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_sse",
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_sse",
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, $src}">, XS, VEX,
754 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load,
755 "cvtdq2ps\t{$src, $dst|$dst, $src}",
756 SSEPackedSingle>, TB, VEX;
757 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, f256mem, load,
758 "cvtdq2ps\t{$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, $src}">, XS, REX_W;
766 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load /*dummy*/,
767 "cvtdq2ps\t{$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:$src))))]>,
895 XS, Requires<[HasSSE2]>;
897 // Convert packed single/double fp to doubleword
898 let isAsmParserOnly = 1 in {
899 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
900 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
901 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
902 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
903 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
904 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
905 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
906 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
908 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
909 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
910 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
911 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
913 let isAsmParserOnly = 1 in {
914 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
915 "cvtps2dq\t{$src, $dst|$dst, $src}",
916 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
918 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
920 "cvtps2dq\t{$src, $dst|$dst, $src}",
921 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
922 (memop addr:$src)))]>, VEX;
924 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
925 "cvtps2dq\t{$src, $dst|$dst, $src}",
926 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
927 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
928 "cvtps2dq\t{$src, $dst|$dst, $src}",
929 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
930 (memop addr:$src)))]>;
932 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
933 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
934 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
935 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
936 XD, VEX, Requires<[HasAVX]>;
937 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
938 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
939 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
940 (memop addr:$src)))]>,
941 XD, VEX, Requires<[HasAVX]>;
943 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
944 "cvtpd2dq\t{$src, $dst|$dst, $src}",
945 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
946 XD, Requires<[HasSSE2]>;
947 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
948 "cvtpd2dq\t{$src, $dst|$dst, $src}",
949 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
950 (memop addr:$src)))]>,
951 XD, Requires<[HasSSE2]>;
954 // Convert with truncation packed single/double fp to doubleword
955 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
956 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
960 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
961 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
963 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
965 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
966 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
967 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
968 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
971 let isAsmParserOnly = 1 in {
972 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
973 "vcvttps2dq\t{$src, $dst|$dst, $src}",
975 (int_x86_sse2_cvttps2dq VR128:$src))]>,
976 XS, VEX, Requires<[HasAVX]>;
977 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
978 "vcvttps2dq\t{$src, $dst|$dst, $src}",
979 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
980 (memop addr:$src)))]>,
981 XS, VEX, Requires<[HasAVX]>;
983 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
984 "cvttps2dq\t{$src, $dst|$dst, $src}",
986 (int_x86_sse2_cvttps2dq VR128:$src))]>,
987 XS, Requires<[HasSSE2]>;
988 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
989 "cvttps2dq\t{$src, $dst|$dst, $src}",
990 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
991 (memop addr:$src)))]>,
992 XS, Requires<[HasSSE2]>;
994 let isAsmParserOnly = 1 in {
995 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
997 "cvttpd2dq\t{$src, $dst|$dst, $src}",
998 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1000 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1002 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1003 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1004 (memop addr:$src)))]>, VEX;
1006 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1007 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1008 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1009 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1010 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1011 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1012 (memop addr:$src)))]>;
1014 let isAsmParserOnly = 1 in {
1015 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1016 // register, but the same isn't true when using memory operands instead.
1017 // Provide other assembly rr and rm forms to address this explicitly.
1018 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1019 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1020 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1021 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1024 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1025 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1026 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1027 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1030 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1031 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1032 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1033 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1036 // Convert packed single to packed double
1037 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1038 // SSE2 instructions without OpSize prefix
1039 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1040 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1041 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1042 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1043 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1044 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1045 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1046 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1048 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1049 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1050 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1051 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1053 let isAsmParserOnly = 1 in {
1054 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1055 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1056 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1057 VEX, Requires<[HasAVX]>;
1058 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1059 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1060 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1061 (load addr:$src)))]>,
1062 VEX, Requires<[HasAVX]>;
1064 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1065 "cvtps2pd\t{$src, $dst|$dst, $src}",
1066 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1067 TB, Requires<[HasSSE2]>;
1068 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1069 "cvtps2pd\t{$src, $dst|$dst, $src}",
1070 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1071 (load addr:$src)))]>,
1072 TB, Requires<[HasSSE2]>;
1074 // Convert packed double to packed single
1075 let isAsmParserOnly = 1 in {
1076 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1077 // register, but the same isn't true when using memory operands instead.
1078 // Provide other assembly rr and rm forms to address this explicitly.
1079 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1080 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1081 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1082 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1085 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1086 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1087 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1088 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1091 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1092 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1093 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1094 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1096 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1097 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1098 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1099 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1102 let isAsmParserOnly = 1 in {
1103 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1104 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1105 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1106 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1108 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1109 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1110 (memop addr:$src)))]>;
1112 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1113 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1114 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1115 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1116 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1117 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1118 (memop addr:$src)))]>;
1120 //===----------------------------------------------------------------------===//
1121 // SSE 1 & 2 - Compare Instructions
1122 //===----------------------------------------------------------------------===//
1124 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1125 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1126 string asm, string asm_alt> {
1127 def rr : SIi8<0xC2, MRMSrcReg,
1128 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1131 def rm : SIi8<0xC2, MRMSrcMem,
1132 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1134 // Accept explicit immediate argument form instead of comparison code.
1135 let isAsmParserOnly = 1 in {
1136 def rr_alt : SIi8<0xC2, MRMSrcReg,
1137 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1140 def rm_alt : SIi8<0xC2, MRMSrcMem,
1141 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1146 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1147 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1148 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1149 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1151 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1152 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1153 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1157 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1158 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1159 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1160 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1161 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1162 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1163 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1166 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1167 Intrinsic Int, string asm> {
1168 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1169 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1170 [(set VR128:$dst, (Int VR128:$src1,
1171 VR128:$src, imm:$cc))]>;
1172 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1173 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1174 [(set VR128:$dst, (Int VR128:$src1,
1175 (load addr:$src), imm:$cc))]>;
1178 // Aliases to match intrinsics which expect XMM operand(s).
1179 let isAsmParserOnly = 1 in {
1180 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1181 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1183 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1184 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1187 let Constraints = "$src1 = $dst" in {
1188 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1189 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1190 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1191 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1195 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1196 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1197 ValueType vt, X86MemOperand x86memop,
1198 PatFrag ld_frag, string OpcodeStr, Domain d> {
1199 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1200 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1201 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1202 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1203 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1204 [(set EFLAGS, (OpNode (vt RC:$src1),
1205 (ld_frag addr:$src2)))], d>;
1208 let Defs = [EFLAGS] in {
1209 let isAsmParserOnly = 1 in {
1210 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1211 "ucomiss", SSEPackedSingle>, VEX;
1212 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1213 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1214 let Pattern = []<dag> in {
1215 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1216 "comiss", SSEPackedSingle>, VEX;
1217 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1218 "comisd", SSEPackedDouble>, OpSize, VEX;
1221 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1222 load, "ucomiss", SSEPackedSingle>, VEX;
1223 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1224 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1226 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1227 load, "comiss", SSEPackedSingle>, VEX;
1228 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1229 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1231 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1232 "ucomiss", SSEPackedSingle>, TB;
1233 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1234 "ucomisd", SSEPackedDouble>, TB, OpSize;
1236 let Pattern = []<dag> in {
1237 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1238 "comiss", SSEPackedSingle>, TB;
1239 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1240 "comisd", SSEPackedDouble>, TB, OpSize;
1243 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1244 load, "ucomiss", SSEPackedSingle>, TB;
1245 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1246 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1248 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1249 "comiss", SSEPackedSingle>, TB;
1250 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1251 "comisd", SSEPackedDouble>, TB, OpSize;
1252 } // Defs = [EFLAGS]
1254 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1255 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1256 Intrinsic Int, string asm, string asm_alt,
1258 def rri : PIi8<0xC2, MRMSrcReg,
1259 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1260 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1261 def rmi : PIi8<0xC2, MRMSrcMem,
1262 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1263 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1264 // Accept explicit immediate argument form instead of comparison code.
1265 let isAsmParserOnly = 1 in {
1266 def rri_alt : PIi8<0xC2, MRMSrcReg,
1267 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1269 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1270 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1275 let isAsmParserOnly = 1 in {
1276 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1277 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1278 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1279 SSEPackedSingle>, VEX_4V;
1280 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1281 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1282 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1283 SSEPackedDouble>, OpSize, VEX_4V;
1284 let Pattern = []<dag> in {
1285 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_sse_cmp_ps,
1286 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1287 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1288 SSEPackedSingle>, VEX_4V;
1289 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_sse2_cmp_pd,
1290 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1291 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1292 SSEPackedDouble>, OpSize, VEX_4V;
1295 let Constraints = "$src1 = $dst" in {
1296 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1297 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1298 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1299 SSEPackedSingle>, TB;
1300 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1301 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1302 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1303 SSEPackedDouble>, TB, OpSize;
1306 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1307 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1308 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1309 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1310 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1311 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1312 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1313 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1315 //===----------------------------------------------------------------------===//
1316 // SSE 1 & 2 - Shuffle Instructions
1317 //===----------------------------------------------------------------------===//
1319 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1320 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1321 ValueType vt, string asm, PatFrag mem_frag,
1322 Domain d, bit IsConvertibleToThreeAddress = 0> {
1323 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1324 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1325 [(set RC:$dst, (vt (shufp:$src3
1326 RC:$src1, (mem_frag addr:$src2))))], d>;
1327 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1328 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1329 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1331 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1334 let isAsmParserOnly = 1 in {
1335 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1336 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1337 memopv4f32, SSEPackedSingle>, VEX_4V;
1338 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1339 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1340 memopv8f32, SSEPackedSingle>, VEX_4V;
1341 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1342 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1343 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1344 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1345 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1346 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1349 let Constraints = "$src1 = $dst" in {
1350 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1351 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1352 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1354 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1355 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1356 memopv2f64, SSEPackedDouble>, TB, OpSize;
1359 //===----------------------------------------------------------------------===//
1360 // SSE 1 & 2 - Unpack Instructions
1361 //===----------------------------------------------------------------------===//
1363 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1364 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1365 PatFrag mem_frag, RegisterClass RC,
1366 X86MemOperand x86memop, string asm,
1368 def rr : PI<opc, MRMSrcReg,
1369 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1371 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1372 def rm : PI<opc, MRMSrcMem,
1373 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1375 (vt (OpNode RC:$src1,
1376 (mem_frag addr:$src2))))], d>;
1379 let AddedComplexity = 10 in {
1380 let isAsmParserOnly = 1 in {
1381 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1382 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1383 SSEPackedSingle>, VEX_4V;
1384 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1385 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1386 SSEPackedDouble>, OpSize, VEX_4V;
1387 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1388 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1389 SSEPackedSingle>, VEX_4V;
1390 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1391 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1392 SSEPackedDouble>, OpSize, VEX_4V;
1394 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1395 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1396 SSEPackedSingle>, VEX_4V;
1397 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1398 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1399 SSEPackedDouble>, OpSize, VEX_4V;
1400 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1401 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1402 SSEPackedSingle>, VEX_4V;
1403 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1404 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1405 SSEPackedDouble>, OpSize, VEX_4V;
1408 let Constraints = "$src1 = $dst" in {
1409 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1410 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1411 SSEPackedSingle>, TB;
1412 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1413 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1414 SSEPackedDouble>, TB, OpSize;
1415 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1416 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1417 SSEPackedSingle>, TB;
1418 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1419 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1420 SSEPackedDouble>, TB, OpSize;
1421 } // Constraints = "$src1 = $dst"
1422 } // AddedComplexity
1424 //===----------------------------------------------------------------------===//
1425 // SSE 1 & 2 - Extract Floating-Point Sign mask
1426 //===----------------------------------------------------------------------===//
1428 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1429 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1431 def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1432 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1433 [(set GR32:$dst, (Int RC:$src))], d>;
1437 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1438 SSEPackedSingle>, TB;
1439 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1440 SSEPackedDouble>, TB, OpSize;
1442 let isAsmParserOnly = 1 in {
1443 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1444 "movmskps", SSEPackedSingle>, VEX;
1445 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1446 "movmskpd", SSEPackedDouble>, OpSize,
1449 // FIXME: merge with multiclass above when the intrinsics come.
1450 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1451 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1452 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1453 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1456 def VMOVMSKPSYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1457 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1458 def VMOVMSKPDYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1459 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1462 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1463 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1464 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1465 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1469 //===----------------------------------------------------------------------===//
1470 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1471 //===----------------------------------------------------------------------===//
1473 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1474 // names that start with 'Fs'.
1476 // Alias instructions that map fld0 to pxor for sse.
1477 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1478 canFoldAsLoad = 1 in {
1479 // FIXME: Set encoding to pseudo!
1480 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1481 [(set FR32:$dst, fp32imm0)]>,
1482 Requires<[HasSSE1]>, TB, OpSize;
1483 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1484 [(set FR64:$dst, fpimm0)]>,
1485 Requires<[HasSSE2]>, TB, OpSize;
1488 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1489 // bits are disregarded.
1490 let neverHasSideEffects = 1 in {
1491 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1492 "movaps\t{$src, $dst|$dst, $src}", []>;
1493 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1494 "movapd\t{$src, $dst|$dst, $src}", []>;
1497 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1498 // bits are disregarded.
1499 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1500 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1501 "movaps\t{$src, $dst|$dst, $src}",
1502 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1503 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1504 "movapd\t{$src, $dst|$dst, $src}",
1505 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1508 //===----------------------------------------------------------------------===//
1509 // SSE 1 & 2 - Logical Instructions
1510 //===----------------------------------------------------------------------===//
1512 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1514 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1516 let isAsmParserOnly = 1 in {
1517 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1518 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1520 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1521 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1524 let Constraints = "$src1 = $dst" in {
1525 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1526 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1528 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1529 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1533 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1534 let mayLoad = 0 in {
1535 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1536 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1537 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1540 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1541 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1543 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1545 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1546 SDNode OpNode, int HasPat = 0,
1547 list<list<dag>> Pattern = []> {
1548 let isAsmParserOnly = 1, Pattern = []<dag> in {
1549 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1550 !strconcat(OpcodeStr, "ps"), f128mem,
1551 !if(HasPat, Pattern[0], // rr
1552 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1554 !if(HasPat, Pattern[2], // rm
1555 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1556 (memopv2i64 addr:$src2)))]), 0>,
1559 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1560 !strconcat(OpcodeStr, "pd"), f128mem,
1561 !if(HasPat, Pattern[1], // rr
1562 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1565 !if(HasPat, Pattern[3], // rm
1566 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1567 (memopv2i64 addr:$src2)))]), 0>,
1570 let Constraints = "$src1 = $dst" in {
1571 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1572 !strconcat(OpcodeStr, "ps"), f128mem,
1573 !if(HasPat, Pattern[0], // rr
1574 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1576 !if(HasPat, Pattern[2], // rm
1577 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1578 (memopv2i64 addr:$src2)))])>, TB;
1580 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1581 !strconcat(OpcodeStr, "pd"), f128mem,
1582 !if(HasPat, Pattern[1], // rr
1583 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1586 !if(HasPat, Pattern[3], // rm
1587 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1588 (memopv2i64 addr:$src2)))])>,
1593 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1595 let isAsmParserOnly = 1 in {
1596 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1597 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1598 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1600 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1601 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1605 // AVX 256-bit packed logical ops forms
1606 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1607 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1608 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1609 let isCommutable = 0 in
1610 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1612 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1613 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1614 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1615 let isCommutable = 0 in
1616 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1618 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1619 (bc_v2i64 (v4i32 immAllOnesV))),
1622 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1623 (bc_v2i64 (v2f64 VR128:$src2))))],
1625 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1626 (bc_v2i64 (v4i32 immAllOnesV))),
1627 (memopv2i64 addr:$src2))))],
1629 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1630 (memopv2i64 addr:$src2)))]]>;
1632 //===----------------------------------------------------------------------===//
1633 // SSE 1 & 2 - Arithmetic Instructions
1634 //===----------------------------------------------------------------------===//
1636 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1639 /// In addition, we also have a special variant of the scalar form here to
1640 /// represent the associated intrinsic operation. This form is unlike the
1641 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1642 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1644 /// These three forms can each be reg+reg or reg+mem.
1646 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1648 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1649 OpNode, FR32, f32mem, Is2Addr>, XS;
1650 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1651 OpNode, FR64, f64mem, Is2Addr>, XD;
1654 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1656 let mayLoad = 0 in {
1657 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1658 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1659 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1660 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1664 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1666 let mayLoad = 0 in {
1667 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1668 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1669 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1670 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1674 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1676 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1677 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1678 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1679 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1682 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1684 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1685 !strconcat(OpcodeStr, "ps"), "", "_ps", f128mem, memopv4f32,
1686 SSEPackedSingle, Is2Addr>, TB;
1688 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1689 !strconcat(OpcodeStr, "pd"), "2", "_pd", f128mem, memopv2f64,
1690 SSEPackedDouble, Is2Addr>, TB, OpSize;
1693 // Binary Arithmetic instructions
1694 let isAsmParserOnly = 1 in {
1695 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1696 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1697 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1698 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1699 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1700 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1701 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1702 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1704 let isCommutable = 0 in {
1705 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1706 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1707 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1708 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1709 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1710 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1711 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1712 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1713 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1714 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1715 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1716 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1717 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>, VEX_4V;
1718 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1719 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1720 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1721 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1722 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1726 let Constraints = "$src1 = $dst" in {
1727 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1728 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1729 basic_sse12_fp_binop_s_int<0x58, "add">;
1730 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1731 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1732 basic_sse12_fp_binop_s_int<0x59, "mul">;
1734 let isCommutable = 0 in {
1735 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1736 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1737 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1738 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1739 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1740 basic_sse12_fp_binop_s_int<0x5E, "div">;
1741 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1742 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1743 basic_sse12_fp_binop_s_int<0x5F, "max">,
1744 basic_sse12_fp_binop_p_int<0x5F, "max">;
1745 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1746 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1747 basic_sse12_fp_binop_s_int<0x5D, "min">,
1748 basic_sse12_fp_binop_p_int<0x5D, "min">;
1753 /// In addition, we also have a special variant of the scalar form here to
1754 /// represent the associated intrinsic operation. This form is unlike the
1755 /// plain scalar form, in that it takes an entire vector (instead of a
1756 /// scalar) and leaves the top elements undefined.
1758 /// And, we have a special variant form for a full-vector intrinsic form.
1760 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1761 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1762 SDNode OpNode, Intrinsic F32Int> {
1763 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1764 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1765 [(set FR32:$dst, (OpNode FR32:$src))]>;
1766 // For scalar unary operations, fold a load into the operation
1767 // only in OptForSize mode. It eliminates an instruction, but it also
1768 // eliminates a whole-register clobber (the load), so it introduces a
1769 // partial register update condition.
1770 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1771 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1772 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1773 Requires<[HasSSE1, OptForSize]>;
1774 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1775 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1776 [(set VR128:$dst, (F32Int VR128:$src))]>;
1777 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1778 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1779 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1782 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1783 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1784 SDNode OpNode, Intrinsic F32Int> {
1785 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1786 !strconcat(OpcodeStr,
1787 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1788 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1789 !strconcat(OpcodeStr,
1790 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1791 []>, XS, Requires<[HasAVX, OptForSize]>;
1792 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1793 !strconcat(OpcodeStr,
1794 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1795 [(set VR128:$dst, (F32Int VR128:$src))]>;
1796 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1797 !strconcat(OpcodeStr,
1798 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1799 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1802 /// sse1_fp_unop_p - SSE1 unops in packed form.
1803 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1804 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1805 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1806 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1807 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1808 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1809 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1812 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1813 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1814 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1815 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1816 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1817 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1818 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1819 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1822 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1823 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1824 Intrinsic V4F32Int> {
1825 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1826 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1827 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1828 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1829 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1830 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1834 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1835 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1836 SDNode OpNode, Intrinsic F64Int> {
1837 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1838 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1839 [(set FR64:$dst, (OpNode FR64:$src))]>;
1840 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1841 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1842 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1843 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1844 Requires<[HasSSE2, OptForSize]>;
1845 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1847 [(set VR128:$dst, (F64Int VR128:$src))]>;
1848 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1849 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1850 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1853 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1854 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1855 SDNode OpNode, Intrinsic F64Int> {
1856 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1857 !strconcat(OpcodeStr,
1858 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1859 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1860 (ins FR64:$src1, f64mem:$src2),
1861 !strconcat(OpcodeStr,
1862 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1863 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1864 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1865 [(set VR128:$dst, (F64Int VR128:$src))]>;
1866 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1867 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1868 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1871 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1872 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1874 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1875 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1876 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1877 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1878 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1879 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1882 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1883 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1884 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1885 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1886 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1887 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1888 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1889 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1892 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1893 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1894 Intrinsic V2F64Int> {
1895 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1896 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1897 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1898 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1899 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1900 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1903 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1905 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1906 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1909 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1910 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1911 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1912 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1913 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1914 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1917 // Reciprocal approximations. Note that these typically require refinement
1918 // in order to obtain suitable precision.
1919 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1920 int_x86_sse_rsqrt_ss>, VEX_4V;
1921 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1922 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1923 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1925 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1927 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1928 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1929 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1933 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1934 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1935 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1936 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1937 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1938 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1940 // Reciprocal approximations. Note that these typically require refinement
1941 // in order to obtain suitable precision.
1942 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1943 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1944 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1945 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1946 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1947 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1949 // There is no f64 version of the reciprocal approximation instructions.
1951 //===----------------------------------------------------------------------===//
1952 // SSE 1 & 2 - Non-temporal stores
1953 //===----------------------------------------------------------------------===//
1955 let isAsmParserOnly = 1 in {
1956 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1957 (ins i128mem:$dst, VR128:$src),
1958 "movntps\t{$src, $dst|$dst, $src}",
1959 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1960 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1961 (ins i128mem:$dst, VR128:$src),
1962 "movntpd\t{$src, $dst|$dst, $src}",
1963 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1965 let ExeDomain = SSEPackedInt in
1966 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1967 (ins f128mem:$dst, VR128:$src),
1968 "movntdq\t{$src, $dst|$dst, $src}",
1969 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1971 let AddedComplexity = 400 in { // Prefer non-temporal versions
1972 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1973 (ins f128mem:$dst, VR128:$src),
1974 "movntps\t{$src, $dst|$dst, $src}",
1975 [(alignednontemporalstore (v4f32 VR128:$src),
1977 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1978 (ins f128mem:$dst, VR128:$src),
1979 "movntpd\t{$src, $dst|$dst, $src}",
1980 [(alignednontemporalstore (v2f64 VR128:$src),
1982 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1983 (ins f128mem:$dst, VR128:$src),
1984 "movntdq\t{$src, $dst|$dst, $src}",
1985 [(alignednontemporalstore (v2f64 VR128:$src),
1987 let ExeDomain = SSEPackedInt in
1988 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1989 (ins f128mem:$dst, VR128:$src),
1990 "movntdq\t{$src, $dst|$dst, $src}",
1991 [(alignednontemporalstore (v4f32 VR128:$src),
1994 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1995 (ins f256mem:$dst, VR256:$src),
1996 "movntps\t{$src, $dst|$dst, $src}",
1997 [(alignednontemporalstore (v8f32 VR256:$src),
1999 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2000 (ins f256mem:$dst, VR256:$src),
2001 "movntpd\t{$src, $dst|$dst, $src}",
2002 [(alignednontemporalstore (v4f64 VR256:$src),
2004 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2005 (ins f256mem:$dst, VR256:$src),
2006 "movntdq\t{$src, $dst|$dst, $src}",
2007 [(alignednontemporalstore (v4f64 VR256:$src),
2009 let ExeDomain = SSEPackedInt in
2010 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2011 (ins f256mem:$dst, VR256:$src),
2012 "movntdq\t{$src, $dst|$dst, $src}",
2013 [(alignednontemporalstore (v8f32 VR256:$src),
2018 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2019 "movntps\t{$src, $dst|$dst, $src}",
2020 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2021 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2022 "movntpd\t{$src, $dst|$dst, $src}",
2023 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2025 let ExeDomain = SSEPackedInt in
2026 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2027 "movntdq\t{$src, $dst|$dst, $src}",
2028 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2030 let AddedComplexity = 400 in { // Prefer non-temporal versions
2031 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2032 "movntps\t{$src, $dst|$dst, $src}",
2033 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2034 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2035 "movntpd\t{$src, $dst|$dst, $src}",
2036 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2038 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2039 "movntdq\t{$src, $dst|$dst, $src}",
2040 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2042 let ExeDomain = SSEPackedInt in
2043 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2044 "movntdq\t{$src, $dst|$dst, $src}",
2045 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2047 // There is no AVX form for instructions below this point
2048 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2049 "movnti\t{$src, $dst|$dst, $src}",
2050 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2051 TB, Requires<[HasSSE2]>;
2053 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2054 "movnti\t{$src, $dst|$dst, $src}",
2055 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2056 TB, Requires<[HasSSE2]>;
2059 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2060 "movnti\t{$src, $dst|$dst, $src}",
2061 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2062 TB, Requires<[HasSSE2]>;
2064 //===----------------------------------------------------------------------===//
2065 // SSE 1 & 2 - Misc Instructions (No AVX form)
2066 //===----------------------------------------------------------------------===//
2068 // Prefetch intrinsic.
2069 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2070 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2071 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2072 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2073 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2074 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2075 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2076 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2078 // Load, store, and memory fence
2079 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2080 TB, Requires<[HasSSE1]>;
2081 def : Pat<(X86SFence), (SFENCE)>;
2083 // Alias instructions that map zero vector to pxor / xorp* for sse.
2084 // We set canFoldAsLoad because this can be converted to a constant-pool
2085 // load of an all-zeros value if folding it would be beneficial.
2086 // FIXME: Change encoding to pseudo!
2087 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2088 isCodeGenOnly = 1 in {
2089 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2090 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2091 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2092 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2093 let ExeDomain = SSEPackedInt in
2094 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2095 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2098 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2099 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2100 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2102 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2103 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2105 //===----------------------------------------------------------------------===//
2106 // SSE 1 & 2 - Load/Store XCSR register
2107 //===----------------------------------------------------------------------===//
2109 let isAsmParserOnly = 1 in {
2110 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2111 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2112 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2113 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2116 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2117 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2118 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2119 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2121 //===---------------------------------------------------------------------===//
2122 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2123 //===---------------------------------------------------------------------===//
2125 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2127 let isAsmParserOnly = 1 in {
2128 let neverHasSideEffects = 1 in {
2129 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2130 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2131 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2132 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2134 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2135 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2136 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2137 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2139 let canFoldAsLoad = 1, mayLoad = 1 in {
2140 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2141 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2142 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2143 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2144 let Predicates = [HasAVX] in {
2145 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2146 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2147 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2148 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2152 let mayStore = 1 in {
2153 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2154 (ins i128mem:$dst, VR128:$src),
2155 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2156 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2157 (ins i256mem:$dst, VR256:$src),
2158 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2159 let Predicates = [HasAVX] in {
2160 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2161 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2162 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2163 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2168 let neverHasSideEffects = 1 in
2169 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2170 "movdqa\t{$src, $dst|$dst, $src}", []>;
2172 let canFoldAsLoad = 1, mayLoad = 1 in {
2173 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2174 "movdqa\t{$src, $dst|$dst, $src}",
2175 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2176 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2177 "movdqu\t{$src, $dst|$dst, $src}",
2178 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2179 XS, Requires<[HasSSE2]>;
2182 let mayStore = 1 in {
2183 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2184 "movdqa\t{$src, $dst|$dst, $src}",
2185 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2186 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2187 "movdqu\t{$src, $dst|$dst, $src}",
2188 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2189 XS, Requires<[HasSSE2]>;
2192 // Intrinsic forms of MOVDQU load and store
2193 let isAsmParserOnly = 1 in {
2194 let canFoldAsLoad = 1 in
2195 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2196 "vmovdqu\t{$src, $dst|$dst, $src}",
2197 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2198 XS, VEX, Requires<[HasAVX]>;
2199 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2200 "vmovdqu\t{$src, $dst|$dst, $src}",
2201 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2202 XS, VEX, Requires<[HasAVX]>;
2205 let canFoldAsLoad = 1 in
2206 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2207 "movdqu\t{$src, $dst|$dst, $src}",
2208 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2209 XS, Requires<[HasSSE2]>;
2210 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2211 "movdqu\t{$src, $dst|$dst, $src}",
2212 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2213 XS, Requires<[HasSSE2]>;
2215 } // ExeDomain = SSEPackedInt
2217 //===---------------------------------------------------------------------===//
2218 // SSE2 - Packed Integer Arithmetic Instructions
2219 //===---------------------------------------------------------------------===//
2221 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2223 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2224 bit IsCommutable = 0, bit Is2Addr = 1> {
2225 let isCommutable = IsCommutable in
2226 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2227 (ins VR128:$src1, VR128:$src2),
2229 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2230 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2231 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2232 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2233 (ins VR128:$src1, i128mem:$src2),
2235 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2236 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2237 [(set VR128:$dst, (IntId VR128:$src1,
2238 (bitconvert (memopv2i64 addr:$src2))))]>;
2241 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2242 string OpcodeStr, Intrinsic IntId,
2243 Intrinsic IntId2, bit Is2Addr = 1> {
2244 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2245 (ins VR128:$src1, VR128:$src2),
2247 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2248 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2249 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2250 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2251 (ins VR128:$src1, i128mem:$src2),
2253 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2254 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2255 [(set VR128:$dst, (IntId VR128:$src1,
2256 (bitconvert (memopv2i64 addr:$src2))))]>;
2257 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2258 (ins VR128:$src1, i32i8imm:$src2),
2260 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2261 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2262 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2265 /// PDI_binop_rm - Simple SSE2 binary operator.
2266 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2267 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2268 let isCommutable = IsCommutable in
2269 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2270 (ins VR128:$src1, VR128:$src2),
2272 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2273 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2274 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2275 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2276 (ins VR128:$src1, i128mem:$src2),
2278 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2279 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2280 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2281 (bitconvert (memopv2i64 addr:$src2)))))]>;
2284 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2286 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2287 /// to collapse (bitconvert VT to VT) into its operand.
2289 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2290 bit IsCommutable = 0, bit Is2Addr = 1> {
2291 let isCommutable = IsCommutable in
2292 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2293 (ins VR128:$src1, VR128:$src2),
2295 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2296 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2297 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2298 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2299 (ins VR128:$src1, i128mem:$src2),
2301 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2302 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2303 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2306 } // ExeDomain = SSEPackedInt
2308 // 128-bit Integer Arithmetic
2310 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2311 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2312 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2313 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2314 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2315 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2316 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2317 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2318 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2319 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2322 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2324 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2326 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2328 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2330 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2332 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2334 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2336 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2338 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2340 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2342 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2344 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2346 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2348 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2350 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2352 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2354 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2356 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2358 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2362 let Constraints = "$src1 = $dst" in {
2363 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2364 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2365 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2366 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2367 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2368 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2369 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2370 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2371 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2374 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2375 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2376 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2377 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2378 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2379 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2380 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2381 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2382 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2383 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2384 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2385 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2386 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2387 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2388 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2389 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2390 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2391 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2392 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2394 } // Constraints = "$src1 = $dst"
2396 //===---------------------------------------------------------------------===//
2397 // SSE2 - Packed Integer Logical Instructions
2398 //===---------------------------------------------------------------------===//
2400 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2401 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2402 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2404 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2405 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2407 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2408 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2411 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2412 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2414 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2415 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2417 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2418 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2421 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2422 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2424 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2425 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2428 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2429 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2430 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2432 let ExeDomain = SSEPackedInt in {
2433 let neverHasSideEffects = 1 in {
2434 // 128-bit logical shifts.
2435 def VPSLLDQri : PDIi8<0x73, MRM7r,
2436 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2437 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2439 def VPSRLDQri : PDIi8<0x73, MRM3r,
2440 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2441 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2443 // PSRADQri doesn't exist in SSE[1-3].
2445 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2446 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2447 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2448 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2449 VR128:$src2)))]>, VEX_4V;
2451 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2452 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2453 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2454 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2455 (memopv2i64 addr:$src2))))]>,
2460 let Constraints = "$src1 = $dst" in {
2461 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2462 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2463 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2464 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2465 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2466 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2468 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2469 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2470 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2471 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2472 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2473 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2475 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2476 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2477 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2478 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2480 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2481 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2482 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2484 let ExeDomain = SSEPackedInt in {
2485 let neverHasSideEffects = 1 in {
2486 // 128-bit logical shifts.
2487 def PSLLDQri : PDIi8<0x73, MRM7r,
2488 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2489 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2490 def PSRLDQri : PDIi8<0x73, MRM3r,
2491 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2492 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2493 // PSRADQri doesn't exist in SSE[1-3].
2495 def PANDNrr : PDI<0xDF, MRMSrcReg,
2496 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2497 "pandn\t{$src2, $dst|$dst, $src2}",
2498 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2501 def PANDNrm : PDI<0xDF, MRMSrcMem,
2502 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2503 "pandn\t{$src2, $dst|$dst, $src2}",
2504 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2505 (memopv2i64 addr:$src2))))]>;
2507 } // Constraints = "$src1 = $dst"
2509 let Predicates = [HasAVX] in {
2510 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2511 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2512 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2513 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2514 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2515 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2516 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2517 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2518 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2519 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2521 // Shift up / down and insert zero's.
2522 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2523 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2524 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2525 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2528 let Predicates = [HasSSE2] in {
2529 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2530 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2531 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2532 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2533 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2534 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2535 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2536 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2537 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2538 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2540 // Shift up / down and insert zero's.
2541 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2542 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2543 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2544 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2547 //===---------------------------------------------------------------------===//
2548 // SSE2 - Packed Integer Comparison Instructions
2549 //===---------------------------------------------------------------------===//
2551 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2552 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2554 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2556 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2558 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2560 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2562 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2566 let Constraints = "$src1 = $dst" in {
2567 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2568 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2569 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2570 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2571 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2572 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2573 } // Constraints = "$src1 = $dst"
2575 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2576 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2577 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2578 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2579 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2580 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2581 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2582 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2583 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2584 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2585 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2586 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2588 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2589 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2590 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2591 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2592 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2593 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2594 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2595 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2596 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2597 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2598 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2599 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2601 //===---------------------------------------------------------------------===//
2602 // SSE2 - Packed Integer Pack Instructions
2603 //===---------------------------------------------------------------------===//
2605 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2606 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2608 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2610 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2614 let Constraints = "$src1 = $dst" in {
2615 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2616 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2617 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2618 } // Constraints = "$src1 = $dst"
2620 //===---------------------------------------------------------------------===//
2621 // SSE2 - Packed Integer Shuffle Instructions
2622 //===---------------------------------------------------------------------===//
2624 let ExeDomain = SSEPackedInt in {
2625 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2627 def ri : Ii8<0x70, MRMSrcReg,
2628 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2629 !strconcat(OpcodeStr,
2630 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2631 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2633 def mi : Ii8<0x70, MRMSrcMem,
2634 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2635 !strconcat(OpcodeStr,
2636 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2637 [(set VR128:$dst, (vt (pshuf_frag:$src2
2638 (bc_frag (memopv2i64 addr:$src1)),
2641 } // ExeDomain = SSEPackedInt
2643 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2644 let AddedComplexity = 5 in
2645 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2648 // SSE2 with ImmT == Imm8 and XS prefix.
2649 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2652 // SSE2 with ImmT == Imm8 and XD prefix.
2653 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2657 let Predicates = [HasSSE2] in {
2658 let AddedComplexity = 5 in
2659 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2661 // SSE2 with ImmT == Imm8 and XS prefix.
2662 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2664 // SSE2 with ImmT == Imm8 and XD prefix.
2665 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2668 //===---------------------------------------------------------------------===//
2669 // SSE2 - Packed Integer Unpack Instructions
2670 //===---------------------------------------------------------------------===//
2672 let ExeDomain = SSEPackedInt in {
2673 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2674 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2675 def rr : PDI<opc, MRMSrcReg,
2676 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2678 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2679 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2680 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2681 def rm : PDI<opc, MRMSrcMem,
2682 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2684 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2685 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2686 [(set VR128:$dst, (unp_frag VR128:$src1,
2687 (bc_frag (memopv2i64
2691 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2692 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2694 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2696 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2699 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2700 /// knew to collapse (bitconvert VT to VT) into its operand.
2701 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2702 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2703 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2705 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2706 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2707 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2708 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2710 (v2i64 (unpckl VR128:$src1,
2711 (memopv2i64 addr:$src2))))]>, VEX_4V;
2713 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2715 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2717 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2720 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2721 /// knew to collapse (bitconvert VT to VT) into its operand.
2722 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2723 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2724 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2726 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2727 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2728 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2729 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2731 (v2i64 (unpckh VR128:$src1,
2732 (memopv2i64 addr:$src2))))]>, VEX_4V;
2735 let Constraints = "$src1 = $dst" in {
2736 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2737 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2738 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2740 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2741 /// knew to collapse (bitconvert VT to VT) into its operand.
2742 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2743 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2744 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2746 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2747 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2748 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2749 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2751 (v2i64 (unpckl VR128:$src1,
2752 (memopv2i64 addr:$src2))))]>;
2754 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2755 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2756 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2758 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2759 /// knew to collapse (bitconvert VT to VT) into its operand.
2760 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2761 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2762 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2764 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2765 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2766 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2767 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2769 (v2i64 (unpckh VR128:$src1,
2770 (memopv2i64 addr:$src2))))]>;
2773 } // ExeDomain = SSEPackedInt
2775 //===---------------------------------------------------------------------===//
2776 // SSE2 - Packed Integer Extract and Insert
2777 //===---------------------------------------------------------------------===//
2779 let ExeDomain = SSEPackedInt in {
2780 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2781 def rri : Ii8<0xC4, MRMSrcReg,
2782 (outs VR128:$dst), (ins VR128:$src1,
2783 GR32:$src2, i32i8imm:$src3),
2785 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2786 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2788 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2789 def rmi : Ii8<0xC4, MRMSrcMem,
2790 (outs VR128:$dst), (ins VR128:$src1,
2791 i16mem:$src2, i32i8imm:$src3),
2793 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2794 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2796 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2801 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2802 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2803 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2804 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2805 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2806 imm:$src2))]>, OpSize, VEX;
2807 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2808 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2809 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2810 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2814 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2815 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2816 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2817 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2818 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2819 []>, OpSize, VEX_4V;
2822 let Constraints = "$src1 = $dst" in
2823 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2825 } // ExeDomain = SSEPackedInt
2827 //===---------------------------------------------------------------------===//
2828 // SSE2 - Packed Mask Creation
2829 //===---------------------------------------------------------------------===//
2831 let ExeDomain = SSEPackedInt in {
2833 let isAsmParserOnly = 1 in {
2834 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2835 "pmovmskb\t{$src, $dst|$dst, $src}",
2836 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2837 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2838 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2840 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2841 "pmovmskb\t{$src, $dst|$dst, $src}",
2842 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2844 } // ExeDomain = SSEPackedInt
2846 //===---------------------------------------------------------------------===//
2847 // SSE2 - Conditional Store
2848 //===---------------------------------------------------------------------===//
2850 let ExeDomain = SSEPackedInt in {
2852 let isAsmParserOnly = 1 in {
2854 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2855 (ins VR128:$src, VR128:$mask),
2856 "maskmovdqu\t{$mask, $src|$src, $mask}",
2857 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2859 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2860 (ins VR128:$src, VR128:$mask),
2861 "maskmovdqu\t{$mask, $src|$src, $mask}",
2862 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2866 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2867 "maskmovdqu\t{$mask, $src|$src, $mask}",
2868 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2870 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2871 "maskmovdqu\t{$mask, $src|$src, $mask}",
2872 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2874 } // ExeDomain = SSEPackedInt
2876 //===---------------------------------------------------------------------===//
2877 // SSE2 - Move Doubleword
2878 //===---------------------------------------------------------------------===//
2880 // Move Int Doubleword to Packed Double Int
2881 let isAsmParserOnly = 1 in {
2882 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2883 "movd\t{$src, $dst|$dst, $src}",
2885 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2886 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2887 "movd\t{$src, $dst|$dst, $src}",
2889 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2892 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2893 "movd\t{$src, $dst|$dst, $src}",
2895 (v4i32 (scalar_to_vector GR32:$src)))]>;
2896 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2897 "movd\t{$src, $dst|$dst, $src}",
2899 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2902 // Move Int Doubleword to Single Scalar
2903 let isAsmParserOnly = 1 in {
2904 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2905 "movd\t{$src, $dst|$dst, $src}",
2906 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2908 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2909 "movd\t{$src, $dst|$dst, $src}",
2910 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2913 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2914 "movd\t{$src, $dst|$dst, $src}",
2915 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2917 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2918 "movd\t{$src, $dst|$dst, $src}",
2919 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2921 // Move Packed Doubleword Int to Packed Double Int
2922 let isAsmParserOnly = 1 in {
2923 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2924 "movd\t{$src, $dst|$dst, $src}",
2925 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2927 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2928 (ins i32mem:$dst, VR128:$src),
2929 "movd\t{$src, $dst|$dst, $src}",
2930 [(store (i32 (vector_extract (v4i32 VR128:$src),
2931 (iPTR 0))), addr:$dst)]>, VEX;
2933 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2934 "movd\t{$src, $dst|$dst, $src}",
2935 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2937 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2938 "movd\t{$src, $dst|$dst, $src}",
2939 [(store (i32 (vector_extract (v4i32 VR128:$src),
2940 (iPTR 0))), addr:$dst)]>;
2942 // Move Scalar Single to Double Int
2943 let isAsmParserOnly = 1 in {
2944 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2945 "movd\t{$src, $dst|$dst, $src}",
2946 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2947 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2948 "movd\t{$src, $dst|$dst, $src}",
2949 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2951 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2953 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2954 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2955 "movd\t{$src, $dst|$dst, $src}",
2956 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2958 // movd / movq to XMM register zero-extends
2959 let AddedComplexity = 15, isAsmParserOnly = 1 in {
2960 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2961 "movd\t{$src, $dst|$dst, $src}",
2962 [(set VR128:$dst, (v4i32 (X86vzmovl
2963 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2965 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2966 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2967 [(set VR128:$dst, (v2i64 (X86vzmovl
2968 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2971 let AddedComplexity = 15 in {
2972 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2973 "movd\t{$src, $dst|$dst, $src}",
2974 [(set VR128:$dst, (v4i32 (X86vzmovl
2975 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2976 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2977 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2978 [(set VR128:$dst, (v2i64 (X86vzmovl
2979 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2982 let AddedComplexity = 20 in {
2983 let isAsmParserOnly = 1 in
2984 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2985 "movd\t{$src, $dst|$dst, $src}",
2987 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2988 (loadi32 addr:$src))))))]>,
2990 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2991 "movd\t{$src, $dst|$dst, $src}",
2993 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2994 (loadi32 addr:$src))))))]>;
2996 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2997 (MOVZDI2PDIrm addr:$src)>;
2998 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2999 (MOVZDI2PDIrm addr:$src)>;
3000 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3001 (MOVZDI2PDIrm addr:$src)>;
3004 //===---------------------------------------------------------------------===//
3005 // SSE2 - Move Quadword
3006 //===---------------------------------------------------------------------===//
3008 // Move Quadword Int to Packed Quadword Int
3009 let isAsmParserOnly = 1 in
3010 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3011 "vmovq\t{$src, $dst|$dst, $src}",
3013 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3014 VEX, Requires<[HasAVX]>;
3015 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3016 "movq\t{$src, $dst|$dst, $src}",
3018 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3019 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3021 // Move Packed Quadword Int to Quadword Int
3022 let isAsmParserOnly = 1 in
3023 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3024 "movq\t{$src, $dst|$dst, $src}",
3025 [(store (i64 (vector_extract (v2i64 VR128:$src),
3026 (iPTR 0))), addr:$dst)]>, VEX;
3027 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3028 "movq\t{$src, $dst|$dst, $src}",
3029 [(store (i64 (vector_extract (v2i64 VR128:$src),
3030 (iPTR 0))), addr:$dst)]>;
3032 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3033 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3035 // Store / copy lower 64-bits of a XMM register.
3036 let isAsmParserOnly = 1 in
3037 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3038 "movq\t{$src, $dst|$dst, $src}",
3039 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3040 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3041 "movq\t{$src, $dst|$dst, $src}",
3042 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3044 let AddedComplexity = 20, isAsmParserOnly = 1 in
3045 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3046 "vmovq\t{$src, $dst|$dst, $src}",
3048 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3049 (loadi64 addr:$src))))))]>,
3050 XS, VEX, Requires<[HasAVX]>;
3052 let AddedComplexity = 20 in {
3053 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3054 "movq\t{$src, $dst|$dst, $src}",
3056 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3057 (loadi64 addr:$src))))))]>,
3058 XS, Requires<[HasSSE2]>;
3060 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3061 (MOVZQI2PQIrm addr:$src)>;
3062 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3063 (MOVZQI2PQIrm addr:$src)>;
3064 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3067 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3068 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3069 let isAsmParserOnly = 1, AddedComplexity = 15 in
3070 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3071 "vmovq\t{$src, $dst|$dst, $src}",
3072 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3073 XS, VEX, Requires<[HasAVX]>;
3074 let AddedComplexity = 15 in
3075 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3076 "movq\t{$src, $dst|$dst, $src}",
3077 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3078 XS, Requires<[HasSSE2]>;
3080 let AddedComplexity = 20, isAsmParserOnly = 1 in
3081 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3082 "vmovq\t{$src, $dst|$dst, $src}",
3083 [(set VR128:$dst, (v2i64 (X86vzmovl
3084 (loadv2i64 addr:$src))))]>,
3085 XS, VEX, Requires<[HasAVX]>;
3086 let AddedComplexity = 20 in {
3087 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3088 "movq\t{$src, $dst|$dst, $src}",
3089 [(set VR128:$dst, (v2i64 (X86vzmovl
3090 (loadv2i64 addr:$src))))]>,
3091 XS, Requires<[HasSSE2]>;
3093 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3094 (MOVZPQILo2PQIrm addr:$src)>;
3097 // Instructions to match in the assembler
3098 let isAsmParserOnly = 1 in {
3099 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3100 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3101 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3102 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3103 // Recognize "movd" with GR64 destination, but encode as a "movq"
3104 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3105 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3108 // Instructions for the disassembler
3109 // xr = XMM register
3112 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3113 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3114 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3115 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3116 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3118 //===---------------------------------------------------------------------===//
3119 // SSE2 - Misc Instructions
3120 //===---------------------------------------------------------------------===//
3123 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3124 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3125 TB, Requires<[HasSSE2]>;
3127 // Load, store, and memory fence
3128 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3129 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3130 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3131 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3132 def : Pat<(X86LFence), (LFENCE)>;
3133 def : Pat<(X86MFence), (MFENCE)>;
3136 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3137 // was introduced with SSE2, it's backward compatible.
3138 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3140 // Alias instructions that map zero vector to pxor / xorp* for sse.
3141 // We set canFoldAsLoad because this can be converted to a constant-pool
3142 // load of an all-ones value if folding it would be beneficial.
3143 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3144 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3145 // FIXME: Change encoding to pseudo.
3146 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3147 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3149 //===---------------------------------------------------------------------===//
3150 // SSE3 - Conversion Instructions
3151 //===---------------------------------------------------------------------===//
3153 // Convert Packed Double FP to Packed DW Integers
3154 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3155 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3156 // register, but the same isn't true when using memory operands instead.
3157 // Provide other assembly rr and rm forms to address this explicitly.
3158 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3159 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3160 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3161 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3164 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3165 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3166 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3167 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3170 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3171 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3172 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3173 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3176 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3177 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3178 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3179 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3181 // Convert Packed DW Integers to Packed Double FP
3182 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3183 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3184 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3185 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3186 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3187 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3188 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3189 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3190 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3193 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3194 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3195 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3196 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3198 //===---------------------------------------------------------------------===//
3199 // SSE3 - Move Instructions
3200 //===---------------------------------------------------------------------===//
3202 // Replicate Single FP
3203 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3204 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3205 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3206 [(set VR128:$dst, (v4f32 (rep_frag
3207 VR128:$src, (undef))))]>;
3208 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3209 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3210 [(set VR128:$dst, (rep_frag
3211 (memopv4f32 addr:$src), (undef)))]>;
3214 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3216 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3217 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3218 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3219 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3222 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3223 // FIXME: Merge above classes when we have patterns for the ymm version
3224 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3225 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3226 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3227 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3229 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3230 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3232 // Replicate Double FP
3233 multiclass sse3_replicate_dfp<string OpcodeStr> {
3234 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3235 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3236 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3237 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3238 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3240 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3244 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3245 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3246 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3248 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3249 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3253 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3254 // FIXME: Merge above classes when we have patterns for the ymm version
3255 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3256 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3258 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3260 // Move Unaligned Integer
3261 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3262 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3263 "vlddqu\t{$src, $dst|$dst, $src}",
3264 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3265 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3266 "vlddqu\t{$src, $dst|$dst, $src}", []>, VEX;
3268 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3269 "lddqu\t{$src, $dst|$dst, $src}",
3270 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3272 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3274 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3276 // Several Move patterns
3277 let AddedComplexity = 5 in {
3278 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3279 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3280 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3281 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3282 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3283 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3284 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3285 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3288 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3289 let AddedComplexity = 15 in
3290 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3291 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3292 let AddedComplexity = 20 in
3293 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3294 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3296 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3297 let AddedComplexity = 15 in
3298 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3299 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3300 let AddedComplexity = 20 in
3301 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3302 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3304 //===---------------------------------------------------------------------===//
3305 // SSE3 - Arithmetic
3306 //===---------------------------------------------------------------------===//
3308 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3309 X86MemOperand x86memop, bit Is2Addr = 1> {
3310 def rr : I<0xD0, MRMSrcReg,
3311 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3313 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3314 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3315 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3316 def rm : I<0xD0, MRMSrcMem,
3317 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3319 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3321 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3324 let isAsmParserOnly = 1, Predicates = [HasAVX],
3325 ExeDomain = SSEPackedDouble in {
3326 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3327 f128mem, 0>, XD, VEX_4V;
3328 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3329 f128mem, 0>, OpSize, VEX_4V;
3330 let Pattern = []<dag> in {
3331 defm VADDSUBPSY : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR256,
3332 f256mem, 0>, XD, VEX_4V;
3333 defm VADDSUBPDY : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR256,
3334 f256mem, 0>, OpSize, VEX_4V;
3337 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3338 ExeDomain = SSEPackedDouble in {
3339 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3341 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3342 f128mem>, TB, OpSize;
3345 //===---------------------------------------------------------------------===//
3346 // SSE3 Instructions
3347 //===---------------------------------------------------------------------===//
3350 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3351 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3352 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3354 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3355 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3356 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3358 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3360 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3362 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3364 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3365 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3366 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3368 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3369 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3370 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3372 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3374 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3375 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3376 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3379 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3380 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3381 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3382 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3383 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3384 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3385 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3386 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3387 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3388 let Pattern = []<dag> in {
3389 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3390 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3391 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3392 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3393 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3394 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3395 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3396 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3400 let Constraints = "$src1 = $dst" in {
3401 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3402 int_x86_sse3_hadd_ps>;
3403 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3404 int_x86_sse3_hadd_pd>;
3405 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3406 int_x86_sse3_hsub_ps>;
3407 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3408 int_x86_sse3_hsub_pd>;
3411 //===---------------------------------------------------------------------===//
3412 // SSSE3 - Packed Absolute Instructions
3413 //===---------------------------------------------------------------------===//
3415 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3416 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3417 PatFrag mem_frag64, PatFrag mem_frag128,
3418 Intrinsic IntId64, Intrinsic IntId128> {
3419 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
3420 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3421 [(set VR64:$dst, (IntId64 VR64:$src))]>;
3423 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
3424 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3426 (IntId64 (bitconvert (mem_frag64 addr:$src))))]>;
3428 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3430 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3431 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3434 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3436 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3439 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3442 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3443 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv8i8, memopv16i8,
3444 int_x86_ssse3_pabs_b,
3445 int_x86_ssse3_pabs_b_128>, VEX;
3446 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv4i16, memopv8i16,
3447 int_x86_ssse3_pabs_w,
3448 int_x86_ssse3_pabs_w_128>, VEX;
3449 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv2i32, memopv4i32,
3450 int_x86_ssse3_pabs_d,
3451 int_x86_ssse3_pabs_d_128>, VEX;
3454 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv8i8, memopv16i8,
3455 int_x86_ssse3_pabs_b,
3456 int_x86_ssse3_pabs_b_128>;
3457 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv4i16, memopv8i16,
3458 int_x86_ssse3_pabs_w,
3459 int_x86_ssse3_pabs_w_128>;
3460 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv2i32, memopv4i32,
3461 int_x86_ssse3_pabs_d,
3462 int_x86_ssse3_pabs_d_128>;
3464 //===---------------------------------------------------------------------===//
3465 // SSSE3 - Packed Binary Operator Instructions
3466 //===---------------------------------------------------------------------===//
3468 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3469 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3470 PatFrag mem_frag64, PatFrag mem_frag128,
3471 Intrinsic IntId64, Intrinsic IntId128,
3473 let isCommutable = 1 in
3474 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
3475 (ins VR64:$src1, VR64:$src2),
3477 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3478 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3479 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
3480 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
3481 (ins VR64:$src1, i64mem:$src2),
3483 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3484 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3486 (IntId64 VR64:$src1,
3487 (bitconvert (memopv8i8 addr:$src2))))]>;
3489 let isCommutable = 1 in
3490 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3491 (ins VR128:$src1, VR128:$src2),
3493 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3494 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3495 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3497 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3498 (ins VR128:$src1, i128mem:$src2),
3500 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3501 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3503 (IntId128 VR128:$src1,
3504 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3507 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3508 let isCommutable = 0 in {
3509 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv4i16, memopv8i16,
3510 int_x86_ssse3_phadd_w,
3511 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3512 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv2i32, memopv4i32,
3513 int_x86_ssse3_phadd_d,
3514 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3515 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv4i16, memopv8i16,
3516 int_x86_ssse3_phadd_sw,
3517 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3518 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv4i16, memopv8i16,
3519 int_x86_ssse3_phsub_w,
3520 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3521 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv2i32, memopv4i32,
3522 int_x86_ssse3_phsub_d,
3523 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3524 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv4i16, memopv8i16,
3525 int_x86_ssse3_phsub_sw,
3526 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3527 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv8i8, memopv16i8,
3528 int_x86_ssse3_pmadd_ub_sw,
3529 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3530 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv8i8, memopv16i8,
3531 int_x86_ssse3_pshuf_b,
3532 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3533 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv8i8, memopv16i8,
3534 int_x86_ssse3_psign_b,
3535 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3536 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv4i16, memopv8i16,
3537 int_x86_ssse3_psign_w,
3538 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3539 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv2i32, memopv4i32,
3540 int_x86_ssse3_psign_d,
3541 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3543 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv4i16, memopv8i16,
3544 int_x86_ssse3_pmul_hr_sw,
3545 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3548 // None of these have i8 immediate fields.
3549 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3550 let isCommutable = 0 in {
3551 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv4i16, memopv8i16,
3552 int_x86_ssse3_phadd_w,
3553 int_x86_ssse3_phadd_w_128>;
3554 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv2i32, memopv4i32,
3555 int_x86_ssse3_phadd_d,
3556 int_x86_ssse3_phadd_d_128>;
3557 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv4i16, memopv8i16,
3558 int_x86_ssse3_phadd_sw,
3559 int_x86_ssse3_phadd_sw_128>;
3560 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv4i16, memopv8i16,
3561 int_x86_ssse3_phsub_w,
3562 int_x86_ssse3_phsub_w_128>;
3563 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv2i32, memopv4i32,
3564 int_x86_ssse3_phsub_d,
3565 int_x86_ssse3_phsub_d_128>;
3566 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv4i16, memopv8i16,
3567 int_x86_ssse3_phsub_sw,
3568 int_x86_ssse3_phsub_sw_128>;
3569 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv8i8, memopv16i8,
3570 int_x86_ssse3_pmadd_ub_sw,
3571 int_x86_ssse3_pmadd_ub_sw_128>;
3572 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv8i8, memopv16i8,
3573 int_x86_ssse3_pshuf_b,
3574 int_x86_ssse3_pshuf_b_128>;
3575 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv8i8, memopv16i8,
3576 int_x86_ssse3_psign_b,
3577 int_x86_ssse3_psign_b_128>;
3578 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv4i16, memopv8i16,
3579 int_x86_ssse3_psign_w,
3580 int_x86_ssse3_psign_w_128>;
3581 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv2i32, memopv4i32,
3582 int_x86_ssse3_psign_d,
3583 int_x86_ssse3_psign_d_128>;
3585 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv4i16, memopv8i16,
3586 int_x86_ssse3_pmul_hr_sw,
3587 int_x86_ssse3_pmul_hr_sw_128>;
3590 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3591 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3592 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3593 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3595 //===---------------------------------------------------------------------===//
3596 // SSSE3 - Packed Align Instruction Patterns
3597 //===---------------------------------------------------------------------===//
3599 multiclass sse3_palign<string asm, bit Is2Addr = 1> {
3600 def R64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
3601 (ins VR64:$src1, VR64:$src2, i8imm:$src3),
3603 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3605 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3607 def R64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
3608 (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
3610 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3612 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3615 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3616 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3618 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3620 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3622 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3623 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3625 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3627 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3631 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3632 defm VPALIGN : sse3_palign<"vpalignr", 0>, VEX_4V;
3633 let Constraints = "$src1 = $dst" in
3634 defm PALIGN : sse3_palign<"palignr">;
3636 let AddedComplexity = 5 in {
3638 def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
3639 (PALIGNR64rr VR64:$src2, VR64:$src1,
3640 (SHUFFLE_get_palign_imm VR64:$src3))>,
3641 Requires<[HasSSSE3]>;
3642 def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
3643 (PALIGNR64rr VR64:$src2, VR64:$src1,
3644 (SHUFFLE_get_palign_imm VR64:$src3))>,
3645 Requires<[HasSSSE3]>;
3646 def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
3647 (PALIGNR64rr VR64:$src2, VR64:$src1,
3648 (SHUFFLE_get_palign_imm VR64:$src3))>,
3649 Requires<[HasSSSE3]>;
3650 def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
3651 (PALIGNR64rr VR64:$src2, VR64:$src1,
3652 (SHUFFLE_get_palign_imm VR64:$src3))>,
3653 Requires<[HasSSSE3]>;
3655 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3656 (PALIGNR128rr VR128:$src2, VR128:$src1,
3657 (SHUFFLE_get_palign_imm VR128:$src3))>,
3658 Requires<[HasSSSE3]>;
3659 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3660 (PALIGNR128rr VR128:$src2, VR128:$src1,
3661 (SHUFFLE_get_palign_imm VR128:$src3))>,
3662 Requires<[HasSSSE3]>;
3663 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3664 (PALIGNR128rr VR128:$src2, VR128:$src1,
3665 (SHUFFLE_get_palign_imm VR128:$src3))>,
3666 Requires<[HasSSSE3]>;
3667 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3668 (PALIGNR128rr VR128:$src2, VR128:$src1,
3669 (SHUFFLE_get_palign_imm VR128:$src3))>,
3670 Requires<[HasSSSE3]>;
3673 //===---------------------------------------------------------------------===//
3674 // SSSE3 Misc Instructions
3675 //===---------------------------------------------------------------------===//
3677 // Thread synchronization
3678 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3679 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3680 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3681 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3683 //===---------------------------------------------------------------------===//
3684 // Non-Instruction Patterns
3685 //===---------------------------------------------------------------------===//
3687 // extload f32 -> f64. This matches load+fextend because we have a hack in
3688 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3690 // Since these loads aren't folded into the fextend, we have to match it
3692 let Predicates = [HasSSE2] in
3693 def : Pat<(fextend (loadf32 addr:$src)),
3694 (CVTSS2SDrm addr:$src)>;
3697 let Predicates = [HasSSE2] in {
3698 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3699 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3700 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3701 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3702 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3703 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3704 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3705 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3706 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3707 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3708 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3709 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3710 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3711 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3712 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3713 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3714 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3715 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3716 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3717 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3718 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3719 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3720 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3721 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3722 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3723 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3724 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3725 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3726 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3727 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3730 // Move scalar to XMM zero-extended
3731 // movd to XMM register zero-extends
3732 let AddedComplexity = 15 in {
3733 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3734 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3735 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3736 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3737 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3738 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3739 (MOVSSrr (v4f32 (V_SET0PS)),
3740 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3741 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3742 (MOVSSrr (v4i32 (V_SET0PI)),
3743 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3746 // Splat v2f64 / v2i64
3747 let AddedComplexity = 10 in {
3748 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3749 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3750 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3751 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3752 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3753 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3754 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3755 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3758 // Special unary SHUFPSrri case.
3759 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3760 (SHUFPSrri VR128:$src1, VR128:$src1,
3761 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3762 let AddedComplexity = 5 in
3763 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3764 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3765 Requires<[HasSSE2]>;
3766 // Special unary SHUFPDrri case.
3767 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3768 (SHUFPDrri VR128:$src1, VR128:$src1,
3769 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3770 Requires<[HasSSE2]>;
3771 // Special unary SHUFPDrri case.
3772 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3773 (SHUFPDrri VR128:$src1, VR128:$src1,
3774 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3775 Requires<[HasSSE2]>;
3776 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3777 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3778 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3779 Requires<[HasSSE2]>;
3781 // Special binary v4i32 shuffle cases with SHUFPS.
3782 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3783 (SHUFPSrri VR128:$src1, VR128:$src2,
3784 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3785 Requires<[HasSSE2]>;
3786 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3787 (SHUFPSrmi VR128:$src1, addr:$src2,
3788 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3789 Requires<[HasSSE2]>;
3790 // Special binary v2i64 shuffle cases using SHUFPDrri.
3791 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3792 (SHUFPDrri VR128:$src1, VR128:$src2,
3793 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3794 Requires<[HasSSE2]>;
3796 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3797 let AddedComplexity = 15 in {
3798 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3799 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3800 Requires<[OptForSpeed, HasSSE2]>;
3801 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3802 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3803 Requires<[OptForSpeed, HasSSE2]>;
3805 let AddedComplexity = 10 in {
3806 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3807 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3808 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3809 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3810 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3811 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3812 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3813 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3816 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3817 let AddedComplexity = 15 in {
3818 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3819 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3820 Requires<[OptForSpeed, HasSSE2]>;
3821 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3822 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3823 Requires<[OptForSpeed, HasSSE2]>;
3825 let AddedComplexity = 10 in {
3826 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3827 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3828 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3829 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3830 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3831 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3832 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3833 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3836 let AddedComplexity = 20 in {
3837 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3838 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3839 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3841 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3842 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3843 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3845 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3846 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3847 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3848 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3849 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3852 let AddedComplexity = 20 in {
3853 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3854 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3855 (MOVLPSrm VR128:$src1, addr:$src2)>;
3856 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3857 (MOVLPDrm VR128:$src1, addr:$src2)>;
3858 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3859 (MOVLPSrm VR128:$src1, addr:$src2)>;
3860 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3861 (MOVLPDrm VR128:$src1, addr:$src2)>;
3864 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3865 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3866 (MOVLPSmr addr:$src1, VR128:$src2)>;
3867 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3868 (MOVLPDmr addr:$src1, VR128:$src2)>;
3869 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3871 (MOVLPSmr addr:$src1, VR128:$src2)>;
3872 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3873 (MOVLPDmr addr:$src1, VR128:$src2)>;
3875 let AddedComplexity = 15 in {
3876 // Setting the lowest element in the vector.
3877 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3878 (MOVSSrr (v4i32 VR128:$src1),
3879 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3880 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3881 (MOVSDrr (v2i64 VR128:$src1),
3882 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3884 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3885 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3886 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3887 Requires<[HasSSE2]>;
3888 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3889 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3890 Requires<[HasSSE2]>;
3893 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3894 // fall back to this for SSE1)
3895 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3896 (SHUFPSrri VR128:$src2, VR128:$src1,
3897 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3899 // Set lowest element and zero upper elements.
3900 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3901 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3903 // Some special case pandn patterns.
3904 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3906 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3907 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3909 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3910 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3912 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3914 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3915 (memop addr:$src2))),
3916 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3917 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3918 (memop addr:$src2))),
3919 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3920 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3921 (memop addr:$src2))),
3922 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3924 // vector -> vector casts
3925 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3926 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3927 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3928 (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3929 def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
3930 (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
3931 def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
3932 (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
3934 // Use movaps / movups for SSE integer load / store (one byte shorter).
3935 def : Pat<(alignedloadv4i32 addr:$src),
3936 (MOVAPSrm addr:$src)>;
3937 def : Pat<(loadv4i32 addr:$src),
3938 (MOVUPSrm addr:$src)>;
3939 def : Pat<(alignedloadv2i64 addr:$src),
3940 (MOVAPSrm addr:$src)>;
3941 def : Pat<(loadv2i64 addr:$src),
3942 (MOVUPSrm addr:$src)>;
3944 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3945 (MOVAPSmr addr:$dst, VR128:$src)>;
3946 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3947 (MOVAPSmr addr:$dst, VR128:$src)>;
3948 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3949 (MOVAPSmr addr:$dst, VR128:$src)>;
3950 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3951 (MOVAPSmr addr:$dst, VR128:$src)>;
3952 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3953 (MOVUPSmr addr:$dst, VR128:$src)>;
3954 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3955 (MOVUPSmr addr:$dst, VR128:$src)>;
3956 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3957 (MOVUPSmr addr:$dst, VR128:$src)>;
3958 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3959 (MOVUPSmr addr:$dst, VR128:$src)>;
3961 //===----------------------------------------------------------------------===//
3962 // SSE4.1 - Packed Move with Sign/Zero Extend
3963 //===----------------------------------------------------------------------===//
3965 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3966 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3967 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3968 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3970 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3971 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3973 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3977 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3978 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3980 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3982 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3984 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3986 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3988 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3992 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3993 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3994 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3995 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3996 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3997 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3999 // Common patterns involving scalar load.
4000 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4001 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4002 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4003 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4005 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4006 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4007 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4008 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4010 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4011 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4012 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4013 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4015 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4016 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4017 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4018 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4020 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4021 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4022 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4023 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4025 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4026 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4027 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4028 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4031 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4032 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4033 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4034 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4036 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4037 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4039 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4043 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4044 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4046 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4048 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4050 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4054 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4055 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4056 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4057 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4059 // Common patterns involving scalar load
4060 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4061 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4062 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4063 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4065 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4066 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4067 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4068 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4071 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4072 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4073 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4074 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4076 // Expecting a i16 load any extended to i32 value.
4077 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4078 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4079 [(set VR128:$dst, (IntId (bitconvert
4080 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4084 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4085 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4087 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4090 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4091 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4093 // Common patterns involving scalar load
4094 def : Pat<(int_x86_sse41_pmovsxbq
4095 (bitconvert (v4i32 (X86vzmovl
4096 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4097 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4099 def : Pat<(int_x86_sse41_pmovzxbq
4100 (bitconvert (v4i32 (X86vzmovl
4101 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4102 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4104 //===----------------------------------------------------------------------===//
4105 // SSE4.1 - Extract Instructions
4106 //===----------------------------------------------------------------------===//
4108 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4109 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4110 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4111 (ins VR128:$src1, i32i8imm:$src2),
4112 !strconcat(OpcodeStr,
4113 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4114 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4116 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4117 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4118 !strconcat(OpcodeStr,
4119 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4122 // There's an AssertZext in the way of writing the store pattern
4123 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4126 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4127 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4128 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4129 (ins VR128:$src1, i32i8imm:$src2),
4130 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4133 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4136 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4137 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4138 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4139 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4140 !strconcat(OpcodeStr,
4141 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4144 // There's an AssertZext in the way of writing the store pattern
4145 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4148 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4149 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4151 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4154 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4155 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4156 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4157 (ins VR128:$src1, i32i8imm:$src2),
4158 !strconcat(OpcodeStr,
4159 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4161 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4162 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4163 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4164 !strconcat(OpcodeStr,
4165 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4166 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4167 addr:$dst)]>, OpSize;
4170 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4171 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4173 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4175 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4176 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4177 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4178 (ins VR128:$src1, i32i8imm:$src2),
4179 !strconcat(OpcodeStr,
4180 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4182 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4183 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4184 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4185 !strconcat(OpcodeStr,
4186 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4187 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4188 addr:$dst)]>, OpSize, REX_W;
4191 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4192 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4194 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4196 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4198 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4199 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4200 (ins VR128:$src1, i32i8imm:$src2),
4201 !strconcat(OpcodeStr,
4202 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4204 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4206 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4207 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4208 !strconcat(OpcodeStr,
4209 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4210 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4211 addr:$dst)]>, OpSize;
4214 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4215 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4216 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4217 (ins VR128:$src1, i32i8imm:$src2),
4218 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4221 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4223 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4224 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4227 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4228 Requires<[HasSSE41]>;
4230 //===----------------------------------------------------------------------===//
4231 // SSE4.1 - Insert Instructions
4232 //===----------------------------------------------------------------------===//
4234 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4235 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4236 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4238 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4240 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4242 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4243 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4244 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4246 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4248 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4250 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4251 imm:$src3))]>, OpSize;
4254 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4255 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4256 let Constraints = "$src1 = $dst" in
4257 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4259 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4260 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4261 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4263 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4265 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4267 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4269 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4270 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4272 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4274 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4276 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4277 imm:$src3)))]>, OpSize;
4280 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4281 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4282 let Constraints = "$src1 = $dst" in
4283 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4285 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4286 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4287 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4289 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4291 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4293 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4295 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4296 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4298 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4300 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4302 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4303 imm:$src3)))]>, OpSize;
4306 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4307 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4308 let Constraints = "$src1 = $dst" in
4309 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4311 // insertps has a few different modes, there's the first two here below which
4312 // are optimized inserts that won't zero arbitrary elements in the destination
4313 // vector. The next one matches the intrinsic and could zero arbitrary elements
4314 // in the target vector.
4315 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4316 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4317 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4319 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4321 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4323 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4325 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4326 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4328 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4330 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4332 (X86insrtps VR128:$src1,
4333 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4334 imm:$src3))]>, OpSize;
4337 let Constraints = "$src1 = $dst" in
4338 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4339 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4340 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4342 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4343 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4345 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4346 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4347 Requires<[HasSSE41]>;
4349 //===----------------------------------------------------------------------===//
4350 // SSE4.1 - Round Instructions
4351 //===----------------------------------------------------------------------===//
4353 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd,
4356 Intrinsic V2F64Int> {
4357 // Intrinsic operation, reg.
4358 // Vector intrinsic operation, reg
4359 def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
4360 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4361 !strconcat(OpcodeStr,
4362 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4363 [(set VR128:$dst, (V4F32Int VR128:$src1, imm:$src2))]>,
4366 // Vector intrinsic operation, mem
4367 def PSm_Int : Ii8<opcps, MRMSrcMem,
4368 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4369 !strconcat(OpcodeStr,
4370 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4372 (V4F32Int (memopv4f32 addr:$src1),imm:$src2))]>,
4374 Requires<[HasSSE41]>;
4376 // Vector intrinsic operation, reg
4377 def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
4378 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4379 !strconcat(OpcodeStr,
4380 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4381 [(set VR128:$dst, (V2F64Int VR128:$src1, imm:$src2))]>,
4384 // Vector intrinsic operation, mem
4385 def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
4386 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4387 !strconcat(OpcodeStr,
4388 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4390 (V2F64Int (memopv2f64 addr:$src1),imm:$src2))]>,
4394 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4395 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4396 // Intrinsic operation, reg.
4397 // Vector intrinsic operation, reg
4398 def PSr : SS4AIi8<opcps, MRMSrcReg,
4399 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4400 !strconcat(OpcodeStr,
4401 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4404 // Vector intrinsic operation, mem
4405 def PSm : Ii8<opcps, MRMSrcMem,
4406 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4407 !strconcat(OpcodeStr,
4408 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4409 []>, TA, OpSize, Requires<[HasSSE41]>;
4411 // Vector intrinsic operation, reg
4412 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4413 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4414 !strconcat(OpcodeStr,
4415 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4418 // Vector intrinsic operation, mem
4419 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4420 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4421 !strconcat(OpcodeStr,
4422 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4426 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4429 Intrinsic F64Int, bit Is2Addr = 1> {
4430 // Intrinsic operation, reg.
4431 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
4432 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4434 !strconcat(OpcodeStr,
4435 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4436 !strconcat(OpcodeStr,
4437 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4438 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4441 // Intrinsic operation, mem.
4442 def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
4443 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4445 !strconcat(OpcodeStr,
4446 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4447 !strconcat(OpcodeStr,
4448 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4450 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4453 // Intrinsic operation, reg.
4454 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
4455 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4457 !strconcat(OpcodeStr,
4458 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4459 !strconcat(OpcodeStr,
4460 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4461 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4464 // Intrinsic operation, mem.
4465 def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
4466 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4468 !strconcat(OpcodeStr,
4469 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4470 !strconcat(OpcodeStr,
4471 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4473 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4477 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4479 // Intrinsic operation, reg.
4480 def SSr : SS4AIi8<opcss, MRMSrcReg,
4481 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4482 !strconcat(OpcodeStr,
4483 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4486 // Intrinsic operation, mem.
4487 def SSm : SS4AIi8<opcss, MRMSrcMem,
4488 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4489 !strconcat(OpcodeStr,
4490 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4493 // Intrinsic operation, reg.
4494 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4495 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4496 !strconcat(OpcodeStr,
4497 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4500 // Intrinsic operation, mem.
4501 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4502 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4503 !strconcat(OpcodeStr,
4504 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4508 // FP round - roundss, roundps, roundsd, roundpd
4509 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4511 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround",
4512 int_x86_sse41_round_ps, int_x86_sse41_round_pd>,
4514 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4515 int_x86_sse41_round_ss, int_x86_sse41_round_sd,
4517 // Instructions for the assembler
4518 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4520 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4522 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4525 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round",
4526 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4527 let Constraints = "$src1 = $dst" in
4528 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4529 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4531 //===----------------------------------------------------------------------===//
4532 // SSE4.1 - Packed Bit Test
4533 //===----------------------------------------------------------------------===//
4535 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4536 // the intel intrinsic that corresponds to this.
4537 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4538 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4539 "vptest\t{$src2, $src1|$src1, $src2}",
4540 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4542 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4543 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4545 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4546 "vptest\t{$src2, $src1|$src1, $src2}",
4547 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4549 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4550 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4553 let Defs = [EFLAGS] in {
4554 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4555 "ptest \t{$src2, $src1|$src1, $src2}",
4556 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4558 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4559 "ptest \t{$src2, $src1|$src1, $src2}",
4560 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4564 // The bit test instructions below are AVX only
4565 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4566 X86MemOperand x86memop> {
4567 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4568 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4570 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4571 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4575 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4576 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem>;
4577 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem>;
4578 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem>;
4579 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem>;
4582 //===----------------------------------------------------------------------===//
4583 // SSE4.1 - Misc Instructions
4584 //===----------------------------------------------------------------------===//
4586 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4587 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4588 Intrinsic IntId128> {
4589 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4591 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4592 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4593 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4595 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4598 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4601 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4602 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4603 int_x86_sse41_phminposuw>, VEX;
4604 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4605 int_x86_sse41_phminposuw>;
4607 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4608 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4609 Intrinsic IntId128, bit Is2Addr = 1> {
4610 let isCommutable = 1 in
4611 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4612 (ins VR128:$src1, VR128:$src2),
4614 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4615 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4616 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4617 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4618 (ins VR128:$src1, i128mem:$src2),
4620 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4621 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4623 (IntId128 VR128:$src1,
4624 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4627 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4628 let isCommutable = 0 in
4629 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4631 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4633 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4635 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4637 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4639 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4641 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4643 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4645 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4647 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4649 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4653 let Constraints = "$src1 = $dst" in {
4654 let isCommutable = 0 in
4655 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4656 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4657 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4658 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4659 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4660 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4661 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4662 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4663 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4664 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4665 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4668 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4669 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4670 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4671 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4673 /// SS48I_binop_rm - Simple SSE41 binary operator.
4674 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4675 ValueType OpVT, bit Is2Addr = 1> {
4676 let isCommutable = 1 in
4677 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4678 (ins VR128:$src1, VR128:$src2),
4680 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4681 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4682 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4684 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4685 (ins VR128:$src1, i128mem:$src2),
4687 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4688 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4689 [(set VR128:$dst, (OpNode VR128:$src1,
4690 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4694 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4695 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4696 let Constraints = "$src1 = $dst" in
4697 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4699 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4700 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4701 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4702 X86MemOperand x86memop, bit Is2Addr = 1> {
4703 let isCommutable = 1 in
4704 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4705 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4707 !strconcat(OpcodeStr,
4708 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4709 !strconcat(OpcodeStr,
4710 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4711 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4713 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4714 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4716 !strconcat(OpcodeStr,
4717 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4718 !strconcat(OpcodeStr,
4719 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4722 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4726 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4727 let isCommutable = 0 in {
4728 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4729 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4730 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4731 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4732 let Pattern = []<dag> in {
4733 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4734 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4735 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4736 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4738 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4739 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4740 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4741 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4743 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4744 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4745 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4746 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4747 let Pattern = []<dag> in
4748 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4749 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4752 let Constraints = "$src1 = $dst" in {
4753 let isCommutable = 0 in {
4754 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4755 VR128, memopv16i8, i128mem>;
4756 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4757 VR128, memopv16i8, i128mem>;
4758 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4759 VR128, memopv16i8, i128mem>;
4760 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4761 VR128, memopv16i8, i128mem>;
4763 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4764 VR128, memopv16i8, i128mem>;
4765 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4766 VR128, memopv16i8, i128mem>;
4769 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4770 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4771 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4772 RegisterClass RC, X86MemOperand x86memop,
4773 PatFrag mem_frag, Intrinsic IntId> {
4774 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4775 (ins RC:$src1, RC:$src2, RC:$src3),
4776 !strconcat(OpcodeStr,
4777 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4778 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4779 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4781 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4782 (ins RC:$src1, x86memop:$src2, RC:$src3),
4783 !strconcat(OpcodeStr,
4784 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4786 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4788 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4792 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4793 memopv16i8, int_x86_sse41_blendvpd>;
4794 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4795 memopv16i8, int_x86_sse41_blendvps>;
4796 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4797 memopv16i8, int_x86_sse41_pblendvb>;
4799 let Pattern = []<dag> in { // FIXME: implement 256 intrinsics here.
4800 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4801 memopv32i8, int_x86_sse41_blendvpd>;
4802 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4803 memopv32i8, int_x86_sse41_blendvps>;
4806 /// SS41I_ternary_int - SSE 4.1 ternary operator
4807 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4808 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4809 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4810 (ins VR128:$src1, VR128:$src2),
4811 !strconcat(OpcodeStr,
4812 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4813 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4816 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4817 (ins VR128:$src1, i128mem:$src2),
4818 !strconcat(OpcodeStr,
4819 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4822 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4826 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4827 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4828 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4830 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4831 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4832 "vmovntdqa\t{$src, $dst|$dst, $src}",
4833 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4835 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4836 "movntdqa\t{$src, $dst|$dst, $src}",
4837 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4840 //===----------------------------------------------------------------------===//
4841 // SSE4.2 - Compare Instructions
4842 //===----------------------------------------------------------------------===//
4844 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4845 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4846 Intrinsic IntId128, bit Is2Addr = 1> {
4847 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4848 (ins VR128:$src1, VR128:$src2),
4850 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4851 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4852 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4854 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4855 (ins VR128:$src1, i128mem:$src2),
4857 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4858 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4860 (IntId128 VR128:$src1,
4861 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4864 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4865 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4867 let Constraints = "$src1 = $dst" in
4868 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4870 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4871 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4872 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4873 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4875 //===----------------------------------------------------------------------===//
4876 // SSE4.2 - String/text Processing Instructions
4877 //===----------------------------------------------------------------------===//
4879 // Packed Compare Implicit Length Strings, Return Mask
4880 multiclass pseudo_pcmpistrm<string asm> {
4881 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4882 (ins VR128:$src1, VR128:$src2, i8imm:$src3), !strconcat(asm, "rr PSEUDO"),
4883 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4885 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4886 (ins VR128:$src1, i128mem:$src2, i8imm:$src3), !strconcat(asm, "rm PSEUDO"),
4887 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4888 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4891 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4892 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4893 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4896 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4897 Predicates = [HasAVX] in {
4898 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4899 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4900 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4901 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4902 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4903 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4906 let Defs = [XMM0, EFLAGS] in {
4907 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4908 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4909 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4910 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4911 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4912 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4915 // Packed Compare Explicit Length Strings, Return Mask
4916 multiclass pseudo_pcmpestrm<string asm> {
4917 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4918 (ins VR128:$src1, VR128:$src3, i8imm:$src5), !strconcat(asm, "rr PSEUDO"),
4919 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4920 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4921 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4922 (ins VR128:$src1, i128mem:$src3, i8imm:$src5), !strconcat(asm, "rm PSEUDO"),
4923 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4924 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4927 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4928 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4929 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4932 let isAsmParserOnly = 1, Predicates = [HasAVX],
4933 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4934 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4935 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4936 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4937 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4938 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4939 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4942 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4943 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4944 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4945 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4946 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4947 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4948 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4951 // Packed Compare Implicit Length Strings, Return Index
4952 let Defs = [ECX, EFLAGS] in {
4953 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4954 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4955 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4956 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4957 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4958 (implicit EFLAGS)]>, OpSize;
4959 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4960 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4961 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4962 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4963 (implicit EFLAGS)]>, OpSize;
4967 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4968 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4970 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4972 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4974 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4976 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4978 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4982 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4983 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4984 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4985 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4986 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4987 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4989 // Packed Compare Explicit Length Strings, Return Index
4990 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4991 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4992 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4993 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4994 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4995 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4996 (implicit EFLAGS)]>, OpSize;
4997 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4998 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4999 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5001 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5002 (implicit EFLAGS)]>, OpSize;
5006 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5007 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5009 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5011 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5013 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5015 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5017 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5021 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5022 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5023 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5024 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5025 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5026 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5028 //===----------------------------------------------------------------------===//
5029 // SSE4.2 - CRC Instructions
5030 //===----------------------------------------------------------------------===//
5032 // No CRC instructions have AVX equivalents
5034 // crc intrinsic instruction
5035 // This set of instructions are only rm, the only difference is the size
5037 let Constraints = "$src1 = $dst" in {
5038 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5039 (ins GR32:$src1, i8mem:$src2),
5040 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5042 (int_x86_sse42_crc32_8 GR32:$src1,
5043 (load addr:$src2)))]>;
5044 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5045 (ins GR32:$src1, GR8:$src2),
5046 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5048 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5049 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5050 (ins GR32:$src1, i16mem:$src2),
5051 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5053 (int_x86_sse42_crc32_16 GR32:$src1,
5054 (load addr:$src2)))]>,
5056 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5057 (ins GR32:$src1, GR16:$src2),
5058 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5060 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5062 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5063 (ins GR32:$src1, i32mem:$src2),
5064 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5066 (int_x86_sse42_crc32_32 GR32:$src1,
5067 (load addr:$src2)))]>;
5068 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5069 (ins GR32:$src1, GR32:$src2),
5070 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5072 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5073 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5074 (ins GR64:$src1, i8mem:$src2),
5075 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5077 (int_x86_sse42_crc64_8 GR64:$src1,
5078 (load addr:$src2)))]>,
5080 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5081 (ins GR64:$src1, GR8:$src2),
5082 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5084 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5086 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5087 (ins GR64:$src1, i64mem:$src2),
5088 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5090 (int_x86_sse42_crc64_64 GR64:$src1,
5091 (load addr:$src2)))]>,
5093 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5094 (ins GR64:$src1, GR64:$src2),
5095 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5097 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5101 //===----------------------------------------------------------------------===//
5102 // AES-NI Instructions
5103 //===----------------------------------------------------------------------===//
5105 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5106 Intrinsic IntId128, bit Is2Addr = 1> {
5107 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5108 (ins VR128:$src1, VR128:$src2),
5110 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5111 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5112 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5114 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5115 (ins VR128:$src1, i128mem:$src2),
5117 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5118 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5120 (IntId128 VR128:$src1,
5121 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5124 // Perform One Round of an AES Encryption/Decryption Flow
5125 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5126 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5127 int_x86_aesni_aesenc, 0>, VEX_4V;
5128 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5129 int_x86_aesni_aesenclast, 0>, VEX_4V;
5130 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5131 int_x86_aesni_aesdec, 0>, VEX_4V;
5132 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5133 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5136 let Constraints = "$src1 = $dst" in {
5137 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5138 int_x86_aesni_aesenc>;
5139 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5140 int_x86_aesni_aesenclast>;
5141 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5142 int_x86_aesni_aesdec>;
5143 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5144 int_x86_aesni_aesdeclast>;
5147 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5148 (AESENCrr VR128:$src1, VR128:$src2)>;
5149 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5150 (AESENCrm VR128:$src1, addr:$src2)>;
5151 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5152 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5153 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5154 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5155 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5156 (AESDECrr VR128:$src1, VR128:$src2)>;
5157 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5158 (AESDECrm VR128:$src1, addr:$src2)>;
5159 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5160 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5161 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5162 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5164 // Perform the AES InvMixColumn Transformation
5165 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5166 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5168 "vaesimc\t{$src1, $dst|$dst, $src1}",
5170 (int_x86_aesni_aesimc VR128:$src1))]>,
5172 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5173 (ins i128mem:$src1),
5174 "vaesimc\t{$src1, $dst|$dst, $src1}",
5176 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5179 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5181 "aesimc\t{$src1, $dst|$dst, $src1}",
5183 (int_x86_aesni_aesimc VR128:$src1))]>,
5185 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5186 (ins i128mem:$src1),
5187 "aesimc\t{$src1, $dst|$dst, $src1}",
5189 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5192 // AES Round Key Generation Assist
5193 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5194 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5195 (ins VR128:$src1, i8imm:$src2),
5196 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5198 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5200 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5201 (ins i128mem:$src1, i8imm:$src2),
5202 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5204 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5208 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5209 (ins VR128:$src1, i8imm:$src2),
5210 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5212 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5214 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5215 (ins i128mem:$src1, i8imm:$src2),
5216 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5218 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5222 //===----------------------------------------------------------------------===//
5223 // CLMUL Instructions
5224 //===----------------------------------------------------------------------===//
5226 // Only the AVX version of CLMUL instructions are described here.
5228 // Carry-less Multiplication instructions
5229 let isAsmParserOnly = 1 in {
5230 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5231 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5232 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5235 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5236 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5237 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5241 multiclass avx_vpclmul<string asm> {
5242 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5243 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5246 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5247 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5250 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5251 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5252 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5253 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5255 } // isAsmParserOnly
5257 //===----------------------------------------------------------------------===//
5259 //===----------------------------------------------------------------------===//
5261 let isAsmParserOnly = 1 in {
5263 // Load from memory and broadcast to all elements of the destination operand
5264 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5265 X86MemOperand x86memop> :
5266 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5267 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>, VEX;
5269 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem>;
5270 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem>;
5271 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem>;
5272 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem>;
5274 // Insert packed floating-point values
5275 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5276 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5277 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5279 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5280 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5281 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5284 // Extract packed floating-point values
5285 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5286 (ins VR256:$src1, i8imm:$src2),
5287 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5289 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5290 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5291 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5294 // Conditional SIMD Packed Loads and Stores
5295 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr> {
5296 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5297 (ins VR128:$src1, f128mem:$src2),
5298 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5300 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5301 (ins VR256:$src1, f256mem:$src2),
5302 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5304 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5305 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5306 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5308 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5309 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5310 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5314 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps">;
5315 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd">;
5317 // Permute Floating-Point Values
5318 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5319 RegisterClass RC, X86MemOperand x86memop> {
5320 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5321 (ins RC:$src1, RC:$src2),
5322 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5324 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5325 (ins RC:$src1, x86memop:$src2),
5326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5328 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5329 (ins RC:$src1, i8imm:$src2),
5330 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5332 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5333 (ins x86memop:$src1, i8imm:$src2),
5334 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5338 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem>;
5339 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem>;
5340 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem>;
5341 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem>;
5343 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5344 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5345 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5347 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5348 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5349 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5352 // Zero All YMM registers
5353 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall", []>, VEX, VEX_L,
5356 // Zero Upper bits of YMM registers
5357 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper", []>, VEX,
5360 } // isAsmParserOnly