src2)))]>,
1287 XS, Requires<[HasSSE2]>;
1288 def PSHUFHWrm : Ii8<0x70, MRMDestMem,
1289 (ops VR128:$dst, i128mem:$src1, i8imm:$src2),
1290 "pshufhw {$src2, $src1, $dst|$dst, $src1, $src2}",
1291 [(set VR128:$dst, (v8i16 (vector_shuffle
1292 (bc_v8i16 (loadv2i64 addr:$src1)), (undef),
1293 PSHUFHW_shuffle_mask:$src2)))]>,
1294 XS, Requires<[HasSSE2]>;
1296 // SSE2 with ImmT == Imm8 and XD prefix.
1297 def PSHUFLWrr : Ii8<0x70, MRMDestReg,
1298 (ops VR128:$dst, VR128:$src1, i32i8imm:$src2),
1299 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1300 [(set VR128:$dst, (v8i16 (vector_shuffle
1301 VR128:$src1, (undef),
1302 PSHUFLW_shuffle_mask:$src2)))]>,
1303 XD, Requires<[HasSSE2]>;
1304 def PSHUFLWrm : Ii8<0x70, MRMDestMem,
1305 (ops VR128:$dst, i128mem:$src1, i32i8imm:$src2),
1306 "pshuflw {$src2, $src1, $dst|$dst, $src1, $src2}",
1307 [(set VR128:$dst, (v8i16 (vector_shuffle
1308 (bc_v8i16 (loadv2i64 addr:$src1)), (undef),
1309 PSHUFLW_shuffle_mask:$src2)))]>,
1310 XD, Requires<[HasSSE2]>;
1312 let isTwoAddress = 1 in {
1313 def PUNPCKLBWrr : PDI<0x60, MRMSrcReg,
1314 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1315 "punpcklbw {$src2, $dst|$dst, $src2}",
1317 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1318 UNPCKL_shuffle_mask)))]>;
1319 def PUNPCKLBWrm : PDI<0x60, MRMSrcMem,
1320 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1321 "punpcklbw {$src2, $dst|$dst, $src2}",
1323 (v16i8 (vector_shuffle VR128:$src1, (load addr:$src2),
1324 UNPCKL_shuffle_mask)))]>;
1325 def PUNPCKLWDrr : PDI<0x61, MRMSrcReg,
1326 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1327 "punpcklwd {$src2, $dst|$dst, $src2}",
1329 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1330 UNPCKL_shuffle_mask)))]>;
1331 def PUNPCKLWDrm : PDI<0x61, MRMSrcMem,
1332 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1333 "punpcklwd {$src2, $dst|$dst, $src2}",
1335 (v8i16 (vector_shuffle VR128:$src1, (load addr:$src2),
1336 UNPCKL_shuffle_mask)))]>;
1337 def PUNPCKLDQrr : PDI<0x62, MRMSrcReg,
1338 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1339 "punpckldq {$src2, $dst|$dst, $src2}",
1341 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1342 UNPCKL_shuffle_mask)))]>;
1343 def PUNPCKLDQrm : PDI<0x62, MRMSrcMem,
1344 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1345 "punpckldq {$src2, $dst|$dst, $src2}",
1347 (v4i32 (vector_shuffle VR128:$src1, (load addr:$src2),
1348 UNPCKL_shuffle_mask)))]>;
1349 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
1350 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1351 "punpcklqdq {$src2, $dst|$dst, $src2}",
1353 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1354 UNPCKL_shuffle_mask)))]>;
1355 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
1356 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1357 "punpcklqdq {$src2, $dst|$dst, $src2}",
1359 (v2i64 (vector_shuffle VR128:$src1, (load addr:$src2),
1360 UNPCKL_shuffle_mask)))]>;
1362 def PUNPCKHBWrr : PDI<0x68, MRMSrcReg,
1363 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1364 "punpckhbw {$src2, $dst|$dst, $src2}",
1366 (v16i8 (vector_shuffle VR128:$src1, VR128:$src2,
1367 UNPCKH_shuffle_mask)))]>;
1368 def PUNPCKHBWrm : PDI<0x68, MRMSrcMem,
1369 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1370 "punpckhbw {$src2, $dst|$dst, $src2}",
1372 (v16i8 (vector_shuffle VR128:$src1, (load addr:$src2),
1373 UNPCKH_shuffle_mask)))]>;
1374 def PUNPCKHWDrr : PDI<0x69, MRMSrcReg,
1375 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1376 "punpckhwd {$src2, $dst|$dst, $src2}",
1378 (v8i16 (vector_shuffle VR128:$src1, VR128:$src2,
1379 UNPCKH_shuffle_mask)))]>;
1380 def PUNPCKHWDrm : PDI<0x69, MRMSrcMem,
1381 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1382 "punpckhwd {$src2, $dst|$dst, $src2}",
1384 (v8i16 (vector_shuffle VR128:$src1, (load addr:$src2),
1385 UNPCKH_shuffle_mask)))]>;
1386 def PUNPCKHDQrr : PDI<0x6A, MRMSrcReg,
1387 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1388 "punpckhdq {$src2, $dst|$dst, $src2}",
1390 (v4i32 (vector_shuffle VR128:$src1, VR128:$src2,
1391 UNPCKH_shuffle_mask)))]>;
1392 def PUNPCKHDQrm : PDI<0x6A, MRMSrcMem,
1393 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1394 "punpckhdq {$src2, $dst|$dst, $src2}",
1396 (v4i32 (vector_shuffle VR128:$src1, (load addr:$src2),
1397 UNPCKH_shuffle_mask)))]>;
1398 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
1399 (ops VR128:$dst, VR128:$src1, VR128:$src2),
1400 "punpckhdq {$src2, $dst|$dst, $src2}",
1402 (v2i64 (vector_shuffle VR128:$src1, VR128:$src2,
1403 UNPCKH_shuffle_mask)))]>;
1404 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
1405 (ops VR128:$dst, VR128:$src1, i128mem:$src2),
1406 "punpckhqdq {$src2, $dst|$dst, $src2}",
1408 (v2i64 (vector_shuffle VR128:$src1, (load addr:$src2),
1409 UNPCKH_shuffle_mask)))]>;
1412 //===----------------------------------------------------------------------===//
1413 // Miscellaneous Instructions
1414 //===----------------------------------------------------------------------===//
1417 def MOVMSKPSrr : PSI<0x50, MRMSrcReg, (ops R32:$dst, VR128:$src),
1418 "movmskps {$src, $dst|$dst, $src}",
1419 [(set R32:$dst, (int_x86_sse_movmsk_ps VR128:$src))]>;
1420 def MOVMSKPDrr : PSI<0x50, MRMSrcReg, (ops R32:$dst, VR128:$src),
1421 "movmskpd {$src, $dst|$dst, $src}",
1422 [(set R32:$dst, (int_x86_sse2_movmskpd VR128:$src))]>;
1424 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (ops R32:$dst, VR128:$src),
1425 "pmovmskb {$src, $dst|$dst, $src}",
1426 [(set R32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
1428 // Prefetching loads
1429 def PREFETCHT0 : I<0x18, MRM1m, (ops i8mem:$src),
1430 "prefetcht0 $src", []>, TB,
1431 Requires<[HasSSE1]>;
1432 def PREFETCHT1 : I<0x18, MRM2m, (ops i8mem:$src),
1433 "prefetcht0 $src", []>, TB,
1434 Requires<[HasSSE1]>;
1435 def PREFETCHT2 : I<0x18, MRM3m, (ops i8mem:$src),
1436 "prefetcht0 $src", []>, TB,
1437 Requires<[HasSSE1]>;
1438 def PREFETCHTNTA : I<0x18, MRM0m, (ops i8mem:$src),
1439 "prefetcht0 $src", []>, TB,
1440 Requires<[HasSSE1]>;
1442 // Non-temporal stores
1443 def MOVNTQ : I<0xE7, MRMDestMem, (ops i64mem:$dst, VR64:$src),
1444 "movntq {$src, $dst|$dst, $src}", []>, TB,
1445 Requires<[HasSSE1]>;
1446 def MOVNTPS : I<0x2B, MRMDestMem, (ops i128mem:$dst, VR128:$src),
1447 "movntps {$src, $dst|$dst, $src}", []>, TB,
1448 Requires<[HasSSE1]>;
1449 def MASKMOVQ : I<0xF7, MRMDestMem, (ops i64mem:$dst, VR64:$src),
1450 "maskmovq {$src, $dst|$dst, $src}", []>, TB,
1451 Requires<[HasSSE1]>;
1454 def SFENCE : I<0xAE, MRM7m, (ops),
1455 "sfence", []>, TB, Requires<[HasSSE1]>;
1457 // Load MXCSR register
1458 def LDMXCSR : I<0xAE, MRM2m, (ops i32mem:$src),
1459 "ldmxcsr {$src|$src}", []>, TB, Requires<[HasSSE1]>;
1461 //===----------------------------------------------------------------------===//
1462 // Alias Instructions
1463 //===----------------------------------------------------------------------===//
1465 // Alias instructions that map zero vector to pxor / xorp* for sse.
1466 // FIXME: remove when we can teach regalloc that xor reg, reg is ok.
1467 def V_SET0_PI : PDI<0xEF, MRMInitReg, (ops VR128:$dst),
1469 [(set VR128:$dst, (v2i64 immAllZerosV))]>;
1470 def V_SET0_PS : PSI<0x57, MRMInitReg, (ops VR128:$dst),
1472 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
1473 def V_SET0_PD : PDI<0x57, MRMInitReg, (ops VR128:$dst),
1475 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
1477 def V_SETALLONES : PDI<0x76, MRMInitReg, (ops VR128:$dst),
1478 "pcmpeqd $dst, $dst",
1479 [(set VR128:$dst, (v2f64 immAllOnesV))]>;
1481 // Scalar to 128-bit vector with zero extension.
1482 // Three operand (but two address) aliases.
1483 let isTwoAddress = 1 in {
1484 def MOVZSS128rr : SSI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR32:$src2),
1485 "movss {$src2, $dst|$dst, $src2}", []>;
1486 def MOVZSD128rr : SDI<0x10, MRMSrcReg, (ops VR128:$dst, VR128:$src1, FR64:$src2),
1487 "movsd {$src2, $dst|$dst, $src2}", []>;
1488 def MOVZD128rr : PDI<0x6E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, R32:$src2),
1489 "movd {$src2, $dst|$dst, $src2}", []>;
1490 def MOVZQ128rr : I<0x7E, MRMSrcReg, (ops VR128:$dst, VR128:$src1, VR64:$src2),
1491 "movq {$src2, $dst|$dst, $src2}", []>;
1494 // Loading from memory automatically zeroing upper bits.
1495 def MOVZSS128rm : SSI<0x10, MRMSrcMem, (ops VR128:$dst, f32mem:$src),
1496 "movss {$src, $dst|$dst, $src}",
1498 (v4f32 (X86zexts2vec (loadf32 addr:$src))))]>;
1499 def MOVZSD128rm : SDI<0x10, MRMSrcMem, (ops VR128:$dst, f64mem:$src),
1500 "movsd {$src, $dst|$dst, $src}",
1502 (v2f64 (X86zexts2vec (loadf64 addr:$src))))]>;
1503 def MOVZD128rm : PDI<0x6E, MRMSrcMem, (ops VR128:$dst, i32mem:$src),
1504 "movd {$src, $dst|$dst, $src}",
1506 (v4i32 (X86zexts2vec (loadi32 addr:$src))))]>;
1508 //===----------------------------------------------------------------------===//
1509 // Non-Instruction Patterns
1510 //===----------------------------------------------------------------------===//
1512 // 128-bit vector undef's.
1513 def : Pat<(v2f64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1514 def : Pat<(v16i8 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1515 def : Pat<(v8i16 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1516 def : Pat<(v4i32 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1517 def : Pat<(v2i64 (undef)), (IMPLICIT_DEF_VR128)>, Requires<[HasSSE2]>;
1519 // 128-bit vector all zero's.
1520 def : Pat<(v16i8 immAllZerosV), (v16i8 (V_SET0_PI))>, Requires<[HasSSE2]>;
1521 def : Pat<(v8i16 immAllZerosV), (v8i16 (V_SET0_PI))>, Requires<[HasSSE2]>;
1522 def : Pat<(v4i32 immAllZerosV), (v4i32 (V_SET0_PI))>, Requires<[HasSSE2]>;
1524 // 128-bit vector all one's.
1525 def : Pat<(v16i8 immAllOnesV), (v16i8 (V_SETALLONES))>, Requires<[HasSSE2]>;
1526 def : Pat<(v8i16 immAllOnesV), (v8i16 (V_SETALLONES))>, Requires<[HasSSE2]>;
1527 def : Pat<(v4i32 immAllOnesV), (v4i32 (V_SETALLONES))>, Requires<[HasSSE2]>;
1528 def : Pat<(v2i64 immAllOnesV), (v2i64 (V_SETALLONES))>, Requires<[HasSSE2]>;
1529 def : Pat<(v4f32 immAllOnesV), (v4f32 (V_SETALLONES))>, Requires<[HasSSE1]>;
1531 // Load 128-bit integer vector values.
1532 def : Pat<(v16i8 (load addr:$src)), (MOVDQArm addr:$src)>,
1533 Requires<[HasSSE2]>;
1534 def : Pat<(v8i16 (load addr:$src)), (MOVDQArm addr:$src)>,
1535 Requires<[HasSSE2]>;
1536 def : Pat<(v4i32 (load addr:$src)), (MOVDQArm addr:$src)>,
1537 Requires<[HasSSE2]>;
1538 def : Pat<(v2i64 (load addr:$src)), (MOVDQArm addr:$src)>,
1539 Requires<[HasSSE2]>;
1541 // Store 128-bit integer vector values.
1542 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
1543 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1544 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
1545 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1546 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
1547 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1548 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
1549 (MOVDQAmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
1551 // Scalar to v8i16 / v16i8. The source may be a R32, but only the lower 8 or
1553 def : Pat<(v8i16 (X86s2vec R32:$src)), (MOVD128rr R32:$src)>,
1554 Requires<[HasSSE2]>;
1555 def : Pat<(v16i8 (X86s2vec R32:$src)), (MOVD128rr R32:$src)>,
1556 Requires<[HasSSE2]>;
1559 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>,
1560 Requires<[HasSSE2]>;
1561 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>,
1562 Requires<[HasSSE2]>;
1563 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>,
1564 Requires<[HasSSE2]>;
1565 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
1566 Requires<[HasSSE2]>;
1567 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>,
1568 Requires<[HasSSE2]>;
1569 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>,
1570 Requires<[HasSSE2]>;
1571 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
1572 Requires<[HasSSE2]>;
1573 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v4i32 VR128:$src)>,
1574 Requires<[HasSSE2]>;
1575 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>,
1576 Requires<[HasSSE2]>;
1577 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>,
1578 Requires<[HasSSE2]>;
1579 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v4i32 VR128:$src)>,
1580 Requires<[HasSSE2]>;
1581 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>,
1582 Requires<[HasSSE2]>;
1584 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>,
1585 Requires<[HasSSE2]>;
1586 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>,
1587 Requires<[HasSSE2]>;
1589 // Zeroing a VR128 then do a MOVS* to the lower bits.
1590 def : Pat<(v2f64 (X86zexts2vec FR64:$src)),
1591 (MOVZSD128rr (V_SET0_PD), FR64:$src)>, Requires<[HasSSE2]>;
1592 def : Pat<(v4f32 (X86zexts2vec FR32:$src)),
1593 (MOVZSS128rr (V_SET0_PS), FR32:$src)>, Requires<[HasSSE2]>;
1594 def : Pat<(v2i64 (X86zexts2vec VR64:$src)),
1595 (MOVZQ128rr (V_SET0_PI), VR64:$src)>, Requires<[HasSSE2]>;
1596 def : Pat<(v4i32 (X86zexts2vec R32:$src)),
1597 (MOVZD128rr (V_SET0_PI), R32:$src)>, Requires<[HasSSE2]>;
1598 def : Pat<(v8i16 (X86zexts2vec R16:$src)),
1599 (MOVZD128rr (V_SET0_PI), (MOVZX32rr16 R16:$src))>, Requires<[HasSSE2]>;
1600 def : Pat<(v16i8 (X86zexts2vec R8:$src)),
1601 (MOVZD128rr (V_SET0_PI), (MOVZX32rr8 R8:$src))>, Requires<[HasSSE2]>;
1603 // Splat v2f64 / v2i64
1604 def : Pat<(vector_shuffle (v2f64 VR128:$src), (undef), SSE_splat_mask:$sm),
1605 (v2f64 (UNPCKLPDrr VR128:$src, VR128:$src))>, Requires<[HasSSE2]>;
1606 def : Pat<(vector_shuffle (v2i64 VR128:$src), (undef), SSE_splat_mask:$sm),
1607 (v2i64 (PUNPCKLQDQrr VR128:$src, VR128:$src))>, Requires<[HasSSE2]>;
1610 def : Pat<(vector_shuffle (v4f32 VR128:$src), (undef), SSE_splat_mask:$sm),
1611 (v4f32 (SHUFPSrr VR128:$src, VR128:$src, SSE_splat_mask:$sm))>,
1612 Requires<[HasSSE1]>;
1614 // Shuffle v4i32 with SHUFP* if others do not match.
1615 def : Pat<(vector_shuffle (v4i32 VR128:$src1), (v4i32 VR128:$src2),
1616 SHUFP_int_shuffle_mask:$sm),
1617 (v4i32 (SHUFPSrr VR128:$src1, VR128:$src2,
1618 SHUFP_int_shuffle_mask:$sm))>, Requires<[HasSSE2]>;
1619 def : Pat<(vector_shuffle (v4i32 VR128:$src1), (load addr:$src2),
1620 SHUFP_int_shuffle_mask:$sm),
1621 (v4i32 (SHUFPSrm VR128:$src1, addr:$src2,
1622 SHUFP_int_shuffle_mask:$sm))>, Requires<[HasSSE2]>;
1624 // Shuffle v4f32 with PSHUF* if others do not match.
1625 def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef),
1626 PSHUFD_fp_shuffle_mask:$sm),
1627 (v4f32 (PSHUFDrr VR128:$src1, PSHUFD_fp_shuffle_mask:$sm))>,
1628 Requires<[HasSSE2]>;
1629 def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef),
1630 PSHUFD_fp_shuffle_mask:$sm),
1631 (v4f32 (PSHUFDrm addr:$src1, PSHUFD_fp_shuffle_mask:$sm))>,
1632 Requires<[HasSSE2]>;
1633 def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef),
1634 PSHUFHW_fp_shuffle_mask:$sm),
1635 (v4f32 (PSHUFHWrr VR128:$src1, PSHUFHW_fp_shuffle_mask:$sm))>,
1636 Requires<[HasSSE2]>;
1637 def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef),
1638 PSHUFHW_fp_shuffle_mask:$sm),
1639 (v4f32 (PSHUFHWrm addr:$src1, PSHUFHW_fp_shuffle_mask:$sm))>,
1640 Requires<[HasSSE2]>;
1641 def : Pat<(vector_shuffle (v4f32 VR128:$src1), (undef),
1642 PSHUFLW_fp_shuffle_mask:$sm),
1643 (v4f32 (PSHUFLWrr VR128:$src1, PSHUFLW_fp_shuffle_mask:$sm))>,
1644 Requires<[HasSSE2]>;
1645 def : Pat<(vector_shuffle (loadv4f32 addr:$src1), (undef),
1646 PSHUFLW_fp_shuffle_mask:$sm),
1647 (v4f32 (PSHUFLWrm addr:$src1, PSHUFLW_fp_shuffle_mask:$sm))>,
1648 Requires<[HasSSE2]>;
1651 def : Pat<(and (bc_v4i32 (v4f32 VR128:$src1)), (loadv4i32 addr:$src2)),
1652 (ANDPSrm VR128:$src1, addr:$src2)>;
1653 def : Pat<(and (bc_v2i64 (v2f64 VR128:$src1)), (loadv2i64 addr:$src2)),
1654 (ANDPDrm VR128:$src1, addr:$src2)>;
1655 def : Pat<(or (bc_v4i32 (v4f32 VR128:$src1)), (loadv4i32 addr:$src2)),
1656 (ORPSrm VR128:$src1, addr:$src2)>;
1657 def : Pat<(or (bc_v2i64 (v2f64 VR128:$src1)), (loadv2i64 addr:$src2)),
1658 (ORPDrm VR128:$src1, addr:$src2)>;
1659 def : Pat<(xor (bc_v4i32 (v4f32 VR128:$src1)), (loadv4i32 addr:$src2)),
1660 (XORPSrm VR128:$src1, addr:$src2)>;
1661 def : Pat<(xor (bc_v2i64 (v2f64 VR128:$src1)), (loadv2i64 addr:$src2)),
1662 (XORPDrm VR128:$src1, addr:$src2)>;
1663 def : Pat<(and (vnot (bc_v4i32 (v4f32 VR128:$src1))), (loadv4i32 addr:$src2)),
1664 (ANDNPSrm VR128:$src1, addr:$src2)>;
1665 def : Pat<(and (vnot (bc_v2i64 (v2f64 VR128:$src1))), (loadv2i64 addr:$src2)),
1666 (ANDNPDrm VR128:$src1, addr:$src2)>;
1668 def : Pat<(bc_v4f32 (v4i32 (and VR128:$src1, VR128:$src2))),
1669 (ANDPSrr VR128:$src1, VR128:$src2)>;
1670 def : Pat<(bc_v4f32 (v4i32 (or VR128:$src1, VR128:$src2))),
1671 (ORPSrr VR128:$src1, VR128:$src2)>;
1672 def : Pat<(bc_v4f32 (v4i32 (xor VR128:$src1, VR128:$src2))),
1673 (XORPSrr VR128:$src1, VR128:$src2)>;
1674 def : Pat<(bc_v4f32 (v4i32 (and (vnot VR128:$src1), VR128:$src2))),
1675 (ANDNPSrr VR128:$src1, VR128:$src2)>;
1677 def : Pat<(bc_v4f32 (v4i32 (and VR128:$src1, (load addr:$src2)))),
1678 (ANDPSrm (v4i32 VR128:$src1), addr:$src2)>;
1679 def : Pat<(bc_v4f32 (v4i32 (or VR128:$src1, (load addr:$src2)))),
1680 (ORPSrm VR128:$src1, addr:$src2)>;
1681 def : Pat<(bc_v4f32 (v4i32 (xor VR128:$src1, (load addr:$src2)))),
1682 (XORPSrm VR128:$src1, addr:$src2)>;
1683 def : Pat<(bc_v4f32 (v4i32 (and (vnot VR128:$src1), (load addr:$src2)))),
1684 (ANDNPSrm VR128:$src1, addr:$src2)>;
1686 def : Pat<(bc_v2f64 (v2i64 (and VR128:$src1, VR128:$src2))),
1687 (ANDPDrr VR128:$src1, VR128:$src2)>;
1688 def : Pat<(bc_v2f64 (v2i64 (or VR128:$src1, VR128:$src2))),
1689 (ORPDrr VR128:$src1, VR128:$src2)>;
1690 def : Pat<(bc_v2f64 (v2i64 (xor VR128:$src1, VR128:$src2))),
1691 (XORPDrr VR128:$src1, VR128:$src2)>;
1692 def : Pat<(bc_v2f64 (v2i64 (and (vnot VR128:$src1), VR128:$src2))),
1693 (ANDNPDrr VR128:$src1, VR128:$src2)>;
1695 def : Pat<(bc_v2f64 (v2i64 (and VR128:$src1, (load addr:$src2)))),
1696 (ANDPSrm (v2i64 VR128:$src1), addr:$src2)>;
1697 def : Pat<(bc_v2f64 (v2i64 (or VR128:$src1, (load addr:$src2)))),
1698 (ORPSrm VR128:$src1, addr:$src2)>;
1699 def : Pat<(bc_v2f64 (v2i64 (xor VR128:$src1, (load addr:$src2)))),
1700 (XORPSrm VR128:$src1, addr:$src2)>;
1701 def : Pat<(bc_v2f64 (v2i64 (and (vnot VR128:$src1), (load addr:$src2)))),
1702 (ANDNPSrm VR128:$src1, addr:$src2)>;
1704 def : Pat<(v4i32 (and VR128:$src1, VR128:$src2)),
1705 (PANDrr VR128:$src1, VR128:$src2)>;
1706 def : Pat<(v8i16 (and VR128:$src1, VR128:$src2)),
1707 (PANDrr VR128:$src1, VR128:$src2)>;
1708 def : Pat<(v16i8 (and VR128:$src1, VR128:$src2)),
1709 (PANDrr VR128:$src1, VR128:$src2)>;
1710 def : Pat<(v4i32 (or VR128:$src1, VR128:$src2)),
1711 (PORrr VR128:$src1, VR128:$src2)>;
1712 def : Pat<(v8i16 (or VR128:$src1, VR128:$src2)),
1713 (PORrr VR128:$src1, VR128:$src2)>;
1714 def : Pat<(v16i8 (or VR128:$src1, VR128:$src2)),
1715 (PORrr VR128:$src1, VR128:$src2)>;
1716 def : Pat<(v4i32 (xor VR128:$src1, VR128:$src2)),
1717 (PXORrr VR128:$src1, VR128:$src2)>;
1718 def : Pat<(v8i16 (xor VR128:$src1, VR128:$src2)),
1719 (PXORrr VR128:$src1, VR128:$src2)>;
1720 def : Pat<(v16i8 (xor VR128:$src1, VR128:$src2)),
1721 (PXORrr VR128:$src1, VR128:$src2)>;
1722 def : Pat<(v4i32 (and (vnot VR128:$src1), VR128:$src2)),
1723 (PANDNrr VR128:$src1, VR128:$src2)>;
1724 def : Pat<(v8i16 (and (vnot VR128:$src1), VR128:$src2)),
1725 (PANDNrr VR128:$src1, VR128:$src2)>;
1726 def : Pat<(v16i8 (and (vnot VR128:$src1), VR128:$src2)),
1727 (PANDNrr VR128:$src1, VR128:$src2)>;
1729 def : Pat<(v4i32 (and VR128:$src1, (load addr:$src2))),
1730 (PANDrm VR128:$src1, addr:$src2)>;
1731 def : Pat<(v8i16 (and VR128:$src1, (load addr:$src2))),
1732 (PANDrm VR128:$src1, addr:$src2)>;
1733 def : Pat<(v16i8 (and VR128:$src1, (load addr:$src2))),
1734 (PANDrm VR128:$src1, addr:$src2)>;
1735 def : Pat<(v4i32 (or VR128:$src1, (load addr:$src2))),
1736 (PORrm VR128:$src1, addr:$src2)>;
1737 def : Pat<(v8i16 (or VR128:$src1, (load addr:$src2))),
1738 (PORrm VR128:$src1, addr:$src2)>;
1739 def : Pat<(v16i8 (or VR128:$src1, (load addr:$src2))),
1740 (PORrm VR128:$src1, addr:$src2)>;
1741 def : Pat<(v4i32 (xor VR128:$src1, (load addr:$src2))),
1742 (PXORrm VR128:$src1, addr:$src2)>;
1743 def : Pat<(v8i16 (xor VR128:$src1, (load addr:$src2))),
1744 (PXORrm VR128:$src1, addr:$src2)>;
1745 def : Pat<(v16i8 (xor VR128:$src1, (load addr:$src2))),
1746 (PXORrm VR128:$src1, addr:$src2)>;
1747 def : Pat<(v4i32 (and (vnot VR128:$src1), (load addr:$src2))),
1748 (PANDNrm VR128:$src1, addr:$src2)>;
1749 def : Pat<(v8i16 (and (vnot VR128:$src1), (load addr:$src2))),
1750 (PANDNrm VR128:$src1, addr:$src2)>;
1751 def : Pat<(v16i8 (and (vnot VR128:$src1), (load addr:$src2))),
1752 (PANDNrm VR128:$src1, addr:$src2)>;