1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE scalar FP Instructions
19 //===----------------------------------------------------------------------===//
21 // CMOV* - Used to implement the SSE SELECT DAG operation. Expanded after
22 // instruction selection into a branch sequence.
23 let Uses = [EFLAGS], usesCustomInserter = 1 in {
24 def CMOV_FR32 : I<0, Pseudo,
25 (outs FR32:$dst), (ins FR32:$t, FR32:$f, i8imm:$cond),
27 [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond,
29 def CMOV_FR64 : I<0, Pseudo,
30 (outs FR64:$dst), (ins FR64:$t, FR64:$f, i8imm:$cond),
32 [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond,
34 def CMOV_V4F32 : I<0, Pseudo,
35 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
36 "#CMOV_V4F32 PSEUDO!",
38 (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond,
40 def CMOV_V2F64 : I<0, Pseudo,
41 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
42 "#CMOV_V2F64 PSEUDO!",
44 (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
46 def CMOV_V2I64 : I<0, Pseudo,
47 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
48 "#CMOV_V2I64 PSEUDO!",
50 (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
54 //===----------------------------------------------------------------------===//
55 // SSE 1 & 2 Instructions Classes
56 //===----------------------------------------------------------------------===//
58 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
59 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
60 RegisterClass RC, X86MemOperand x86memop,
62 let isCommutable = 1 in {
63 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
65 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
66 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
67 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
69 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
71 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
72 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
73 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
76 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
77 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
78 string asm, string SSEVer, string FPSizeStr,
79 Operand memopr, ComplexPattern mem_cpat,
81 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
83 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
84 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
85 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
86 !strconcat(SSEVer, !strconcat("_",
87 !strconcat(OpcodeStr, FPSizeStr))))
88 RC:$src1, RC:$src2))]>;
89 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
91 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
92 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
93 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
94 !strconcat(SSEVer, !strconcat("_",
95 !strconcat(OpcodeStr, FPSizeStr))))
96 RC:$src1, mem_cpat:$src2))]>;
99 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
100 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
101 RegisterClass RC, ValueType vt,
102 X86MemOperand x86memop, PatFrag mem_frag,
103 Domain d, bit Is2Addr = 1> {
104 let isCommutable = 1 in
105 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
107 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
108 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
109 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
111 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
113 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
114 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
115 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
118 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
119 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
120 string OpcodeStr, X86MemOperand x86memop,
121 list<dag> pat_rr, list<dag> pat_rm,
123 let isCommutable = 1 in
124 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
126 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
127 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
129 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
131 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
132 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
136 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
137 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
138 string asm, string SSEVer, string FPSizeStr,
139 X86MemOperand x86memop, PatFrag mem_frag,
140 Domain d, bit Is2Addr = 1> {
141 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
143 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
144 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
145 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
146 !strconcat(SSEVer, !strconcat("_",
147 !strconcat(OpcodeStr, FPSizeStr))))
148 RC:$src1, RC:$src2))], d>;
149 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
151 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
152 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
153 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
154 !strconcat(SSEVer, !strconcat("_",
155 !strconcat(OpcodeStr, FPSizeStr))))
156 RC:$src1, (mem_frag addr:$src2)))], d>;
159 //===----------------------------------------------------------------------===//
160 // SSE 1 & 2 - Move Instructions
161 //===----------------------------------------------------------------------===//
163 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
164 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
165 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
167 // Loading from memory automatically zeroing upper bits.
168 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
169 PatFrag mem_pat, string OpcodeStr> :
170 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
171 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
172 [(set RC:$dst, (mem_pat addr:$src))]>;
174 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
175 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
176 // is used instead. Register-to-register movss/movsd is not modeled as an
177 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
178 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
179 let isAsmParserOnly = 1 in {
180 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
181 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
182 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
183 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
185 let canFoldAsLoad = 1, isReMaterializable = 1 in {
186 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
188 let AddedComplexity = 20 in
189 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
193 let Constraints = "$src1 = $dst" in {
194 def MOVSSrr : sse12_move_rr<FR32, v4f32,
195 "movss\t{$src2, $dst|$dst, $src2}">, XS;
196 def MOVSDrr : sse12_move_rr<FR64, v2f64,
197 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
200 let canFoldAsLoad = 1, isReMaterializable = 1 in {
201 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
203 let AddedComplexity = 20 in
204 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
207 let AddedComplexity = 15 in {
208 // Extract the low 32-bit value from one vector and insert it into another.
209 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
210 (MOVSSrr (v4f32 VR128:$src1),
211 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
212 // Extract the low 64-bit value from one vector and insert it into another.
213 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
214 (MOVSDrr (v2f64 VR128:$src1),
215 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
218 // Implicitly promote a 32-bit scalar to a vector.
219 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
220 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
221 // Implicitly promote a 64-bit scalar to a vector.
222 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
223 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
225 let AddedComplexity = 20 in {
226 // MOVSSrm zeros the high parts of the register; represent this
227 // with SUBREG_TO_REG.
228 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
229 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
230 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
231 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
232 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
233 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
234 // MOVSDrm zeros the high parts of the register; represent this
235 // with SUBREG_TO_REG.
236 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
237 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
238 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
239 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
240 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
241 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
242 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
243 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
244 def : Pat<(v2f64 (X86vzload addr:$src)),
245 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
248 // Store scalar value to memory.
249 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
250 "movss\t{$src, $dst|$dst, $src}",
251 [(store FR32:$src, addr:$dst)]>;
252 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
253 "movsd\t{$src, $dst|$dst, $src}",
254 [(store FR64:$src, addr:$dst)]>;
256 let isAsmParserOnly = 1 in {
257 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
258 "movss\t{$src, $dst|$dst, $src}",
259 [(store FR32:$src, addr:$dst)]>, XS, VEX;
260 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
261 "movsd\t{$src, $dst|$dst, $src}",
262 [(store FR64:$src, addr:$dst)]>, XD, VEX;
265 // Extract and store.
266 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
269 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
270 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
273 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
275 // Move Aligned/Unaligned floating point values
276 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
277 X86MemOperand x86memop, PatFrag ld_frag,
278 string asm, Domain d,
279 bit IsReMaterializable = 1> {
280 let neverHasSideEffects = 1 in
281 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
282 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
283 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
284 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
285 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
286 [(set RC:$dst, (ld_frag addr:$src))], d>;
289 let isAsmParserOnly = 1 in {
290 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
291 "movaps", SSEPackedSingle>, VEX;
292 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
293 "movapd", SSEPackedDouble>, OpSize, VEX;
294 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
295 "movups", SSEPackedSingle>, VEX;
296 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
297 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
299 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
300 "movaps", SSEPackedSingle>, VEX;
301 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
302 "movapd", SSEPackedDouble>, OpSize, VEX;
303 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
304 "movups", SSEPackedSingle>, VEX;
305 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
306 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
308 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
309 "movaps", SSEPackedSingle>, TB;
310 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
311 "movapd", SSEPackedDouble>, TB, OpSize;
312 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
313 "movups", SSEPackedSingle>, TB;
314 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
315 "movupd", SSEPackedDouble, 0>, TB, OpSize;
317 let isAsmParserOnly = 1 in {
318 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movaps\t{$src, $dst|$dst, $src}",
320 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
321 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
322 "movapd\t{$src, $dst|$dst, $src}",
323 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
324 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
327 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
328 "movupd\t{$src, $dst|$dst, $src}",
329 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
330 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
331 "movaps\t{$src, $dst|$dst, $src}",
332 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
333 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
334 "movapd\t{$src, $dst|$dst, $src}",
335 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
336 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
337 "movups\t{$src, $dst|$dst, $src}",
338 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
339 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
340 "movupd\t{$src, $dst|$dst, $src}",
341 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
343 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
344 "movaps\t{$src, $dst|$dst, $src}",
345 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
346 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
347 "movapd\t{$src, $dst|$dst, $src}",
348 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
349 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
350 "movups\t{$src, $dst|$dst, $src}",
351 [(store (v4f32 VR128:$src), addr:$dst)]>;
352 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
353 "movupd\t{$src, $dst|$dst, $src}",
354 [(store (v2f64 VR128:$src), addr:$dst)]>;
356 // Intrinsic forms of MOVUPS/D load and store
357 let isAsmParserOnly = 1 in {
358 let canFoldAsLoad = 1, isReMaterializable = 1 in
359 def VMOVUPSrm_Int : VPSI<0x10, MRMSrcMem, (outs VR128:$dst),
361 "movups\t{$src, $dst|$dst, $src}",
362 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>, VEX;
363 def VMOVUPDrm_Int : VPDI<0x10, MRMSrcMem, (outs VR128:$dst),
365 "movupd\t{$src, $dst|$dst, $src}",
366 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>, VEX;
367 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
368 (ins f128mem:$dst, VR128:$src),
369 "movups\t{$src, $dst|$dst, $src}",
370 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
371 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
372 (ins f128mem:$dst, VR128:$src),
373 "movupd\t{$src, $dst|$dst, $src}",
374 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
376 let canFoldAsLoad = 1, isReMaterializable = 1 in
377 def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
378 "movups\t{$src, $dst|$dst, $src}",
379 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
380 def MOVUPDrm_Int : PDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
381 "movupd\t{$src, $dst|$dst, $src}",
382 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
384 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
385 "movups\t{$src, $dst|$dst, $src}",
386 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
387 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
388 "movupd\t{$src, $dst|$dst, $src}",
389 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
391 // Move Low/High packed floating point values
392 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
393 PatFrag mov_frag, string base_opc,
395 def PSrm : PI<opc, MRMSrcMem,
396 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
397 !strconcat(!strconcat(base_opc,"s"), asm_opr),
400 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
401 SSEPackedSingle>, TB;
403 def PDrm : PI<opc, MRMSrcMem,
404 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
405 !strconcat(!strconcat(base_opc,"d"), asm_opr),
406 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
407 (scalar_to_vector (loadf64 addr:$src2)))))],
408 SSEPackedDouble>, TB, OpSize;
411 let isAsmParserOnly = 1, AddedComplexity = 20 in {
412 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
413 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
414 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
415 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
417 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
418 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
419 "\t{$src2, $dst|$dst, $src2}">;
420 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
421 "\t{$src2, $dst|$dst, $src2}">;
424 let isAsmParserOnly = 1 in {
425 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
426 "movlps\t{$src, $dst|$dst, $src}",
427 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
428 (iPTR 0))), addr:$dst)]>, VEX;
429 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
430 "movlpd\t{$src, $dst|$dst, $src}",
431 [(store (f64 (vector_extract (v2f64 VR128:$src),
432 (iPTR 0))), addr:$dst)]>, VEX;
434 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
435 "movlps\t{$src, $dst|$dst, $src}",
436 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
437 (iPTR 0))), addr:$dst)]>;
438 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
439 "movlpd\t{$src, $dst|$dst, $src}",
440 [(store (f64 (vector_extract (v2f64 VR128:$src),
441 (iPTR 0))), addr:$dst)]>;
443 // v2f64 extract element 1 is always custom lowered to unpack high to low
444 // and extract element 0 so the non-store version isn't too horrible.
445 let isAsmParserOnly = 1 in {
446 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
447 "movhps\t{$src, $dst|$dst, $src}",
448 [(store (f64 (vector_extract
449 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
450 (undef)), (iPTR 0))), addr:$dst)]>,
452 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
453 "movhpd\t{$src, $dst|$dst, $src}",
454 [(store (f64 (vector_extract
455 (v2f64 (unpckh VR128:$src, (undef))),
456 (iPTR 0))), addr:$dst)]>,
459 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
460 "movhps\t{$src, $dst|$dst, $src}",
461 [(store (f64 (vector_extract
462 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
463 (undef)), (iPTR 0))), addr:$dst)]>;
464 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
465 "movhpd\t{$src, $dst|$dst, $src}",
466 [(store (f64 (vector_extract
467 (v2f64 (unpckh VR128:$src, (undef))),
468 (iPTR 0))), addr:$dst)]>;
470 let isAsmParserOnly = 1, AddedComplexity = 20 in {
471 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
472 (ins VR128:$src1, VR128:$src2),
473 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
475 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
477 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
478 (ins VR128:$src1, VR128:$src2),
479 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
481 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
484 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
485 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
486 (ins VR128:$src1, VR128:$src2),
487 "movlhps\t{$src2, $dst|$dst, $src2}",
489 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
490 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
491 (ins VR128:$src1, VR128:$src2),
492 "movhlps\t{$src2, $dst|$dst, $src2}",
494 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
497 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
498 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
499 let AddedComplexity = 20 in {
500 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
501 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
502 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
503 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
506 //===----------------------------------------------------------------------===//
507 // SSE 1 & 2 - Conversion Instructions
508 //===----------------------------------------------------------------------===//
510 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
511 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
513 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
514 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
515 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
516 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
519 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
520 X86MemOperand x86memop, string asm> {
521 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
523 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
527 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
528 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
529 string asm, Domain d> {
530 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
531 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
532 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
533 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
536 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
537 X86MemOperand x86memop, string asm> {
538 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
539 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
540 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
541 (ins DstRC:$src1, x86memop:$src),
542 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
545 let isAsmParserOnly = 1 in {
546 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
547 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
548 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
549 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
551 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
552 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
553 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
554 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
557 // The assembler can recognize rr 64-bit instructions by seeing a rxx
558 // register, but the same isn't true when only using memory operands,
559 // provide other assembly "l" and "q" forms to address this explicitly
560 // where appropriate to do so.
561 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
563 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
565 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
567 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
569 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
573 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
574 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
575 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
576 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
577 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
578 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
579 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
580 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
581 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
582 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
583 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
584 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
585 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
586 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
587 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
588 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
590 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
591 // and/or XMM operand(s).
592 multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
593 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
594 string asm, Domain d> {
595 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
596 [(set DstRC:$dst, (Int SrcRC:$src))], d>;
597 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
598 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], d>;
601 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
602 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
604 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
605 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
606 [(set DstRC:$dst, (Int SrcRC:$src))]>;
607 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
608 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
609 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
612 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
613 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
614 PatFrag ld_frag, string asm, Domain d> {
615 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
616 asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], d>;
617 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
618 (ins DstRC:$src1, x86memop:$src2), asm,
619 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], d>;
622 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
623 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
624 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
625 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
627 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
628 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
629 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
630 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
631 (ins DstRC:$src1, x86memop:$src2),
633 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
634 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
635 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
638 let isAsmParserOnly = 1 in {
639 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
640 f32mem, load, "cvtss2si">, XS, VEX;
641 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
642 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
644 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
645 f128mem, load, "cvtsd2si">, XD, VEX;
646 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
647 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
650 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
651 // Get rid of this hack or rename the intrinsics, there are several
652 // intructions that only match with the intrinsic form, why create duplicates
653 // to let them be recognized by the assembler?
654 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
655 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
656 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
657 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
659 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
660 f32mem, load, "cvtss2si">, XS;
661 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
662 f32mem, load, "cvtss2si{q}">, XS, REX_W;
663 defm Int_CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
664 f128mem, load, "cvtsd2si">, XD;
665 defm Int_CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
666 f128mem, load, "cvtsd2si">, XD, REX_W;
668 defm CVTSD2SI64 : sse12_cvt_s_np<0x2D, VR128, GR64, f64mem, "cvtsd2si{q}">, XD,
671 let isAsmParserOnly = 1 in {
672 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
673 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
674 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
675 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
677 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
678 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
679 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
680 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
684 let Constraints = "$src1 = $dst" in {
685 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
686 int_x86_sse_cvtsi2ss, i32mem, loadi32,
688 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
689 int_x86_sse_cvtsi642ss, i64mem, loadi64,
690 "cvtsi2ss{q}">, XS, REX_W;
691 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
692 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
694 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
695 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
696 "cvtsi2sd">, XD, REX_W;
699 // Instructions below don't have an AVX form.
700 defm Int_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
701 f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
702 SSEPackedSingle>, TB;
703 defm Int_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
704 f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
705 SSEPackedDouble>, TB, OpSize;
706 defm Int_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
707 f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
708 SSEPackedSingle>, TB;
709 defm Int_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
710 f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
711 SSEPackedDouble>, TB, OpSize;
712 defm Int_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
713 i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
714 SSEPackedDouble>, TB, OpSize;
715 let Constraints = "$src1 = $dst" in {
716 defm Int_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
717 int_x86_sse_cvtpi2ps,
718 i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
719 SSEPackedSingle>, TB;
724 // Aliases for intrinsics
725 let isAsmParserOnly = 1 in {
726 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
727 f32mem, load, "cvttss2si">, XS, VEX;
728 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
729 int_x86_sse_cvttss2si64, f32mem, load,
730 "cvttss2si">, XS, VEX, VEX_W;
731 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
732 f128mem, load, "cvttss2si">, XD, VEX;
733 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
734 int_x86_sse2_cvttsd2si64, f128mem, load,
735 "cvttss2si">, XD, VEX, VEX_W;
737 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
738 f32mem, load, "cvttss2si">, XS;
739 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
740 int_x86_sse_cvttss2si64, f32mem, load,
741 "cvttss2si{q}">, XS, REX_W;
742 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
743 f128mem, load, "cvttss2si">, XD;
744 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
745 int_x86_sse2_cvttsd2si64, f128mem, load,
746 "cvttss2si{q}">, XD, REX_W;
748 let isAsmParserOnly = 1, Pattern = []<dag> in {
749 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
750 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
751 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
752 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
754 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
755 "cvtdq2ps\t{$src, $dst|$dst, $src}",
756 SSEPackedSingle>, TB, VEX;
757 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
758 "cvtdq2ps\t{$src, $dst|$dst, $src}",
759 SSEPackedSingle>, TB, VEX;
761 let Pattern = []<dag> in {
762 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
763 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
764 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
765 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
766 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
767 "cvtdq2ps\t{$src, $dst|$dst, $src}",
768 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
773 // Convert scalar double to scalar single
774 let isAsmParserOnly = 1 in {
775 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
776 (ins FR64:$src1, FR64:$src2),
777 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
779 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
780 (ins FR64:$src1, f64mem:$src2),
781 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
782 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
784 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
785 "cvtsd2ss\t{$src, $dst|$dst, $src}",
786 [(set FR32:$dst, (fround FR64:$src))]>;
787 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
788 "cvtsd2ss\t{$src, $dst|$dst, $src}",
789 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
790 Requires<[HasSSE2, OptForSize]>;
792 let isAsmParserOnly = 1 in
793 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
794 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
796 let Constraints = "$src1 = $dst" in
797 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
798 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
800 // Convert scalar single to scalar double
801 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
802 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
803 (ins FR32:$src1, FR32:$src2),
804 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
805 []>, XS, Requires<[HasAVX]>, VEX_4V;
806 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
807 (ins FR32:$src1, f32mem:$src2),
808 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
809 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
811 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
812 "cvtss2sd\t{$src, $dst|$dst, $src}",
813 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
815 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
816 "cvtss2sd\t{$src, $dst|$dst, $src}",
817 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
818 Requires<[HasSSE2, OptForSize]>;
820 let isAsmParserOnly = 1 in {
821 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
822 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
823 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
824 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
825 VR128:$src2))]>, XS, VEX_4V,
827 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
828 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
829 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
830 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
831 (load addr:$src2)))]>, XS, VEX_4V,
834 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
835 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
836 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
837 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
838 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
841 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
842 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
843 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
844 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
845 (load addr:$src2)))]>, XS,
849 def : Pat<(extloadf32 addr:$src),
850 (CVTSS2SDrr (MOVSSrm addr:$src))>,
851 Requires<[HasSSE2, OptForSpeed]>;
853 // Convert doubleword to packed single/double fp
854 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
855 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
856 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
857 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
858 TB, VEX, Requires<[HasAVX]>;
859 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
860 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
861 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
862 (bitconvert (memopv2i64 addr:$src))))]>,
863 TB, VEX, Requires<[HasAVX]>;
865 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
866 "cvtdq2ps\t{$src, $dst|$dst, $src}",
867 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
868 TB, Requires<[HasSSE2]>;
869 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
870 "cvtdq2ps\t{$src, $dst|$dst, $src}",
871 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
872 (bitconvert (memopv2i64 addr:$src))))]>,
873 TB, Requires<[HasSSE2]>;
875 // FIXME: why the non-intrinsic version is described as SSE3?
876 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
877 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
878 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
879 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
880 XS, VEX, Requires<[HasAVX]>;
881 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
882 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
883 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
884 (bitconvert (memopv2i64 addr:$src))))]>,
885 XS, VEX, Requires<[HasAVX]>;
887 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
888 "cvtdq2pd\t{$src, $dst|$dst, $src}",
889 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
890 XS, Requires<[HasSSE2]>;
891 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
892 "cvtdq2pd\t{$src, $dst|$dst, $src}",
893 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
894 (bitconvert (memopv2i64 addr:$src))))]>,
895 XS, Requires<[HasSSE2]>;
898 // Convert packed single/double fp to doubleword
899 let isAsmParserOnly = 1 in {
900 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
901 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
902 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
903 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
904 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
905 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
906 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
907 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
909 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
910 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
911 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
912 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
914 let isAsmParserOnly = 1 in {
915 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
916 "cvtps2dq\t{$src, $dst|$dst, $src}",
917 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
919 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
921 "cvtps2dq\t{$src, $dst|$dst, $src}",
922 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
923 (memop addr:$src)))]>, VEX;
925 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
926 "cvtps2dq\t{$src, $dst|$dst, $src}",
927 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
928 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
929 "cvtps2dq\t{$src, $dst|$dst, $src}",
930 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
931 (memop addr:$src)))]>;
933 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
934 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
935 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
936 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
937 XD, VEX, Requires<[HasAVX]>;
938 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
939 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
940 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
941 (memop addr:$src)))]>,
942 XD, VEX, Requires<[HasAVX]>;
944 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
945 "cvtpd2dq\t{$src, $dst|$dst, $src}",
946 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
947 XD, Requires<[HasSSE2]>;
948 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
949 "cvtpd2dq\t{$src, $dst|$dst, $src}",
950 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
951 (memop addr:$src)))]>,
952 XD, Requires<[HasSSE2]>;
955 // Convert with truncation packed single/double fp to doubleword
956 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
957 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
958 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
959 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
960 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
961 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
962 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
963 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
964 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
966 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
967 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
968 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
969 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
972 let isAsmParserOnly = 1 in {
973 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
974 "vcvttps2dq\t{$src, $dst|$dst, $src}",
976 (int_x86_sse2_cvttps2dq VR128:$src))]>,
977 XS, VEX, Requires<[HasAVX]>;
978 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
979 "vcvttps2dq\t{$src, $dst|$dst, $src}",
980 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
981 (memop addr:$src)))]>,
982 XS, VEX, Requires<[HasAVX]>;
984 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
985 "cvttps2dq\t{$src, $dst|$dst, $src}",
987 (int_x86_sse2_cvttps2dq VR128:$src))]>,
988 XS, Requires<[HasSSE2]>;
989 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
990 "cvttps2dq\t{$src, $dst|$dst, $src}",
991 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
992 (memop addr:$src)))]>,
993 XS, Requires<[HasSSE2]>;
995 let isAsmParserOnly = 1 in {
996 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
998 "cvttpd2dq\t{$src, $dst|$dst, $src}",
999 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1001 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1003 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1004 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1005 (memop addr:$src)))]>, VEX;
1007 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1008 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1009 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1010 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1011 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1012 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1013 (memop addr:$src)))]>;
1015 let isAsmParserOnly = 1 in {
1016 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1017 // register, but the same isn't true when using memory operands instead.
1018 // Provide other assembly rr and rm forms to address this explicitly.
1019 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1020 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1021 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1022 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1025 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1026 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1027 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1028 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1031 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1032 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1033 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1034 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1037 // Convert packed single to packed double
1038 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1039 // SSE2 instructions without OpSize prefix
1040 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1041 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1042 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1043 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1044 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1045 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1046 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1047 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1049 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1050 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1051 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1052 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1054 let isAsmParserOnly = 1 in {
1055 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1056 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1057 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1058 VEX, Requires<[HasAVX]>;
1059 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1060 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1061 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1062 (load addr:$src)))]>,
1063 VEX, Requires<[HasAVX]>;
1065 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1066 "cvtps2pd\t{$src, $dst|$dst, $src}",
1067 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1068 TB, Requires<[HasSSE2]>;
1069 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1070 "cvtps2pd\t{$src, $dst|$dst, $src}",
1071 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1072 (load addr:$src)))]>,
1073 TB, Requires<[HasSSE2]>;
1075 // Convert packed double to packed single
1076 let isAsmParserOnly = 1 in {
1077 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1078 // register, but the same isn't true when using memory operands instead.
1079 // Provide other assembly rr and rm forms to address this explicitly.
1080 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1081 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1082 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1083 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1086 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1087 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1088 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1089 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1092 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1093 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1094 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1095 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1097 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1098 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1099 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1100 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1103 let isAsmParserOnly = 1 in {
1104 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1105 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1106 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1107 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1109 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1110 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1111 (memop addr:$src)))]>;
1113 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1114 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1115 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1116 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1117 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1118 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1119 (memop addr:$src)))]>;
1121 // AVX 256-bit register conversion intrinsics
1122 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1123 // whenever possible to avoid declaring two versions of each one.
1124 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1125 (VCVTDQ2PSYrr VR256:$src)>;
1126 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1127 (VCVTDQ2PSYrm addr:$src)>;
1129 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1130 (VCVTPD2PSYrr VR256:$src)>;
1131 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1132 (VCVTPD2PSYrm addr:$src)>;
1134 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1135 (VCVTPS2DQYrr VR256:$src)>;
1136 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1137 (VCVTPS2DQYrm addr:$src)>;
1139 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1140 (VCVTPS2PDYrr VR128:$src)>;
1141 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1142 (VCVTPS2PDYrm addr:$src)>;
1144 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1145 (VCVTTPD2DQYrr VR256:$src)>;
1146 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1147 (VCVTTPD2DQYrm addr:$src)>;
1149 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1150 (VCVTTPS2DQYrr VR256:$src)>;
1151 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1152 (VCVTTPS2DQYrm addr:$src)>;
1154 //===----------------------------------------------------------------------===//
1155 // SSE 1 & 2 - Compare Instructions
1156 //===----------------------------------------------------------------------===//
1158 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1159 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1160 string asm, string asm_alt> {
1161 def rr : SIi8<0xC2, MRMSrcReg,
1162 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1165 def rm : SIi8<0xC2, MRMSrcMem,
1166 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1168 // Accept explicit immediate argument form instead of comparison code.
1169 let isAsmParserOnly = 1 in {
1170 def rr_alt : SIi8<0xC2, MRMSrcReg,
1171 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1174 def rm_alt : SIi8<0xC2, MRMSrcMem,
1175 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1180 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1181 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1182 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1183 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1185 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1186 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1187 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1191 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1192 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1193 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1194 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1195 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1196 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1197 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1200 multiclass sse12_cmp_scalar_int<RegisterClass RC, Operand memopr,
1201 ComplexPattern mem_cpat, Intrinsic Int, string asm> {
1202 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1203 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1204 [(set VR128:$dst, (Int VR128:$src1,
1205 VR128:$src, imm:$cc))]>;
1206 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1207 (ins VR128:$src1, memopr:$src, SSECC:$cc), asm,
1208 [(set VR128:$dst, (Int VR128:$src1,
1209 mem_cpat:$src, imm:$cc))]>;
1212 // Aliases to match intrinsics which expect XMM operand(s).
1214 let isAsmParserOnly = 1 in {
1215 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, ssmem, sse_load_f32,
1217 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1219 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, sdmem, sse_load_f64,
1220 int_x86_sse2_cmp_sd,
1221 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1224 let Constraints = "$src1 = $dst" in {
1225 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, ssmem, sse_load_f32,
1227 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1228 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, sdmem, sse_load_f64,
1229 int_x86_sse2_cmp_sd,
1230 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1234 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1235 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1236 ValueType vt, X86MemOperand x86memop,
1237 PatFrag ld_frag, string OpcodeStr, Domain d> {
1238 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1239 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1240 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1241 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1242 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1243 [(set EFLAGS, (OpNode (vt RC:$src1),
1244 (ld_frag addr:$src2)))], d>;
1247 let Defs = [EFLAGS] in {
1248 let isAsmParserOnly = 1 in {
1249 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1250 "ucomiss", SSEPackedSingle>, VEX;
1251 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1252 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1253 let Pattern = []<dag> in {
1254 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1255 "comiss", SSEPackedSingle>, VEX;
1256 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1257 "comisd", SSEPackedDouble>, OpSize, VEX;
1260 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1261 load, "ucomiss", SSEPackedSingle>, VEX;
1262 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1263 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1265 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1266 load, "comiss", SSEPackedSingle>, VEX;
1267 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1268 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1270 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1271 "ucomiss", SSEPackedSingle>, TB;
1272 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1273 "ucomisd", SSEPackedDouble>, TB, OpSize;
1275 let Pattern = []<dag> in {
1276 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1277 "comiss", SSEPackedSingle>, TB;
1278 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1279 "comisd", SSEPackedDouble>, TB, OpSize;
1282 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1283 load, "ucomiss", SSEPackedSingle>, TB;
1284 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1285 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1287 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1288 "comiss", SSEPackedSingle>, TB;
1289 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1290 "comisd", SSEPackedDouble>, TB, OpSize;
1291 } // Defs = [EFLAGS]
1293 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1294 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1295 Intrinsic Int, string asm, string asm_alt,
1297 def rri : PIi8<0xC2, MRMSrcReg,
1298 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1299 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1300 def rmi : PIi8<0xC2, MRMSrcMem,
1301 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1302 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1303 // Accept explicit immediate argument form instead of comparison code.
1304 let isAsmParserOnly = 1 in {
1305 def rri_alt : PIi8<0xC2, MRMSrcReg,
1306 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1308 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1309 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1314 let isAsmParserOnly = 1 in {
1315 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1316 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1317 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1318 SSEPackedSingle>, VEX_4V;
1319 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1320 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1321 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1322 SSEPackedDouble>, OpSize, VEX_4V;
1323 let Pattern = []<dag> in {
1324 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_sse_cmp_ps,
1325 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1326 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1327 SSEPackedSingle>, VEX_4V;
1328 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_sse2_cmp_pd,
1329 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1330 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1331 SSEPackedDouble>, OpSize, VEX_4V;
1334 let Constraints = "$src1 = $dst" in {
1335 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1336 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1337 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1338 SSEPackedSingle>, TB;
1339 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1340 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1341 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1342 SSEPackedDouble>, TB, OpSize;
1345 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1346 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1347 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1348 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1349 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1350 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1351 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1352 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1354 //===----------------------------------------------------------------------===//
1355 // SSE 1 & 2 - Shuffle Instructions
1356 //===----------------------------------------------------------------------===//
1358 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1359 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1360 ValueType vt, string asm, PatFrag mem_frag,
1361 Domain d, bit IsConvertibleToThreeAddress = 0> {
1362 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1363 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1364 [(set RC:$dst, (vt (shufp:$src3
1365 RC:$src1, (mem_frag addr:$src2))))], d>;
1366 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1367 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1368 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1370 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1373 let isAsmParserOnly = 1 in {
1374 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1375 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1376 memopv4f32, SSEPackedSingle>, VEX_4V;
1377 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1378 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1379 memopv8f32, SSEPackedSingle>, VEX_4V;
1380 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1381 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1382 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1383 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1384 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1385 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1388 let Constraints = "$src1 = $dst" in {
1389 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1390 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1391 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1393 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1394 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1395 memopv2f64, SSEPackedDouble>, TB, OpSize;
1398 //===----------------------------------------------------------------------===//
1399 // SSE 1 & 2 - Unpack Instructions
1400 //===----------------------------------------------------------------------===//
1402 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1403 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1404 PatFrag mem_frag, RegisterClass RC,
1405 X86MemOperand x86memop, string asm,
1407 def rr : PI<opc, MRMSrcReg,
1408 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1410 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1411 def rm : PI<opc, MRMSrcMem,
1412 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1414 (vt (OpNode RC:$src1,
1415 (mem_frag addr:$src2))))], d>;
1418 let AddedComplexity = 10 in {
1419 let isAsmParserOnly = 1 in {
1420 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1421 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1422 SSEPackedSingle>, VEX_4V;
1423 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1424 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1425 SSEPackedDouble>, OpSize, VEX_4V;
1426 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1427 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1428 SSEPackedSingle>, VEX_4V;
1429 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1430 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1431 SSEPackedDouble>, OpSize, VEX_4V;
1433 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1434 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1435 SSEPackedSingle>, VEX_4V;
1436 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1437 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1438 SSEPackedDouble>, OpSize, VEX_4V;
1439 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1440 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1441 SSEPackedSingle>, VEX_4V;
1442 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1443 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1444 SSEPackedDouble>, OpSize, VEX_4V;
1447 let Constraints = "$src1 = $dst" in {
1448 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1449 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1450 SSEPackedSingle>, TB;
1451 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1452 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1453 SSEPackedDouble>, TB, OpSize;
1454 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1455 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1456 SSEPackedSingle>, TB;
1457 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1458 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1459 SSEPackedDouble>, TB, OpSize;
1460 } // Constraints = "$src1 = $dst"
1461 } // AddedComplexity
1463 //===----------------------------------------------------------------------===//
1464 // SSE 1 & 2 - Extract Floating-Point Sign mask
1465 //===----------------------------------------------------------------------===//
1467 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1468 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1470 def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1471 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1472 [(set GR32:$dst, (Int RC:$src))], d>;
1476 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1477 SSEPackedSingle>, TB;
1478 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1479 SSEPackedDouble>, TB, OpSize;
1481 let isAsmParserOnly = 1 in {
1482 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1483 "movmskps", SSEPackedSingle>, VEX;
1484 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1485 "movmskpd", SSEPackedDouble>, OpSize,
1488 // FIXME: merge with multiclass above when the intrinsics come.
1489 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1490 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1491 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1492 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1495 def VMOVMSKPSYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1496 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1497 def VMOVMSKPDYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1498 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1501 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1502 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1503 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1504 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1508 //===----------------------------------------------------------------------===//
1509 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1510 //===----------------------------------------------------------------------===//
1512 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1513 // names that start with 'Fs'.
1515 // Alias instructions that map fld0 to pxor for sse.
1516 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1517 canFoldAsLoad = 1 in {
1518 // FIXME: Set encoding to pseudo!
1519 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1520 [(set FR32:$dst, fp32imm0)]>,
1521 Requires<[HasSSE1]>, TB, OpSize;
1522 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1523 [(set FR64:$dst, fpimm0)]>,
1524 Requires<[HasSSE2]>, TB, OpSize;
1527 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1528 // bits are disregarded.
1529 let neverHasSideEffects = 1 in {
1530 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1531 "movaps\t{$src, $dst|$dst, $src}", []>;
1532 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1533 "movapd\t{$src, $dst|$dst, $src}", []>;
1536 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1537 // bits are disregarded.
1538 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1539 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1540 "movaps\t{$src, $dst|$dst, $src}",
1541 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1542 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1543 "movapd\t{$src, $dst|$dst, $src}",
1544 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1547 //===----------------------------------------------------------------------===//
1548 // SSE 1 & 2 - Logical Instructions
1549 //===----------------------------------------------------------------------===//
1551 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1553 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1555 let isAsmParserOnly = 1 in {
1556 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1557 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1559 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1560 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1563 let Constraints = "$src1 = $dst" in {
1564 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1565 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1567 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1568 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1572 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1573 let mayLoad = 0 in {
1574 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1575 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1576 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1579 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1580 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1582 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1584 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1585 SDNode OpNode, int HasPat = 0,
1586 list<list<dag>> Pattern = []> {
1587 let isAsmParserOnly = 1, Pattern = []<dag> in {
1588 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1589 !strconcat(OpcodeStr, "ps"), f128mem,
1590 !if(HasPat, Pattern[0], // rr
1591 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1593 !if(HasPat, Pattern[2], // rm
1594 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1595 (memopv2i64 addr:$src2)))]), 0>,
1598 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1599 !strconcat(OpcodeStr, "pd"), f128mem,
1600 !if(HasPat, Pattern[1], // rr
1601 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1604 !if(HasPat, Pattern[3], // rm
1605 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1606 (memopv2i64 addr:$src2)))]), 0>,
1609 let Constraints = "$src1 = $dst" in {
1610 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1611 !strconcat(OpcodeStr, "ps"), f128mem,
1612 !if(HasPat, Pattern[0], // rr
1613 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1615 !if(HasPat, Pattern[2], // rm
1616 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1617 (memopv2i64 addr:$src2)))])>, TB;
1619 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1620 !strconcat(OpcodeStr, "pd"), f128mem,
1621 !if(HasPat, Pattern[1], // rr
1622 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1625 !if(HasPat, Pattern[3], // rm
1626 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1627 (memopv2i64 addr:$src2)))])>,
1632 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1634 let isAsmParserOnly = 1 in {
1635 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1636 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1637 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1639 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1640 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1644 // AVX 256-bit packed logical ops forms
1645 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1646 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1647 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1648 let isCommutable = 0 in
1649 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1651 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1652 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1653 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1654 let isCommutable = 0 in
1655 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1657 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1658 (bc_v2i64 (v4i32 immAllOnesV))),
1661 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1662 (bc_v2i64 (v2f64 VR128:$src2))))],
1664 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1665 (bc_v2i64 (v4i32 immAllOnesV))),
1666 (memopv2i64 addr:$src2))))],
1668 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1669 (memopv2i64 addr:$src2)))]]>;
1671 //===----------------------------------------------------------------------===//
1672 // SSE 1 & 2 - Arithmetic Instructions
1673 //===----------------------------------------------------------------------===//
1675 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1678 /// In addition, we also have a special variant of the scalar form here to
1679 /// represent the associated intrinsic operation. This form is unlike the
1680 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1681 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1683 /// These three forms can each be reg+reg or reg+mem.
1686 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1688 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1690 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1691 OpNode, FR32, f32mem, Is2Addr>, XS;
1692 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1693 OpNode, FR64, f64mem, Is2Addr>, XD;
1696 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1698 let mayLoad = 0 in {
1699 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1700 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1701 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1702 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1706 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1708 let mayLoad = 0 in {
1709 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1710 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1711 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1712 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1716 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1718 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1719 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1720 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1721 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1724 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1726 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1727 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1728 SSEPackedSingle, Is2Addr>, TB;
1730 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1731 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1732 SSEPackedDouble, Is2Addr>, TB, OpSize;
1735 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1736 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1737 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1738 SSEPackedSingle, 0>, TB;
1740 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1741 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1742 SSEPackedDouble, 0>, TB, OpSize;
1745 // Binary Arithmetic instructions
1746 let isAsmParserOnly = 1 in {
1747 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1748 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1749 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1750 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1751 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1752 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1753 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1754 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1756 let isCommutable = 0 in {
1757 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1758 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1759 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1760 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1761 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1762 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1763 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1764 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1765 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1766 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1767 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1768 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1769 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1770 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1771 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1772 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1773 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1774 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1775 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1776 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1780 let Constraints = "$src1 = $dst" in {
1781 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1782 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1783 basic_sse12_fp_binop_s_int<0x58, "add">;
1784 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1785 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1786 basic_sse12_fp_binop_s_int<0x59, "mul">;
1788 let isCommutable = 0 in {
1789 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1790 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1791 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1792 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1793 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1794 basic_sse12_fp_binop_s_int<0x5E, "div">;
1795 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1796 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1797 basic_sse12_fp_binop_s_int<0x5F, "max">,
1798 basic_sse12_fp_binop_p_int<0x5F, "max">;
1799 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1800 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1801 basic_sse12_fp_binop_s_int<0x5D, "min">,
1802 basic_sse12_fp_binop_p_int<0x5D, "min">;
1807 /// In addition, we also have a special variant of the scalar form here to
1808 /// represent the associated intrinsic operation. This form is unlike the
1809 /// plain scalar form, in that it takes an entire vector (instead of a
1810 /// scalar) and leaves the top elements undefined.
1812 /// And, we have a special variant form for a full-vector intrinsic form.
1814 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1815 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1816 SDNode OpNode, Intrinsic F32Int> {
1817 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1818 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1819 [(set FR32:$dst, (OpNode FR32:$src))]>;
1820 // For scalar unary operations, fold a load into the operation
1821 // only in OptForSize mode. It eliminates an instruction, but it also
1822 // eliminates a whole-register clobber (the load), so it introduces a
1823 // partial register update condition.
1824 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1825 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1826 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1827 Requires<[HasSSE1, OptForSize]>;
1828 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1829 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1830 [(set VR128:$dst, (F32Int VR128:$src))]>;
1831 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1832 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1833 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1836 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1837 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1838 SDNode OpNode, Intrinsic F32Int> {
1839 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1840 !strconcat(OpcodeStr,
1841 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1842 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1843 !strconcat(OpcodeStr,
1844 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1845 []>, XS, Requires<[HasAVX, OptForSize]>;
1846 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1847 !strconcat(OpcodeStr,
1848 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1849 [(set VR128:$dst, (F32Int VR128:$src))]>;
1850 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1851 !strconcat(OpcodeStr,
1852 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1853 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1856 /// sse1_fp_unop_p - SSE1 unops in packed form.
1857 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1858 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1859 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1860 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1861 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1862 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1863 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1866 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1867 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1868 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1869 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1870 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1871 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1872 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1873 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1876 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1877 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1878 Intrinsic V4F32Int> {
1879 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1880 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1881 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1882 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1883 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1884 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1887 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1888 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1889 Intrinsic V4F32Int> {
1890 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1891 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1892 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1893 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1894 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1895 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1898 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1899 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1900 SDNode OpNode, Intrinsic F64Int> {
1901 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1902 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1903 [(set FR64:$dst, (OpNode FR64:$src))]>;
1904 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1905 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1906 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1907 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1908 Requires<[HasSSE2, OptForSize]>;
1909 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1910 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1911 [(set VR128:$dst, (F64Int VR128:$src))]>;
1912 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1913 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1914 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1917 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1918 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1919 SDNode OpNode, Intrinsic F64Int> {
1920 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1921 !strconcat(OpcodeStr,
1922 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1923 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1924 (ins FR64:$src1, f64mem:$src2),
1925 !strconcat(OpcodeStr,
1926 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1927 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1928 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1929 [(set VR128:$dst, (F64Int VR128:$src))]>;
1930 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1931 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1932 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1935 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1936 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1938 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1939 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1940 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1941 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1942 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1943 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1946 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1947 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1948 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1949 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1950 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1951 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1952 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1953 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1956 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1957 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1958 Intrinsic V2F64Int> {
1959 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1960 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1961 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1962 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1963 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1964 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1967 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1968 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1969 Intrinsic V2F64Int> {
1970 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1971 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1972 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1973 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1974 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1975 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1978 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1980 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1981 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1984 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1985 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1986 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1987 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1988 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1989 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1990 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1991 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1994 // Reciprocal approximations. Note that these typically require refinement
1995 // in order to obtain suitable precision.
1996 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1997 int_x86_sse_rsqrt_ss>, VEX_4V;
1998 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1999 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2000 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2001 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2003 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
2005 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2006 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2007 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2008 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2012 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2013 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2014 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2015 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2016 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2017 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2019 // Reciprocal approximations. Note that these typically require refinement
2020 // in order to obtain suitable precision.
2021 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2022 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2023 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2024 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2025 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2026 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2028 // There is no f64 version of the reciprocal approximation instructions.
2030 //===----------------------------------------------------------------------===//
2031 // SSE 1 & 2 - Non-temporal stores
2032 //===----------------------------------------------------------------------===//
2034 let isAsmParserOnly = 1 in {
2035 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
2036 (ins i128mem:$dst, VR128:$src),
2037 "movntps\t{$src, $dst|$dst, $src}",
2038 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
2039 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
2040 (ins i128mem:$dst, VR128:$src),
2041 "movntpd\t{$src, $dst|$dst, $src}",
2042 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
2044 let ExeDomain = SSEPackedInt in
2045 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
2046 (ins f128mem:$dst, VR128:$src),
2047 "movntdq\t{$src, $dst|$dst, $src}",
2048 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
2050 let AddedComplexity = 400 in { // Prefer non-temporal versions
2051 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2052 (ins f128mem:$dst, VR128:$src),
2053 "movntps\t{$src, $dst|$dst, $src}",
2054 [(alignednontemporalstore (v4f32 VR128:$src),
2056 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2057 (ins f128mem:$dst, VR128:$src),
2058 "movntpd\t{$src, $dst|$dst, $src}",
2059 [(alignednontemporalstore (v2f64 VR128:$src),
2061 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2062 (ins f128mem:$dst, VR128:$src),
2063 "movntdq\t{$src, $dst|$dst, $src}",
2064 [(alignednontemporalstore (v2f64 VR128:$src),
2066 let ExeDomain = SSEPackedInt in
2067 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2068 (ins f128mem:$dst, VR128:$src),
2069 "movntdq\t{$src, $dst|$dst, $src}",
2070 [(alignednontemporalstore (v4f32 VR128:$src),
2073 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2074 (ins f256mem:$dst, VR256:$src),
2075 "movntps\t{$src, $dst|$dst, $src}",
2076 [(alignednontemporalstore (v8f32 VR256:$src),
2078 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2079 (ins f256mem:$dst, VR256:$src),
2080 "movntpd\t{$src, $dst|$dst, $src}",
2081 [(alignednontemporalstore (v4f64 VR256:$src),
2083 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2084 (ins f256mem:$dst, VR256:$src),
2085 "movntdq\t{$src, $dst|$dst, $src}",
2086 [(alignednontemporalstore (v4f64 VR256:$src),
2088 let ExeDomain = SSEPackedInt in
2089 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2090 (ins f256mem:$dst, VR256:$src),
2091 "movntdq\t{$src, $dst|$dst, $src}",
2092 [(alignednontemporalstore (v8f32 VR256:$src),
2097 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2098 "movntps\t{$src, $dst|$dst, $src}",
2099 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2100 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2101 "movntpd\t{$src, $dst|$dst, $src}",
2102 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2104 let ExeDomain = SSEPackedInt in
2105 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2106 "movntdq\t{$src, $dst|$dst, $src}",
2107 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2109 let AddedComplexity = 400 in { // Prefer non-temporal versions
2110 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2111 "movntps\t{$src, $dst|$dst, $src}",
2112 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2113 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2114 "movntpd\t{$src, $dst|$dst, $src}",
2115 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2117 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2118 "movntdq\t{$src, $dst|$dst, $src}",
2119 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2121 let ExeDomain = SSEPackedInt in
2122 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2123 "movntdq\t{$src, $dst|$dst, $src}",
2124 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2126 // There is no AVX form for instructions below this point
2127 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2128 "movnti\t{$src, $dst|$dst, $src}",
2129 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2130 TB, Requires<[HasSSE2]>;
2132 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2133 "movnti\t{$src, $dst|$dst, $src}",
2134 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2135 TB, Requires<[HasSSE2]>;
2138 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2139 "movnti\t{$src, $dst|$dst, $src}",
2140 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2141 TB, Requires<[HasSSE2]>;
2143 //===----------------------------------------------------------------------===//
2144 // SSE 1 & 2 - Misc Instructions (No AVX form)
2145 //===----------------------------------------------------------------------===//
2147 // Prefetch intrinsic.
2148 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2149 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2150 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2151 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2152 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2153 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2154 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2155 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2157 // Load, store, and memory fence
2158 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2159 TB, Requires<[HasSSE1]>;
2160 def : Pat<(X86SFence), (SFENCE)>;
2162 // Alias instructions that map zero vector to pxor / xorp* for sse.
2163 // We set canFoldAsLoad because this can be converted to a constant-pool
2164 // load of an all-zeros value if folding it would be beneficial.
2165 // FIXME: Change encoding to pseudo!
2166 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2167 isCodeGenOnly = 1 in {
2168 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2169 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2170 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2171 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2172 let ExeDomain = SSEPackedInt in
2173 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2174 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2177 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2178 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2179 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2181 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2182 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2184 //===----------------------------------------------------------------------===//
2185 // SSE 1 & 2 - Load/Store XCSR register
2186 //===----------------------------------------------------------------------===//
2188 let isAsmParserOnly = 1 in {
2189 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2190 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2191 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2192 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2195 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2196 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2197 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2198 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2200 //===---------------------------------------------------------------------===//
2201 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2202 //===---------------------------------------------------------------------===//
2204 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2206 let isAsmParserOnly = 1 in {
2207 let neverHasSideEffects = 1 in {
2208 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2209 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2210 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2211 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2213 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2214 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2215 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2216 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2218 let canFoldAsLoad = 1, mayLoad = 1 in {
2219 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2220 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2221 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2222 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2223 let Predicates = [HasAVX] in {
2224 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2225 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2226 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2227 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2231 let mayStore = 1 in {
2232 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2233 (ins i128mem:$dst, VR128:$src),
2234 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2235 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2236 (ins i256mem:$dst, VR256:$src),
2237 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2238 let Predicates = [HasAVX] in {
2239 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2240 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2241 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2242 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2247 let neverHasSideEffects = 1 in
2248 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2249 "movdqa\t{$src, $dst|$dst, $src}", []>;
2251 let canFoldAsLoad = 1, mayLoad = 1 in {
2252 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2253 "movdqa\t{$src, $dst|$dst, $src}",
2254 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2255 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2256 "movdqu\t{$src, $dst|$dst, $src}",
2257 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2258 XS, Requires<[HasSSE2]>;
2261 let mayStore = 1 in {
2262 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2263 "movdqa\t{$src, $dst|$dst, $src}",
2264 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2265 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2266 "movdqu\t{$src, $dst|$dst, $src}",
2267 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2268 XS, Requires<[HasSSE2]>;
2271 // Intrinsic forms of MOVDQU load and store
2272 let isAsmParserOnly = 1 in {
2273 let canFoldAsLoad = 1 in
2274 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2275 "vmovdqu\t{$src, $dst|$dst, $src}",
2276 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2277 XS, VEX, Requires<[HasAVX]>;
2278 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2279 "vmovdqu\t{$src, $dst|$dst, $src}",
2280 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2281 XS, VEX, Requires<[HasAVX]>;
2284 let canFoldAsLoad = 1 in
2285 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2286 "movdqu\t{$src, $dst|$dst, $src}",
2287 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2288 XS, Requires<[HasSSE2]>;
2289 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2290 "movdqu\t{$src, $dst|$dst, $src}",
2291 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2292 XS, Requires<[HasSSE2]>;
2294 } // ExeDomain = SSEPackedInt
2296 //===---------------------------------------------------------------------===//
2297 // SSE2 - Packed Integer Arithmetic Instructions
2298 //===---------------------------------------------------------------------===//
2300 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2302 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2303 bit IsCommutable = 0, bit Is2Addr = 1> {
2304 let isCommutable = IsCommutable in
2305 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2306 (ins VR128:$src1, VR128:$src2),
2308 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2309 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2310 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2311 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2312 (ins VR128:$src1, i128mem:$src2),
2314 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2316 [(set VR128:$dst, (IntId VR128:$src1,
2317 (bitconvert (memopv2i64 addr:$src2))))]>;
2320 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2321 string OpcodeStr, Intrinsic IntId,
2322 Intrinsic IntId2, bit Is2Addr = 1> {
2323 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2324 (ins VR128:$src1, VR128:$src2),
2326 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2327 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2328 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2329 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2330 (ins VR128:$src1, i128mem:$src2),
2332 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2333 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2334 [(set VR128:$dst, (IntId VR128:$src1,
2335 (bitconvert (memopv2i64 addr:$src2))))]>;
2336 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2337 (ins VR128:$src1, i32i8imm:$src2),
2339 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2341 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2344 /// PDI_binop_rm - Simple SSE2 binary operator.
2345 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2346 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2347 let isCommutable = IsCommutable in
2348 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2349 (ins VR128:$src1, VR128:$src2),
2351 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2352 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2353 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2354 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2355 (ins VR128:$src1, i128mem:$src2),
2357 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2359 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2360 (bitconvert (memopv2i64 addr:$src2)))))]>;
2363 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2365 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2366 /// to collapse (bitconvert VT to VT) into its operand.
2368 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2369 bit IsCommutable = 0, bit Is2Addr = 1> {
2370 let isCommutable = IsCommutable in
2371 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2372 (ins VR128:$src1, VR128:$src2),
2374 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2375 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2376 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2377 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2378 (ins VR128:$src1, i128mem:$src2),
2380 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2381 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2382 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2385 } // ExeDomain = SSEPackedInt
2387 // 128-bit Integer Arithmetic
2389 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2390 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2391 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2392 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2393 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2394 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2395 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2396 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2397 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2398 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2401 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2403 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2405 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2407 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2409 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2411 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2413 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2415 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2417 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2419 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2421 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2423 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2425 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2427 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2429 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2431 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2433 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2435 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2437 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2441 let Constraints = "$src1 = $dst" in {
2442 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2443 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2444 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2445 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2446 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2447 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2448 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2449 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2450 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2453 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2454 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2455 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2456 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2457 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2458 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2459 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2460 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2461 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2462 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2463 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2464 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2465 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2466 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2467 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2468 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2469 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2470 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2471 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2473 } // Constraints = "$src1 = $dst"
2475 //===---------------------------------------------------------------------===//
2476 // SSE2 - Packed Integer Logical Instructions
2477 //===---------------------------------------------------------------------===//
2479 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2480 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2481 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2483 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2484 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2486 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2487 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2490 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2491 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2493 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2494 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2496 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2497 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2500 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2501 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2503 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2504 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2507 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2508 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2509 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2511 let ExeDomain = SSEPackedInt in {
2512 let neverHasSideEffects = 1 in {
2513 // 128-bit logical shifts.
2514 def VPSLLDQri : PDIi8<0x73, MRM7r,
2515 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2516 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2518 def VPSRLDQri : PDIi8<0x73, MRM3r,
2519 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2520 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2522 // PSRADQri doesn't exist in SSE[1-3].
2524 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2525 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2526 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2527 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2528 VR128:$src2)))]>, VEX_4V;
2530 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2531 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2532 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2533 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2534 (memopv2i64 addr:$src2))))]>,
2539 let Constraints = "$src1 = $dst" in {
2540 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2541 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2542 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2543 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2544 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2545 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2547 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2548 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2549 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2550 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2551 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2552 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2554 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2555 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2556 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2557 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2559 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2560 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2561 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2563 let ExeDomain = SSEPackedInt in {
2564 let neverHasSideEffects = 1 in {
2565 // 128-bit logical shifts.
2566 def PSLLDQri : PDIi8<0x73, MRM7r,
2567 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2568 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2569 def PSRLDQri : PDIi8<0x73, MRM3r,
2570 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2571 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2572 // PSRADQri doesn't exist in SSE[1-3].
2574 def PANDNrr : PDI<0xDF, MRMSrcReg,
2575 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2576 "pandn\t{$src2, $dst|$dst, $src2}",
2577 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2580 def PANDNrm : PDI<0xDF, MRMSrcMem,
2581 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2582 "pandn\t{$src2, $dst|$dst, $src2}",
2583 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2584 (memopv2i64 addr:$src2))))]>;
2586 } // Constraints = "$src1 = $dst"
2588 let Predicates = [HasAVX] in {
2589 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2590 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2591 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2592 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2593 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2594 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2595 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2596 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2597 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2598 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2600 // Shift up / down and insert zero's.
2601 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2602 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2603 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2604 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2607 let Predicates = [HasSSE2] in {
2608 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2609 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2610 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2611 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2612 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2613 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2614 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2615 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2616 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2617 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2619 // Shift up / down and insert zero's.
2620 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2621 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2622 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2623 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2626 //===---------------------------------------------------------------------===//
2627 // SSE2 - Packed Integer Comparison Instructions
2628 //===---------------------------------------------------------------------===//
2630 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2631 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2633 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2635 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2637 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2639 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2641 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2645 let Constraints = "$src1 = $dst" in {
2646 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2647 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2648 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2649 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2650 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2651 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2652 } // Constraints = "$src1 = $dst"
2654 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2655 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2656 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2657 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2658 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2659 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2660 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2661 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2662 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2663 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2664 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2665 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2667 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2668 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2669 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2670 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2671 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2672 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2673 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2674 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2675 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2676 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2677 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2678 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2680 //===---------------------------------------------------------------------===//
2681 // SSE2 - Packed Integer Pack Instructions
2682 //===---------------------------------------------------------------------===//
2684 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2685 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2687 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2689 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2693 let Constraints = "$src1 = $dst" in {
2694 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2695 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2696 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2697 } // Constraints = "$src1 = $dst"
2699 //===---------------------------------------------------------------------===//
2700 // SSE2 - Packed Integer Shuffle Instructions
2701 //===---------------------------------------------------------------------===//
2703 let ExeDomain = SSEPackedInt in {
2704 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2706 def ri : Ii8<0x70, MRMSrcReg,
2707 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2708 !strconcat(OpcodeStr,
2709 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2710 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2712 def mi : Ii8<0x70, MRMSrcMem,
2713 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2714 !strconcat(OpcodeStr,
2715 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2716 [(set VR128:$dst, (vt (pshuf_frag:$src2
2717 (bc_frag (memopv2i64 addr:$src1)),
2720 } // ExeDomain = SSEPackedInt
2722 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2723 let AddedComplexity = 5 in
2724 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2727 // SSE2 with ImmT == Imm8 and XS prefix.
2728 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2731 // SSE2 with ImmT == Imm8 and XD prefix.
2732 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2736 let Predicates = [HasSSE2] in {
2737 let AddedComplexity = 5 in
2738 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2740 // SSE2 with ImmT == Imm8 and XS prefix.
2741 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2743 // SSE2 with ImmT == Imm8 and XD prefix.
2744 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2747 //===---------------------------------------------------------------------===//
2748 // SSE2 - Packed Integer Unpack Instructions
2749 //===---------------------------------------------------------------------===//
2751 let ExeDomain = SSEPackedInt in {
2752 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2753 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2754 def rr : PDI<opc, MRMSrcReg,
2755 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2757 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2758 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2759 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2760 def rm : PDI<opc, MRMSrcMem,
2761 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2763 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2764 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2765 [(set VR128:$dst, (unp_frag VR128:$src1,
2766 (bc_frag (memopv2i64
2770 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2771 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2773 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2775 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2778 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2779 /// knew to collapse (bitconvert VT to VT) into its operand.
2780 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2781 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2782 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2784 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2785 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2786 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2787 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2789 (v2i64 (unpckl VR128:$src1,
2790 (memopv2i64 addr:$src2))))]>, VEX_4V;
2792 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2794 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2796 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2799 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2800 /// knew to collapse (bitconvert VT to VT) into its operand.
2801 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2802 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2803 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2805 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2806 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2807 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2808 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2810 (v2i64 (unpckh VR128:$src1,
2811 (memopv2i64 addr:$src2))))]>, VEX_4V;
2814 let Constraints = "$src1 = $dst" in {
2815 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2816 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2817 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2819 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2820 /// knew to collapse (bitconvert VT to VT) into its operand.
2821 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2822 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2823 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2825 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2826 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2827 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2828 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2830 (v2i64 (unpckl VR128:$src1,
2831 (memopv2i64 addr:$src2))))]>;
2833 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2834 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2835 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2837 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2838 /// knew to collapse (bitconvert VT to VT) into its operand.
2839 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2840 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2841 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2843 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2844 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2845 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2846 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2848 (v2i64 (unpckh VR128:$src1,
2849 (memopv2i64 addr:$src2))))]>;
2852 } // ExeDomain = SSEPackedInt
2854 //===---------------------------------------------------------------------===//
2855 // SSE2 - Packed Integer Extract and Insert
2856 //===---------------------------------------------------------------------===//
2858 let ExeDomain = SSEPackedInt in {
2859 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2860 def rri : Ii8<0xC4, MRMSrcReg,
2861 (outs VR128:$dst), (ins VR128:$src1,
2862 GR32:$src2, i32i8imm:$src3),
2864 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2865 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2867 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2868 def rmi : Ii8<0xC4, MRMSrcMem,
2869 (outs VR128:$dst), (ins VR128:$src1,
2870 i16mem:$src2, i32i8imm:$src3),
2872 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2873 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2875 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2880 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2881 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2882 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2883 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2884 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2885 imm:$src2))]>, OpSize, VEX;
2886 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2887 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2888 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2889 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2893 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2894 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2895 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2896 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2897 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2898 []>, OpSize, VEX_4V;
2901 let Constraints = "$src1 = $dst" in
2902 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2904 } // ExeDomain = SSEPackedInt
2906 //===---------------------------------------------------------------------===//
2907 // SSE2 - Packed Mask Creation
2908 //===---------------------------------------------------------------------===//
2910 let ExeDomain = SSEPackedInt in {
2912 let isAsmParserOnly = 1 in {
2913 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2914 "pmovmskb\t{$src, $dst|$dst, $src}",
2915 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2916 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2917 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2919 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2920 "pmovmskb\t{$src, $dst|$dst, $src}",
2921 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2923 } // ExeDomain = SSEPackedInt
2925 //===---------------------------------------------------------------------===//
2926 // SSE2 - Conditional Store
2927 //===---------------------------------------------------------------------===//
2929 let ExeDomain = SSEPackedInt in {
2931 let isAsmParserOnly = 1 in {
2933 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2934 (ins VR128:$src, VR128:$mask),
2935 "maskmovdqu\t{$mask, $src|$src, $mask}",
2936 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2938 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2939 (ins VR128:$src, VR128:$mask),
2940 "maskmovdqu\t{$mask, $src|$src, $mask}",
2941 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2945 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2946 "maskmovdqu\t{$mask, $src|$src, $mask}",
2947 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2949 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2950 "maskmovdqu\t{$mask, $src|$src, $mask}",
2951 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2953 } // ExeDomain = SSEPackedInt
2955 //===---------------------------------------------------------------------===//
2956 // SSE2 - Move Doubleword
2957 //===---------------------------------------------------------------------===//
2959 // Move Int Doubleword to Packed Double Int
2960 let isAsmParserOnly = 1 in {
2961 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2962 "movd\t{$src, $dst|$dst, $src}",
2964 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2965 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2966 "movd\t{$src, $dst|$dst, $src}",
2968 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2971 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2972 "movd\t{$src, $dst|$dst, $src}",
2974 (v4i32 (scalar_to_vector GR32:$src)))]>;
2975 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2976 "movd\t{$src, $dst|$dst, $src}",
2978 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2981 // Move Int Doubleword to Single Scalar
2982 let isAsmParserOnly = 1 in {
2983 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2984 "movd\t{$src, $dst|$dst, $src}",
2985 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2987 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2988 "movd\t{$src, $dst|$dst, $src}",
2989 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2992 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2993 "movd\t{$src, $dst|$dst, $src}",
2994 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2996 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2997 "movd\t{$src, $dst|$dst, $src}",
2998 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
3000 // Move Packed Doubleword Int to Packed Double Int
3001 let isAsmParserOnly = 1 in {
3002 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3003 "movd\t{$src, $dst|$dst, $src}",
3004 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3006 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
3007 (ins i32mem:$dst, VR128:$src),
3008 "movd\t{$src, $dst|$dst, $src}",
3009 [(store (i32 (vector_extract (v4i32 VR128:$src),
3010 (iPTR 0))), addr:$dst)]>, VEX;
3012 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3013 "movd\t{$src, $dst|$dst, $src}",
3014 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3016 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3017 "movd\t{$src, $dst|$dst, $src}",
3018 [(store (i32 (vector_extract (v4i32 VR128:$src),
3019 (iPTR 0))), addr:$dst)]>;
3021 // Move Scalar Single to Double Int
3022 let isAsmParserOnly = 1 in {
3023 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3024 "movd\t{$src, $dst|$dst, $src}",
3025 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3026 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3027 "movd\t{$src, $dst|$dst, $src}",
3028 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3030 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3031 "movd\t{$src, $dst|$dst, $src}",
3032 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3033 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3034 "movd\t{$src, $dst|$dst, $src}",
3035 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3037 // movd / movq to XMM register zero-extends
3038 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3039 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3040 "movd\t{$src, $dst|$dst, $src}",
3041 [(set VR128:$dst, (v4i32 (X86vzmovl
3042 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3044 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3045 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3046 [(set VR128:$dst, (v2i64 (X86vzmovl
3047 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3050 let AddedComplexity = 15 in {
3051 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3052 "movd\t{$src, $dst|$dst, $src}",
3053 [(set VR128:$dst, (v4i32 (X86vzmovl
3054 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3055 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3056 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3057 [(set VR128:$dst, (v2i64 (X86vzmovl
3058 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3061 let AddedComplexity = 20 in {
3062 let isAsmParserOnly = 1 in
3063 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3064 "movd\t{$src, $dst|$dst, $src}",
3066 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3067 (loadi32 addr:$src))))))]>,
3069 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3070 "movd\t{$src, $dst|$dst, $src}",
3072 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3073 (loadi32 addr:$src))))))]>;
3075 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3076 (MOVZDI2PDIrm addr:$src)>;
3077 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3078 (MOVZDI2PDIrm addr:$src)>;
3079 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3080 (MOVZDI2PDIrm addr:$src)>;
3083 //===---------------------------------------------------------------------===//
3084 // SSE2 - Move Quadword
3085 //===---------------------------------------------------------------------===//
3087 // Move Quadword Int to Packed Quadword Int
3088 let isAsmParserOnly = 1 in
3089 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3090 "vmovq\t{$src, $dst|$dst, $src}",
3092 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3093 VEX, Requires<[HasAVX]>;
3094 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3095 "movq\t{$src, $dst|$dst, $src}",
3097 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3098 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3100 // Move Packed Quadword Int to Quadword Int
3101 let isAsmParserOnly = 1 in
3102 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3103 "movq\t{$src, $dst|$dst, $src}",
3104 [(store (i64 (vector_extract (v2i64 VR128:$src),
3105 (iPTR 0))), addr:$dst)]>, VEX;
3106 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3107 "movq\t{$src, $dst|$dst, $src}",
3108 [(store (i64 (vector_extract (v2i64 VR128:$src),
3109 (iPTR 0))), addr:$dst)]>;
3111 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3112 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3114 // Store / copy lower 64-bits of a XMM register.
3115 let isAsmParserOnly = 1 in
3116 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3117 "movq\t{$src, $dst|$dst, $src}",
3118 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3119 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3120 "movq\t{$src, $dst|$dst, $src}",
3121 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3123 let AddedComplexity = 20, isAsmParserOnly = 1 in
3124 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3125 "vmovq\t{$src, $dst|$dst, $src}",
3127 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3128 (loadi64 addr:$src))))))]>,
3129 XS, VEX, Requires<[HasAVX]>;
3131 let AddedComplexity = 20 in {
3132 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3133 "movq\t{$src, $dst|$dst, $src}",
3135 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3136 (loadi64 addr:$src))))))]>,
3137 XS, Requires<[HasSSE2]>;
3139 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3140 (MOVZQI2PQIrm addr:$src)>;
3141 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3142 (MOVZQI2PQIrm addr:$src)>;
3143 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3146 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3147 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3148 let isAsmParserOnly = 1, AddedComplexity = 15 in
3149 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3150 "vmovq\t{$src, $dst|$dst, $src}",
3151 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3152 XS, VEX, Requires<[HasAVX]>;
3153 let AddedComplexity = 15 in
3154 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3155 "movq\t{$src, $dst|$dst, $src}",
3156 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3157 XS, Requires<[HasSSE2]>;
3159 let AddedComplexity = 20, isAsmParserOnly = 1 in
3160 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3161 "vmovq\t{$src, $dst|$dst, $src}",
3162 [(set VR128:$dst, (v2i64 (X86vzmovl
3163 (loadv2i64 addr:$src))))]>,
3164 XS, VEX, Requires<[HasAVX]>;
3165 let AddedComplexity = 20 in {
3166 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3167 "movq\t{$src, $dst|$dst, $src}",
3168 [(set VR128:$dst, (v2i64 (X86vzmovl
3169 (loadv2i64 addr:$src))))]>,
3170 XS, Requires<[HasSSE2]>;
3172 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3173 (MOVZPQILo2PQIrm addr:$src)>;
3176 // Instructions to match in the assembler
3177 let isAsmParserOnly = 1 in {
3178 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3179 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3180 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3181 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3182 // Recognize "movd" with GR64 destination, but encode as a "movq"
3183 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3184 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3187 // Instructions for the disassembler
3188 // xr = XMM register
3191 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3192 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3193 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3194 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3195 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3197 //===---------------------------------------------------------------------===//
3198 // SSE2 - Misc Instructions
3199 //===---------------------------------------------------------------------===//
3202 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3203 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3204 TB, Requires<[HasSSE2]>;
3206 // Load, store, and memory fence
3207 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3208 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3209 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3210 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3211 def : Pat<(X86LFence), (LFENCE)>;
3212 def : Pat<(X86MFence), (MFENCE)>;
3215 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3216 // was introduced with SSE2, it's backward compatible.
3217 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3219 // Alias instructions that map zero vector to pxor / xorp* for sse.
3220 // We set canFoldAsLoad because this can be converted to a constant-pool
3221 // load of an all-ones value if folding it would be beneficial.
3222 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3223 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3224 // FIXME: Change encoding to pseudo.
3225 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3226 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3228 //===---------------------------------------------------------------------===//
3229 // SSE3 - Conversion Instructions
3230 //===---------------------------------------------------------------------===//
3232 // Convert Packed Double FP to Packed DW Integers
3233 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3234 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3235 // register, but the same isn't true when using memory operands instead.
3236 // Provide other assembly rr and rm forms to address this explicitly.
3237 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3238 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3239 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3240 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3243 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3244 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3245 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3246 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3249 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3250 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3251 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3252 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3255 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3256 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3257 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3258 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3260 // Convert Packed DW Integers to Packed Double FP
3261 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3262 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3263 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3264 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3265 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3266 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3267 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3268 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3269 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3272 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3273 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3274 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3275 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3277 // AVX 256-bit register conversion intrinsics
3278 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3279 (VCVTDQ2PDYrr VR128:$src)>;
3280 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3281 (VCVTDQ2PDYrm addr:$src)>;
3283 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3284 (VCVTPD2DQYrr VR256:$src)>;
3285 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3286 (VCVTPD2DQYrm addr:$src)>;
3288 //===---------------------------------------------------------------------===//
3289 // SSE3 - Move Instructions
3290 //===---------------------------------------------------------------------===//
3292 // Replicate Single FP
3293 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3294 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3295 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3296 [(set VR128:$dst, (v4f32 (rep_frag
3297 VR128:$src, (undef))))]>;
3298 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3299 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3300 [(set VR128:$dst, (rep_frag
3301 (memopv4f32 addr:$src), (undef)))]>;
3304 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3306 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3307 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3308 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3309 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3312 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3313 // FIXME: Merge above classes when we have patterns for the ymm version
3314 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3315 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3316 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3317 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3319 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3320 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3322 // Replicate Double FP
3323 multiclass sse3_replicate_dfp<string OpcodeStr> {
3324 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3325 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3326 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3327 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3328 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3330 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3334 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3335 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3338 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3339 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3343 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3344 // FIXME: Merge above classes when we have patterns for the ymm version
3345 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3346 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3348 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3350 // Move Unaligned Integer
3351 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3352 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3353 "vlddqu\t{$src, $dst|$dst, $src}",
3354 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3355 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3356 "vlddqu\t{$src, $dst|$dst, $src}", []>, VEX;
3358 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3359 "lddqu\t{$src, $dst|$dst, $src}",
3360 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3362 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3364 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3366 // Several Move patterns
3367 let AddedComplexity = 5 in {
3368 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3369 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3370 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3371 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3372 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3373 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3374 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3375 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3378 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3379 let AddedComplexity = 15 in
3380 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3381 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3382 let AddedComplexity = 20 in
3383 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3384 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3386 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3387 let AddedComplexity = 15 in
3388 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3389 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3390 let AddedComplexity = 20 in
3391 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3392 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3394 //===---------------------------------------------------------------------===//
3395 // SSE3 - Arithmetic
3396 //===---------------------------------------------------------------------===//
3398 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3399 X86MemOperand x86memop, bit Is2Addr = 1> {
3400 def rr : I<0xD0, MRMSrcReg,
3401 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3403 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3404 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3405 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3406 def rm : I<0xD0, MRMSrcMem,
3407 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3409 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3410 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3411 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3414 let isAsmParserOnly = 1, Predicates = [HasAVX],
3415 ExeDomain = SSEPackedDouble in {
3416 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3417 f128mem, 0>, XD, VEX_4V;
3418 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3419 f128mem, 0>, OpSize, VEX_4V;
3420 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3421 f256mem, 0>, XD, VEX_4V;
3422 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3423 f256mem, 0>, OpSize, VEX_4V;
3425 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3426 ExeDomain = SSEPackedDouble in {
3427 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3429 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3430 f128mem>, TB, OpSize;
3433 //===---------------------------------------------------------------------===//
3434 // SSE3 Instructions
3435 //===---------------------------------------------------------------------===//
3438 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3439 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3440 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3444 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3446 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3448 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3449 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3450 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3452 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3453 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3454 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3456 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3458 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3460 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3462 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3463 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3464 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3467 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3468 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3469 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3470 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3471 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3472 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3473 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3474 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3475 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3476 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3477 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3478 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3479 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3480 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3481 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3482 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3483 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3486 let Constraints = "$src1 = $dst" in {
3487 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3488 int_x86_sse3_hadd_ps>;
3489 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3490 int_x86_sse3_hadd_pd>;
3491 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3492 int_x86_sse3_hsub_ps>;
3493 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3494 int_x86_sse3_hsub_pd>;
3497 //===---------------------------------------------------------------------===//
3498 // SSSE3 - Packed Absolute Instructions
3499 //===---------------------------------------------------------------------===//
3501 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3502 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3503 PatFrag mem_frag64, PatFrag mem_frag128,
3504 Intrinsic IntId64, Intrinsic IntId128> {
3505 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
3506 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3507 [(set VR64:$dst, (IntId64 VR64:$src))]>;
3509 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
3510 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3512 (IntId64 (bitconvert (mem_frag64 addr:$src))))]>;
3514 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3516 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3517 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3520 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3522 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3525 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3528 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3529 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv8i8, memopv16i8,
3530 int_x86_ssse3_pabs_b,
3531 int_x86_ssse3_pabs_b_128>, VEX;
3532 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv4i16, memopv8i16,
3533 int_x86_ssse3_pabs_w,
3534 int_x86_ssse3_pabs_w_128>, VEX;
3535 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv2i32, memopv4i32,
3536 int_x86_ssse3_pabs_d,
3537 int_x86_ssse3_pabs_d_128>, VEX;
3540 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv8i8, memopv16i8,
3541 int_x86_ssse3_pabs_b,
3542 int_x86_ssse3_pabs_b_128>;
3543 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv4i16, memopv8i16,
3544 int_x86_ssse3_pabs_w,
3545 int_x86_ssse3_pabs_w_128>;
3546 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv2i32, memopv4i32,
3547 int_x86_ssse3_pabs_d,
3548 int_x86_ssse3_pabs_d_128>;
3550 //===---------------------------------------------------------------------===//
3551 // SSSE3 - Packed Binary Operator Instructions
3552 //===---------------------------------------------------------------------===//
3554 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3555 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3556 PatFrag mem_frag64, PatFrag mem_frag128,
3557 Intrinsic IntId64, Intrinsic IntId128,
3559 let isCommutable = 1 in
3560 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
3561 (ins VR64:$src1, VR64:$src2),
3563 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3564 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3565 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
3566 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
3567 (ins VR64:$src1, i64mem:$src2),
3569 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3570 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3572 (IntId64 VR64:$src1,
3573 (bitconvert (memopv8i8 addr:$src2))))]>;
3575 let isCommutable = 1 in
3576 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3577 (ins VR128:$src1, VR128:$src2),
3579 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3580 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3581 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3583 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3584 (ins VR128:$src1, i128mem:$src2),
3586 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3587 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3589 (IntId128 VR128:$src1,
3590 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3593 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3594 let isCommutable = 0 in {
3595 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv4i16, memopv8i16,
3596 int_x86_ssse3_phadd_w,
3597 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3598 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv2i32, memopv4i32,
3599 int_x86_ssse3_phadd_d,
3600 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3601 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv4i16, memopv8i16,
3602 int_x86_ssse3_phadd_sw,
3603 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3604 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv4i16, memopv8i16,
3605 int_x86_ssse3_phsub_w,
3606 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3607 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv2i32, memopv4i32,
3608 int_x86_ssse3_phsub_d,
3609 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3610 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv4i16, memopv8i16,
3611 int_x86_ssse3_phsub_sw,
3612 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3613 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv8i8, memopv16i8,
3614 int_x86_ssse3_pmadd_ub_sw,
3615 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3616 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv8i8, memopv16i8,
3617 int_x86_ssse3_pshuf_b,
3618 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3619 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv8i8, memopv16i8,
3620 int_x86_ssse3_psign_b,
3621 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3622 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv4i16, memopv8i16,
3623 int_x86_ssse3_psign_w,
3624 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3625 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv2i32, memopv4i32,
3626 int_x86_ssse3_psign_d,
3627 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3629 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv4i16, memopv8i16,
3630 int_x86_ssse3_pmul_hr_sw,
3631 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3634 // None of these have i8 immediate fields.
3635 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3636 let isCommutable = 0 in {
3637 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv4i16, memopv8i16,
3638 int_x86_ssse3_phadd_w,
3639 int_x86_ssse3_phadd_w_128>;
3640 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv2i32, memopv4i32,
3641 int_x86_ssse3_phadd_d,
3642 int_x86_ssse3_phadd_d_128>;
3643 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv4i16, memopv8i16,
3644 int_x86_ssse3_phadd_sw,
3645 int_x86_ssse3_phadd_sw_128>;
3646 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv4i16, memopv8i16,
3647 int_x86_ssse3_phsub_w,
3648 int_x86_ssse3_phsub_w_128>;
3649 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv2i32, memopv4i32,
3650 int_x86_ssse3_phsub_d,
3651 int_x86_ssse3_phsub_d_128>;
3652 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv4i16, memopv8i16,
3653 int_x86_ssse3_phsub_sw,
3654 int_x86_ssse3_phsub_sw_128>;
3655 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv8i8, memopv16i8,
3656 int_x86_ssse3_pmadd_ub_sw,
3657 int_x86_ssse3_pmadd_ub_sw_128>;
3658 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv8i8, memopv16i8,
3659 int_x86_ssse3_pshuf_b,
3660 int_x86_ssse3_pshuf_b_128>;
3661 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv8i8, memopv16i8,
3662 int_x86_ssse3_psign_b,
3663 int_x86_ssse3_psign_b_128>;
3664 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv4i16, memopv8i16,
3665 int_x86_ssse3_psign_w,
3666 int_x86_ssse3_psign_w_128>;
3667 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv2i32, memopv4i32,
3668 int_x86_ssse3_psign_d,
3669 int_x86_ssse3_psign_d_128>;
3671 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv4i16, memopv8i16,
3672 int_x86_ssse3_pmul_hr_sw,
3673 int_x86_ssse3_pmul_hr_sw_128>;
3676 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3677 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3678 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3679 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3681 //===---------------------------------------------------------------------===//
3682 // SSSE3 - Packed Align Instruction Patterns
3683 //===---------------------------------------------------------------------===//
3685 multiclass sse3_palign<string asm, bit Is2Addr = 1> {
3686 def R64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
3687 (ins VR64:$src1, VR64:$src2, i8imm:$src3),
3689 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3691 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3693 def R64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
3694 (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
3696 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3698 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3701 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3702 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3704 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3706 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3708 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3709 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3711 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3713 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3717 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3718 defm VPALIGN : sse3_palign<"vpalignr", 0>, VEX_4V;
3719 let Constraints = "$src1 = $dst" in
3720 defm PALIGN : sse3_palign<"palignr">;
3722 let AddedComplexity = 5 in {
3724 def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
3725 (PALIGNR64rr VR64:$src2, VR64:$src1,
3726 (SHUFFLE_get_palign_imm VR64:$src3))>,
3727 Requires<[HasSSSE3]>;
3728 def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
3729 (PALIGNR64rr VR64:$src2, VR64:$src1,
3730 (SHUFFLE_get_palign_imm VR64:$src3))>,
3731 Requires<[HasSSSE3]>;
3732 def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
3733 (PALIGNR64rr VR64:$src2, VR64:$src1,
3734 (SHUFFLE_get_palign_imm VR64:$src3))>,
3735 Requires<[HasSSSE3]>;
3736 def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
3737 (PALIGNR64rr VR64:$src2, VR64:$src1,
3738 (SHUFFLE_get_palign_imm VR64:$src3))>,
3739 Requires<[HasSSSE3]>;
3741 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3742 (PALIGNR128rr VR128:$src2, VR128:$src1,
3743 (SHUFFLE_get_palign_imm VR128:$src3))>,
3744 Requires<[HasSSSE3]>;
3745 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3746 (PALIGNR128rr VR128:$src2, VR128:$src1,
3747 (SHUFFLE_get_palign_imm VR128:$src3))>,
3748 Requires<[HasSSSE3]>;
3749 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3750 (PALIGNR128rr VR128:$src2, VR128:$src1,
3751 (SHUFFLE_get_palign_imm VR128:$src3))>,
3752 Requires<[HasSSSE3]>;
3753 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3754 (PALIGNR128rr VR128:$src2, VR128:$src1,
3755 (SHUFFLE_get_palign_imm VR128:$src3))>,
3756 Requires<[HasSSSE3]>;
3759 //===---------------------------------------------------------------------===//
3760 // SSSE3 Misc Instructions
3761 //===---------------------------------------------------------------------===//
3763 // Thread synchronization
3764 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3765 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3766 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3767 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3769 //===---------------------------------------------------------------------===//
3770 // Non-Instruction Patterns
3771 //===---------------------------------------------------------------------===//
3773 // extload f32 -> f64. This matches load+fextend because we have a hack in
3774 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3776 // Since these loads aren't folded into the fextend, we have to match it
3778 let Predicates = [HasSSE2] in
3779 def : Pat<(fextend (loadf32 addr:$src)),
3780 (CVTSS2SDrm addr:$src)>;
3783 let Predicates = [HasSSE2] in {
3784 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3785 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3786 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3787 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3788 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3789 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3790 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3791 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3792 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3793 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3794 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3795 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3796 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3797 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3798 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3799 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3800 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3801 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3802 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3803 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3804 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3805 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3806 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3807 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3808 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3809 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3810 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3811 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3812 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3813 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3816 // Move scalar to XMM zero-extended
3817 // movd to XMM register zero-extends
3818 let AddedComplexity = 15 in {
3819 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3820 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3821 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3822 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3823 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3824 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3825 (MOVSSrr (v4f32 (V_SET0PS)),
3826 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3827 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3828 (MOVSSrr (v4i32 (V_SET0PI)),
3829 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3832 // Splat v2f64 / v2i64
3833 let AddedComplexity = 10 in {
3834 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3835 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3836 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3837 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3838 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3839 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3840 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3841 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3844 // Special unary SHUFPSrri case.
3845 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3846 (SHUFPSrri VR128:$src1, VR128:$src1,
3847 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3848 let AddedComplexity = 5 in
3849 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3850 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3851 Requires<[HasSSE2]>;
3852 // Special unary SHUFPDrri case.
3853 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3854 (SHUFPDrri VR128:$src1, VR128:$src1,
3855 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3856 Requires<[HasSSE2]>;
3857 // Special unary SHUFPDrri case.
3858 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3859 (SHUFPDrri VR128:$src1, VR128:$src1,
3860 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3861 Requires<[HasSSE2]>;
3862 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3863 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3864 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3865 Requires<[HasSSE2]>;
3867 // Special binary v4i32 shuffle cases with SHUFPS.
3868 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3869 (SHUFPSrri VR128:$src1, VR128:$src2,
3870 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3871 Requires<[HasSSE2]>;
3872 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3873 (SHUFPSrmi VR128:$src1, addr:$src2,
3874 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3875 Requires<[HasSSE2]>;
3876 // Special binary v2i64 shuffle cases using SHUFPDrri.
3877 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3878 (SHUFPDrri VR128:$src1, VR128:$src2,
3879 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3880 Requires<[HasSSE2]>;
3882 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3883 let AddedComplexity = 15 in {
3884 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3885 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3886 Requires<[OptForSpeed, HasSSE2]>;
3887 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3888 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3889 Requires<[OptForSpeed, HasSSE2]>;
3891 let AddedComplexity = 10 in {
3892 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3893 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3894 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3895 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3896 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3897 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3898 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3899 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3902 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3903 let AddedComplexity = 15 in {
3904 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3905 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3906 Requires<[OptForSpeed, HasSSE2]>;
3907 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3908 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3909 Requires<[OptForSpeed, HasSSE2]>;
3911 let AddedComplexity = 10 in {
3912 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3913 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3914 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3915 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3916 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3917 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3918 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3919 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3922 let AddedComplexity = 20 in {
3923 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3924 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3925 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3927 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3928 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3929 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3931 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3932 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3933 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3934 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3935 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3938 let AddedComplexity = 20 in {
3939 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3940 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3941 (MOVLPSrm VR128:$src1, addr:$src2)>;
3942 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3943 (MOVLPDrm VR128:$src1, addr:$src2)>;
3944 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3945 (MOVLPSrm VR128:$src1, addr:$src2)>;
3946 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3947 (MOVLPDrm VR128:$src1, addr:$src2)>;
3950 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3951 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3952 (MOVLPSmr addr:$src1, VR128:$src2)>;
3953 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3954 (MOVLPDmr addr:$src1, VR128:$src2)>;
3955 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3957 (MOVLPSmr addr:$src1, VR128:$src2)>;
3958 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3959 (MOVLPDmr addr:$src1, VR128:$src2)>;
3961 let AddedComplexity = 15 in {
3962 // Setting the lowest element in the vector.
3963 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3964 (MOVSSrr (v4i32 VR128:$src1),
3965 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3966 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3967 (MOVSDrr (v2i64 VR128:$src1),
3968 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3970 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3971 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3972 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3973 Requires<[HasSSE2]>;
3974 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3975 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3976 Requires<[HasSSE2]>;
3979 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3980 // fall back to this for SSE1)
3981 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3982 (SHUFPSrri VR128:$src2, VR128:$src1,
3983 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3985 // Set lowest element and zero upper elements.
3986 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3987 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3989 // Some special case pandn patterns.
3990 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3992 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3993 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3995 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3996 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3998 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
4000 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
4001 (memop addr:$src2))),
4002 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
4003 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
4004 (memop addr:$src2))),
4005 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
4006 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
4007 (memop addr:$src2))),
4008 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
4010 // vector -> vector casts
4011 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
4012 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
4013 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
4014 (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
4015 def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
4016 (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
4017 def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
4018 (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
4020 // Use movaps / movups for SSE integer load / store (one byte shorter).
4021 def : Pat<(alignedloadv4i32 addr:$src),
4022 (MOVAPSrm addr:$src)>;
4023 def : Pat<(loadv4i32 addr:$src),
4024 (MOVUPSrm addr:$src)>;
4025 def : Pat<(alignedloadv2i64 addr:$src),
4026 (MOVAPSrm addr:$src)>;
4027 def : Pat<(loadv2i64 addr:$src),
4028 (MOVUPSrm addr:$src)>;
4030 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4031 (MOVAPSmr addr:$dst, VR128:$src)>;
4032 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4033 (MOVAPSmr addr:$dst, VR128:$src)>;
4034 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4035 (MOVAPSmr addr:$dst, VR128:$src)>;
4036 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4037 (MOVAPSmr addr:$dst, VR128:$src)>;
4038 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4039 (MOVUPSmr addr:$dst, VR128:$src)>;
4040 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4041 (MOVUPSmr addr:$dst, VR128:$src)>;
4042 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4043 (MOVUPSmr addr:$dst, VR128:$src)>;
4044 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4045 (MOVUPSmr addr:$dst, VR128:$src)>;
4047 //===----------------------------------------------------------------------===//
4048 // SSE4.1 - Packed Move with Sign/Zero Extend
4049 //===----------------------------------------------------------------------===//
4051 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4052 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4053 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4054 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4056 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4057 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4059 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4063 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4064 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4066 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4068 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4070 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4072 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4074 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4078 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4079 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4080 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4081 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4082 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4083 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4085 // Common patterns involving scalar load.
4086 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4087 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4088 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4089 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4091 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4092 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4093 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4094 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4096 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4097 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4098 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4099 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4101 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4102 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4103 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4104 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4106 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4107 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4108 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4109 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4111 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4112 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4113 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4114 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4117 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4118 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4119 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4120 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4122 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4123 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4125 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4129 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4130 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4132 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4134 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4136 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4140 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4141 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4142 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4143 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4145 // Common patterns involving scalar load
4146 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4147 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4148 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4149 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4151 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4152 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4153 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4154 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4157 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4158 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4159 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4160 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4162 // Expecting a i16 load any extended to i32 value.
4163 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4164 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4165 [(set VR128:$dst, (IntId (bitconvert
4166 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4170 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4171 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4173 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4176 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4177 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4179 // Common patterns involving scalar load
4180 def : Pat<(int_x86_sse41_pmovsxbq
4181 (bitconvert (v4i32 (X86vzmovl
4182 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4183 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4185 def : Pat<(int_x86_sse41_pmovzxbq
4186 (bitconvert (v4i32 (X86vzmovl
4187 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4188 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4190 //===----------------------------------------------------------------------===//
4191 // SSE4.1 - Extract Instructions
4192 //===----------------------------------------------------------------------===//
4194 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4195 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4196 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4197 (ins VR128:$src1, i32i8imm:$src2),
4198 !strconcat(OpcodeStr,
4199 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4200 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4202 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4203 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4204 !strconcat(OpcodeStr,
4205 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4208 // There's an AssertZext in the way of writing the store pattern
4209 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4212 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4213 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4214 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4215 (ins VR128:$src1, i32i8imm:$src2),
4216 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4219 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4222 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4223 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4224 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4225 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4226 !strconcat(OpcodeStr,
4227 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4230 // There's an AssertZext in the way of writing the store pattern
4231 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4234 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4235 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4237 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4240 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4241 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4242 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4243 (ins VR128:$src1, i32i8imm:$src2),
4244 !strconcat(OpcodeStr,
4245 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4247 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4248 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4249 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4250 !strconcat(OpcodeStr,
4251 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4252 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4253 addr:$dst)]>, OpSize;
4256 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4257 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4259 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4261 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4262 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4263 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4264 (ins VR128:$src1, i32i8imm:$src2),
4265 !strconcat(OpcodeStr,
4266 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4268 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4269 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4270 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4271 !strconcat(OpcodeStr,
4272 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4273 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4274 addr:$dst)]>, OpSize, REX_W;
4277 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4278 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4280 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4282 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4284 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4285 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4286 (ins VR128:$src1, i32i8imm:$src2),
4287 !strconcat(OpcodeStr,
4288 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4290 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4292 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4293 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4294 !strconcat(OpcodeStr,
4295 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4296 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4297 addr:$dst)]>, OpSize;
4300 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4301 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4302 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4303 (ins VR128:$src1, i32i8imm:$src2),
4304 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4307 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4309 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4310 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4313 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4314 Requires<[HasSSE41]>;
4316 //===----------------------------------------------------------------------===//
4317 // SSE4.1 - Insert Instructions
4318 //===----------------------------------------------------------------------===//
4320 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4321 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4322 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4324 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4326 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4328 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4329 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4330 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4332 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4334 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4336 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4337 imm:$src3))]>, OpSize;
4340 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4341 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4342 let Constraints = "$src1 = $dst" in
4343 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4345 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4346 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4347 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4349 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4351 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4353 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4355 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4356 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4358 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4360 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4362 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4363 imm:$src3)))]>, OpSize;
4366 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4367 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4368 let Constraints = "$src1 = $dst" in
4369 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4371 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4372 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4373 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4375 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4377 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4379 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4381 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4382 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4384 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4386 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4388 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4389 imm:$src3)))]>, OpSize;
4392 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4393 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4394 let Constraints = "$src1 = $dst" in
4395 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4397 // insertps has a few different modes, there's the first two here below which
4398 // are optimized inserts that won't zero arbitrary elements in the destination
4399 // vector. The next one matches the intrinsic and could zero arbitrary elements
4400 // in the target vector.
4401 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4402 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4403 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4405 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4407 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4409 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4411 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4412 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4414 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4416 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4418 (X86insrtps VR128:$src1,
4419 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4420 imm:$src3))]>, OpSize;
4423 let Constraints = "$src1 = $dst" in
4424 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4425 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4426 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4428 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4429 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4431 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4432 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4433 Requires<[HasSSE41]>;
4435 //===----------------------------------------------------------------------===//
4436 // SSE4.1 - Round Instructions
4437 //===----------------------------------------------------------------------===//
4439 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4440 X86MemOperand x86memop, RegisterClass RC,
4441 PatFrag mem_frag32, PatFrag mem_frag64,
4442 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4443 // Intrinsic operation, reg.
4444 // Vector intrinsic operation, reg
4445 def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
4446 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4447 !strconcat(OpcodeStr,
4448 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4449 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4452 // Vector intrinsic operation, mem
4453 def PSm_Int : Ii8<opcps, MRMSrcMem,
4454 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4455 !strconcat(OpcodeStr,
4456 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4458 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4460 Requires<[HasSSE41]>;
4462 // Vector intrinsic operation, reg
4463 def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
4464 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4465 !strconcat(OpcodeStr,
4466 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4467 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4470 // Vector intrinsic operation, mem
4471 def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
4472 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4473 !strconcat(OpcodeStr,
4474 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4476 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4480 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4481 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4482 // Intrinsic operation, reg.
4483 // Vector intrinsic operation, reg
4484 def PSr : SS4AIi8<opcps, MRMSrcReg,
4485 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4486 !strconcat(OpcodeStr,
4487 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4490 // Vector intrinsic operation, mem
4491 def PSm : Ii8<opcps, MRMSrcMem,
4492 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4493 !strconcat(OpcodeStr,
4494 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4495 []>, TA, OpSize, Requires<[HasSSE41]>;
4497 // Vector intrinsic operation, reg
4498 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4499 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4500 !strconcat(OpcodeStr,
4501 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4504 // Vector intrinsic operation, mem
4505 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4506 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4507 !strconcat(OpcodeStr,
4508 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4512 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4515 Intrinsic F64Int, bit Is2Addr = 1> {
4516 // Intrinsic operation, reg.
4517 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
4518 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4520 !strconcat(OpcodeStr,
4521 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4522 !strconcat(OpcodeStr,
4523 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4524 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4527 // Intrinsic operation, mem.
4528 def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
4529 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4531 !strconcat(OpcodeStr,
4532 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4533 !strconcat(OpcodeStr,
4534 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4536 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4539 // Intrinsic operation, reg.
4540 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
4541 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4543 !strconcat(OpcodeStr,
4544 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4545 !strconcat(OpcodeStr,
4546 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4547 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4550 // Intrinsic operation, mem.
4551 def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
4552 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4554 !strconcat(OpcodeStr,
4555 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4556 !strconcat(OpcodeStr,
4557 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4559 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4563 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4565 // Intrinsic operation, reg.
4566 def SSr : SS4AIi8<opcss, MRMSrcReg,
4567 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4568 !strconcat(OpcodeStr,
4569 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4572 // Intrinsic operation, mem.
4573 def SSm : SS4AIi8<opcss, MRMSrcMem,
4574 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4575 !strconcat(OpcodeStr,
4576 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4579 // Intrinsic operation, reg.
4580 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4581 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4582 !strconcat(OpcodeStr,
4583 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4586 // Intrinsic operation, mem.
4587 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4588 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4589 !strconcat(OpcodeStr,
4590 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4594 // FP round - roundss, roundps, roundsd, roundpd
4595 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4597 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4598 memopv4f32, memopv2f64,
4599 int_x86_sse41_round_ps,
4600 int_x86_sse41_round_pd>, VEX;
4601 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4602 memopv8f32, memopv4f64,
4603 int_x86_avx_round_ps_256,
4604 int_x86_avx_round_pd_256>, VEX;
4605 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4606 int_x86_sse41_round_ss,
4607 int_x86_sse41_round_sd, 0>, VEX_4V;
4609 // Instructions for the assembler
4610 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4612 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4614 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4617 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4618 memopv4f32, memopv2f64,
4619 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4620 let Constraints = "$src1 = $dst" in
4621 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4622 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4624 //===----------------------------------------------------------------------===//
4625 // SSE4.1 - Packed Bit Test
4626 //===----------------------------------------------------------------------===//
4628 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4629 // the intel intrinsic that corresponds to this.
4630 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4631 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4632 "vptest\t{$src2, $src1|$src1, $src2}",
4633 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4635 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4636 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4638 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4639 "vptest\t{$src2, $src1|$src1, $src2}",
4640 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4642 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4643 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4646 let Defs = [EFLAGS] in {
4647 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4648 "ptest \t{$src2, $src1|$src1, $src2}",
4649 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4651 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4652 "ptest \t{$src2, $src1|$src1, $src2}",
4653 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4657 // The bit test instructions below are AVX only
4658 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4659 X86MemOperand x86memop> {
4660 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4663 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4664 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4668 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4669 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem>;
4670 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem>;
4671 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem>;
4672 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem>;
4675 //===----------------------------------------------------------------------===//
4676 // SSE4.1 - Misc Instructions
4677 //===----------------------------------------------------------------------===//
4679 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4680 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4681 Intrinsic IntId128> {
4682 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4684 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4685 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4686 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4688 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4691 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4694 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4695 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4696 int_x86_sse41_phminposuw>, VEX;
4697 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4698 int_x86_sse41_phminposuw>;
4700 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4701 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4702 Intrinsic IntId128, bit Is2Addr = 1> {
4703 let isCommutable = 1 in
4704 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4705 (ins VR128:$src1, VR128:$src2),
4707 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4708 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4709 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4710 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4711 (ins VR128:$src1, i128mem:$src2),
4713 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4714 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4716 (IntId128 VR128:$src1,
4717 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4720 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4721 let isCommutable = 0 in
4722 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4724 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4726 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4728 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4730 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4732 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4734 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4736 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4738 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4740 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4742 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4746 let Constraints = "$src1 = $dst" in {
4747 let isCommutable = 0 in
4748 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4749 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4750 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4751 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4752 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4753 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4754 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4755 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4756 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4757 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4758 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4761 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4762 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4763 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4764 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4766 /// SS48I_binop_rm - Simple SSE41 binary operator.
4767 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4768 ValueType OpVT, bit Is2Addr = 1> {
4769 let isCommutable = 1 in
4770 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4771 (ins VR128:$src1, VR128:$src2),
4773 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4774 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4775 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4777 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4778 (ins VR128:$src1, i128mem:$src2),
4780 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4781 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4782 [(set VR128:$dst, (OpNode VR128:$src1,
4783 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4787 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4788 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4789 let Constraints = "$src1 = $dst" in
4790 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4792 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4793 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4794 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4795 X86MemOperand x86memop, bit Is2Addr = 1> {
4796 let isCommutable = 1 in
4797 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4798 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4800 !strconcat(OpcodeStr,
4801 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4802 !strconcat(OpcodeStr,
4803 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4804 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4806 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4807 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4809 !strconcat(OpcodeStr,
4810 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4811 !strconcat(OpcodeStr,
4812 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4815 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4819 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4820 let isCommutable = 0 in {
4821 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4822 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4823 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4824 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4825 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4826 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4827 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4828 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4829 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4830 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4831 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4832 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4834 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4835 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4836 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4837 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4838 let Pattern = []<dag> in
4839 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4840 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4843 let Constraints = "$src1 = $dst" in {
4844 let isCommutable = 0 in {
4845 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4846 VR128, memopv16i8, i128mem>;
4847 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4848 VR128, memopv16i8, i128mem>;
4849 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4850 VR128, memopv16i8, i128mem>;
4851 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4852 VR128, memopv16i8, i128mem>;
4854 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4855 VR128, memopv16i8, i128mem>;
4856 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4857 VR128, memopv16i8, i128mem>;
4860 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4861 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4862 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4863 RegisterClass RC, X86MemOperand x86memop,
4864 PatFrag mem_frag, Intrinsic IntId> {
4865 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4866 (ins RC:$src1, RC:$src2, RC:$src3),
4867 !strconcat(OpcodeStr,
4868 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4869 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4870 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4872 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4873 (ins RC:$src1, x86memop:$src2, RC:$src3),
4874 !strconcat(OpcodeStr,
4875 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4877 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4879 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4883 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4884 memopv16i8, int_x86_sse41_blendvpd>;
4885 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4886 memopv16i8, int_x86_sse41_blendvps>;
4887 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4888 memopv16i8, int_x86_sse41_pblendvb>;
4889 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4890 memopv32i8, int_x86_avx_blendv_pd_256>;
4891 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4892 memopv32i8, int_x86_avx_blendv_ps_256>;
4894 /// SS41I_ternary_int - SSE 4.1 ternary operator
4895 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4896 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4897 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4898 (ins VR128:$src1, VR128:$src2),
4899 !strconcat(OpcodeStr,
4900 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4901 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4904 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4905 (ins VR128:$src1, i128mem:$src2),
4906 !strconcat(OpcodeStr,
4907 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4910 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4914 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4915 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4916 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4918 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4919 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4920 "vmovntdqa\t{$src, $dst|$dst, $src}",
4921 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4923 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4924 "movntdqa\t{$src, $dst|$dst, $src}",
4925 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4928 //===----------------------------------------------------------------------===//
4929 // SSE4.2 - Compare Instructions
4930 //===----------------------------------------------------------------------===//
4932 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4933 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4934 Intrinsic IntId128, bit Is2Addr = 1> {
4935 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4936 (ins VR128:$src1, VR128:$src2),
4938 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4939 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4940 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4942 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4943 (ins VR128:$src1, i128mem:$src2),
4945 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4946 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4948 (IntId128 VR128:$src1,
4949 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4952 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4953 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4955 let Constraints = "$src1 = $dst" in
4956 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4958 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4959 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4960 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4961 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4963 //===----------------------------------------------------------------------===//
4964 // SSE4.2 - String/text Processing Instructions
4965 //===----------------------------------------------------------------------===//
4967 // Packed Compare Implicit Length Strings, Return Mask
4968 multiclass pseudo_pcmpistrm<string asm> {
4969 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4970 (ins VR128:$src1, VR128:$src2, i8imm:$src3), !strconcat(asm, "rr PSEUDO"),
4971 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4973 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4974 (ins VR128:$src1, i128mem:$src2, i8imm:$src3), !strconcat(asm, "rm PSEUDO"),
4975 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4976 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4979 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4980 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4981 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4984 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4985 Predicates = [HasAVX] in {
4986 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4987 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4988 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4989 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4990 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4991 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4994 let Defs = [XMM0, EFLAGS] in {
4995 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4996 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4997 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4998 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4999 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5000 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5003 // Packed Compare Explicit Length Strings, Return Mask
5004 multiclass pseudo_pcmpestrm<string asm> {
5005 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
5006 (ins VR128:$src1, VR128:$src3, i8imm:$src5), !strconcat(asm, "rr PSEUDO"),
5007 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5008 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5009 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
5010 (ins VR128:$src1, i128mem:$src3, i8imm:$src5), !strconcat(asm, "rm PSEUDO"),
5011 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5012 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5015 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5016 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5017 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5020 let isAsmParserOnly = 1, Predicates = [HasAVX],
5021 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5022 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5023 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5024 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5025 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5026 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5027 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5030 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5031 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5032 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5033 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5034 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5035 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5036 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5039 // Packed Compare Implicit Length Strings, Return Index
5040 let Defs = [ECX, EFLAGS] in {
5041 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5042 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5043 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5044 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5045 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5046 (implicit EFLAGS)]>, OpSize;
5047 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5048 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5049 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5050 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5051 (implicit EFLAGS)]>, OpSize;
5055 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5056 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5058 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5060 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5062 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5064 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5066 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5070 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5071 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5072 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5073 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5074 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5075 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5077 // Packed Compare Explicit Length Strings, Return Index
5078 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5079 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5080 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5081 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5082 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5083 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5084 (implicit EFLAGS)]>, OpSize;
5085 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5086 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5087 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5089 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5090 (implicit EFLAGS)]>, OpSize;
5094 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5095 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5097 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5099 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5101 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5103 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5105 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5109 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5110 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5111 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5112 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5113 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5114 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5116 //===----------------------------------------------------------------------===//
5117 // SSE4.2 - CRC Instructions
5118 //===----------------------------------------------------------------------===//
5120 // No CRC instructions have AVX equivalents
5122 // crc intrinsic instruction
5123 // This set of instructions are only rm, the only difference is the size
5125 let Constraints = "$src1 = $dst" in {
5126 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5127 (ins GR32:$src1, i8mem:$src2),
5128 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5130 (int_x86_sse42_crc32_8 GR32:$src1,
5131 (load addr:$src2)))]>;
5132 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5133 (ins GR32:$src1, GR8:$src2),
5134 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5136 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5137 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5138 (ins GR32:$src1, i16mem:$src2),
5139 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5141 (int_x86_sse42_crc32_16 GR32:$src1,
5142 (load addr:$src2)))]>,
5144 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5145 (ins GR32:$src1, GR16:$src2),
5146 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5148 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5150 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5151 (ins GR32:$src1, i32mem:$src2),
5152 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5154 (int_x86_sse42_crc32_32 GR32:$src1,
5155 (load addr:$src2)))]>;
5156 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5157 (ins GR32:$src1, GR32:$src2),
5158 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5160 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5161 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5162 (ins GR64:$src1, i8mem:$src2),
5163 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5165 (int_x86_sse42_crc64_8 GR64:$src1,
5166 (load addr:$src2)))]>,
5168 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5169 (ins GR64:$src1, GR8:$src2),
5170 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5172 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5174 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5175 (ins GR64:$src1, i64mem:$src2),
5176 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5178 (int_x86_sse42_crc64_64 GR64:$src1,
5179 (load addr:$src2)))]>,
5181 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5182 (ins GR64:$src1, GR64:$src2),
5183 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5185 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5189 //===----------------------------------------------------------------------===//
5190 // AES-NI Instructions
5191 //===----------------------------------------------------------------------===//
5193 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5194 Intrinsic IntId128, bit Is2Addr = 1> {
5195 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5196 (ins VR128:$src1, VR128:$src2),
5198 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5199 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5200 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5202 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5203 (ins VR128:$src1, i128mem:$src2),
5205 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5206 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5208 (IntId128 VR128:$src1,
5209 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5212 // Perform One Round of an AES Encryption/Decryption Flow
5213 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5214 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5215 int_x86_aesni_aesenc, 0>, VEX_4V;
5216 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5217 int_x86_aesni_aesenclast, 0>, VEX_4V;
5218 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5219 int_x86_aesni_aesdec, 0>, VEX_4V;
5220 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5221 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5224 let Constraints = "$src1 = $dst" in {
5225 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5226 int_x86_aesni_aesenc>;
5227 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5228 int_x86_aesni_aesenclast>;
5229 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5230 int_x86_aesni_aesdec>;
5231 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5232 int_x86_aesni_aesdeclast>;
5235 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5236 (AESENCrr VR128:$src1, VR128:$src2)>;
5237 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5238 (AESENCrm VR128:$src1, addr:$src2)>;
5239 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5240 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5241 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5242 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5243 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5244 (AESDECrr VR128:$src1, VR128:$src2)>;
5245 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5246 (AESDECrm VR128:$src1, addr:$src2)>;
5247 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5248 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5249 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5250 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5252 // Perform the AES InvMixColumn Transformation
5253 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5254 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5256 "vaesimc\t{$src1, $dst|$dst, $src1}",
5258 (int_x86_aesni_aesimc VR128:$src1))]>,
5260 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5261 (ins i128mem:$src1),
5262 "vaesimc\t{$src1, $dst|$dst, $src1}",
5264 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5267 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5269 "aesimc\t{$src1, $dst|$dst, $src1}",
5271 (int_x86_aesni_aesimc VR128:$src1))]>,
5273 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5274 (ins i128mem:$src1),
5275 "aesimc\t{$src1, $dst|$dst, $src1}",
5277 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5280 // AES Round Key Generation Assist
5281 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5282 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5283 (ins VR128:$src1, i8imm:$src2),
5284 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5286 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5288 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5289 (ins i128mem:$src1, i8imm:$src2),
5290 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5292 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5296 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5297 (ins VR128:$src1, i8imm:$src2),
5298 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5300 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5302 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5303 (ins i128mem:$src1, i8imm:$src2),
5304 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5306 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5310 //===----------------------------------------------------------------------===//
5311 // CLMUL Instructions
5312 //===----------------------------------------------------------------------===//
5314 // Only the AVX version of CLMUL instructions are described here.
5316 // Carry-less Multiplication instructions
5317 let isAsmParserOnly = 1 in {
5318 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5319 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5320 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5323 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5324 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5325 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5329 multiclass avx_vpclmul<string asm> {
5330 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5331 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5334 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5335 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5338 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5339 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5340 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5341 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5343 } // isAsmParserOnly
5345 //===----------------------------------------------------------------------===//
5347 //===----------------------------------------------------------------------===//
5349 let isAsmParserOnly = 1 in {
5351 // Load from memory and broadcast to all elements of the destination operand
5352 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5353 X86MemOperand x86memop, Intrinsic Int> :
5354 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5355 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5356 [(set RC:$dst, (Int addr:$src))]>, VEX;
5358 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5359 int_x86_avx_vbroadcastss>;
5360 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5361 int_x86_avx_vbroadcastss_256>;
5362 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5363 int_x86_avx_vbroadcast_sd_256>;
5364 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5365 int_x86_avx_vbroadcastf128_pd_256>;
5367 // Insert packed floating-point values
5368 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5369 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5370 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5372 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5373 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5374 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5377 // Extract packed floating-point values
5378 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5379 (ins VR256:$src1, i8imm:$src2),
5380 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5382 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5383 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5384 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5387 // Conditional SIMD Packed Loads and Stores
5388 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5389 Intrinsic IntLd, Intrinsic IntLd256,
5390 Intrinsic IntSt, Intrinsic IntSt256,
5391 PatFrag pf128, PatFrag pf256> {
5392 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5393 (ins VR128:$src1, f128mem:$src2),
5394 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5395 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5397 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5398 (ins VR256:$src1, f256mem:$src2),
5399 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5400 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5402 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5403 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5404 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5405 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5406 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5407 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5408 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5409 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5412 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5413 int_x86_avx_maskload_ps,
5414 int_x86_avx_maskload_ps_256,
5415 int_x86_avx_maskstore_ps,
5416 int_x86_avx_maskstore_ps_256,
5417 memopv4f32, memopv8f32>;
5418 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5419 int_x86_avx_maskload_pd,
5420 int_x86_avx_maskload_pd_256,
5421 int_x86_avx_maskstore_pd,
5422 int_x86_avx_maskstore_pd_256,
5423 memopv2f64, memopv4f64>;
5425 // Permute Floating-Point Values
5426 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5427 RegisterClass RC, X86MemOperand x86memop_f,
5428 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5429 Intrinsic IntVar, Intrinsic IntImm> {
5430 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5431 (ins RC:$src1, RC:$src2),
5432 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5433 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5434 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5435 (ins RC:$src1, x86memop_i:$src2),
5436 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5437 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5439 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5440 (ins RC:$src1, i8imm:$src2),
5441 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5442 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5443 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5444 (ins x86memop_f:$src1, i8imm:$src2),
5445 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5446 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5449 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5450 memopv4f32, memopv4i32,
5451 int_x86_avx_vpermilvar_ps,
5452 int_x86_avx_vpermil_ps>;
5453 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5454 memopv8f32, memopv8i32,
5455 int_x86_avx_vpermilvar_ps_256,
5456 int_x86_avx_vpermil_ps_256>;
5457 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5458 memopv2f64, memopv2i64,
5459 int_x86_avx_vpermilvar_pd,
5460 int_x86_avx_vpermil_pd>;
5461 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5462 memopv4f64, memopv4i64,
5463 int_x86_avx_vpermilvar_pd_256,
5464 int_x86_avx_vpermil_pd_256>;
5466 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5467 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5468 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5470 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5471 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5472 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5475 // Zero All YMM registers
5476 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5477 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5479 // Zero Upper bits of YMM registers
5480 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5481 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5483 } // isAsmParserOnly
5485 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5486 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5487 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5488 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5489 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5490 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5492 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5493 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5494 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5495 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5496 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5497 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5499 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5500 (VBROADCASTF128 addr:$src)>;
5502 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5503 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5504 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5505 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5506 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5507 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5509 def : Pat<(int_x86_avx_vperm2f128_ps_256
5510 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5511 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5512 def : Pat<(int_x86_avx_vperm2f128_pd_256
5513 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5514 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5515 def : Pat<(int_x86_avx_vperm2f128_si_256
5516 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5517 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;