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, $src}">, XS, VEX;
548 defm VCVTTSS2SIr64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
549 "cvttss2si\t{$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, $src}">, XD, VEX;
553 defm VCVTTSD2SIr64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
554 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
557 // The assembler can recognize rr 64-bit instructions by seeing a rxx
558 // register, but the same isn't true when only using memory operands,
559 // provide other assembly "l" and "q" forms to address this explicitly
560 // where appropriate to do so.
561 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
563 defm VCVTSI2SSQ : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ssq">, XS,
565 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
567 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sdl">, XD,
569 defm VCVTSI2SDQ : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sdq">, XD,
573 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
574 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
575 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
576 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
577 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
578 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
579 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
580 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
582 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
583 // and/or XMM operand(s).
584 multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
585 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
586 string asm, Domain d> {
587 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
588 [(set DstRC:$dst, (Int SrcRC:$src))], d>;
589 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
590 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], d>;
593 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
594 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
596 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
597 [(set DstRC:$dst, (Int SrcRC:$src))]>;
598 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
599 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
602 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
603 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
604 PatFrag ld_frag, string asm, Domain d> {
605 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
606 asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], d>;
607 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
608 (ins DstRC:$src1, x86memop:$src2), asm,
609 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], d>;
612 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
613 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
614 PatFrag ld_frag, string asm> {
615 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
616 asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
617 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
618 (ins DstRC:$src1, x86memop:$src2), asm,
619 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
622 let isAsmParserOnly = 1 in {
623 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
624 f32mem, load, "cvtss2si\t{$src, $dst|$dst, $src}">, XS,
626 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
627 f128mem, load, "cvtsd2si\t{$src, $dst|$dst, $src}">, XD,
629 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
630 // Get rid of this hack or rename the intrinsics, there are several
631 // intructions that only match with the intrinsic form, why create duplicates
632 // to let them be recognized by the assembler?
633 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
634 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
635 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
636 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
638 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
639 f32mem, load, "cvtss2si\t{$src, $dst|$dst, $src}">, XS;
640 defm Int_CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
641 f128mem, load, "cvtsd2si\t{$src, $dst|$dst, $src}">, XD;
644 let Constraints = "$src1 = $dst" in {
645 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
646 int_x86_sse_cvtsi2ss, i32mem, loadi32,
647 "cvtsi2ss\t{$src2, $dst|$dst, $src2}">, XS;
648 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
649 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
650 "cvtsi2ss\t{$src2, $dst|$dst, $src2}">, XD;
653 // Instructions below don't have an AVX form.
654 defm Int_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
655 f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
656 SSEPackedSingle>, TB;
657 defm Int_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
658 f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
659 SSEPackedDouble>, TB, OpSize;
660 defm Int_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
661 f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
662 SSEPackedSingle>, TB;
663 defm Int_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
664 f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
665 SSEPackedDouble>, TB, OpSize;
666 defm Int_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
667 i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
668 SSEPackedDouble>, TB, OpSize;
669 let Constraints = "$src1 = $dst" in {
670 defm Int_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
671 int_x86_sse_cvtpi2ps,
672 i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
673 SSEPackedSingle>, TB;
678 // Aliases for intrinsics
679 let isAsmParserOnly = 1, Pattern = []<dag> in {
680 defm Int_VCVTTSS2SI : sse12_cvt_sint_3addr<0x2C, VR128, GR32,
681 int_x86_sse_cvttss2si, f32mem, load,
682 "cvttss2si\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS;
683 defm Int_VCVTTSD2SI : sse12_cvt_sint_3addr<0x2C, VR128, GR32,
684 int_x86_sse2_cvttsd2si, f128mem, load,
685 "cvttss2si\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD;
687 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
688 f32mem, load, "cvttss2si\t{$src, $dst|$dst, $src}">,
690 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
691 f128mem, load, "cvttss2si\t{$src, $dst|$dst, $src}">,
694 let isAsmParserOnly = 1, Pattern = []<dag> in {
695 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
696 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
697 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
698 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
700 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load,
701 "cvtdq2ps\t{$src, $dst|$dst, $src}",
702 SSEPackedSingle>, TB, VEX;
703 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, f256mem, load,
704 "cvtdq2ps\t{$src, $dst|$dst, $src}",
705 SSEPackedSingle>, TB, VEX;
707 let Pattern = []<dag> in {
708 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
709 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
710 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load /*dummy*/,
711 "cvtdq2ps\t{$src, $dst|$dst, $src}",
712 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
717 // Convert scalar double to scalar single
718 let isAsmParserOnly = 1 in {
719 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
720 (ins FR64:$src1, FR64:$src2),
721 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
723 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
724 (ins FR64:$src1, f64mem:$src2),
725 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
726 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
728 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
729 "cvtsd2ss\t{$src, $dst|$dst, $src}",
730 [(set FR32:$dst, (fround FR64:$src))]>;
731 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
732 "cvtsd2ss\t{$src, $dst|$dst, $src}",
733 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
734 Requires<[HasSSE2, OptForSize]>;
736 let isAsmParserOnly = 1 in
737 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
738 int_x86_sse2_cvtsd2ss, f64mem, load,
739 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
741 let Constraints = "$src1 = $dst" in
742 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
743 int_x86_sse2_cvtsd2ss, f64mem, load,
744 "cvtsd2ss\t{$src2, $dst|$dst, $src2}">, XS;
746 // Convert scalar single to scalar double
747 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
748 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
749 (ins FR32:$src1, FR32:$src2),
750 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
751 []>, XS, Requires<[HasAVX]>, VEX_4V;
752 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
753 (ins FR32:$src1, f32mem:$src2),
754 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
755 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
757 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
758 "cvtss2sd\t{$src, $dst|$dst, $src}",
759 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
761 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
762 "cvtss2sd\t{$src, $dst|$dst, $src}",
763 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
764 Requires<[HasSSE2, OptForSize]>;
766 let isAsmParserOnly = 1 in {
767 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
768 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
769 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
770 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
771 VR128:$src2))]>, XS, VEX_4V,
773 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
774 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
775 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
776 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
777 (load addr:$src2)))]>, XS, VEX_4V,
780 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
781 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
782 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
783 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
784 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
787 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
788 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
789 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
790 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
791 (load addr:$src2)))]>, XS,
795 def : Pat<(extloadf32 addr:$src),
796 (CVTSS2SDrr (MOVSSrm addr:$src))>,
797 Requires<[HasSSE2, OptForSpeed]>;
799 // Convert doubleword to packed single/double fp
800 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
801 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
802 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
803 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
804 TB, VEX, Requires<[HasAVX]>;
805 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
806 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
807 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
808 (bitconvert (memopv2i64 addr:$src))))]>,
809 TB, VEX, Requires<[HasAVX]>;
811 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
812 "cvtdq2ps\t{$src, $dst|$dst, $src}",
813 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
814 TB, Requires<[HasSSE2]>;
815 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
816 "cvtdq2ps\t{$src, $dst|$dst, $src}",
817 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
818 (bitconvert (memopv2i64 addr:$src))))]>,
819 TB, Requires<[HasSSE2]>;
821 // FIXME: why the non-intrinsic version is described as SSE3?
822 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
823 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
824 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
825 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
826 XS, VEX, Requires<[HasAVX]>;
827 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
828 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
829 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
830 (bitconvert (memopv2i64 addr:$src))))]>,
831 XS, VEX, Requires<[HasAVX]>;
833 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
834 "cvtdq2pd\t{$src, $dst|$dst, $src}",
835 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
836 XS, Requires<[HasSSE2]>;
837 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
838 "cvtdq2pd\t{$src, $dst|$dst, $src}",
839 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
840 (bitconvert (memopv2i64 addr:$src))))]>,
841 XS, Requires<[HasSSE2]>;
843 // Convert packed single/double fp to doubleword
844 let isAsmParserOnly = 1 in {
845 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
846 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
847 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
848 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
849 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
850 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
851 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
852 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
854 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
855 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
856 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
857 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
859 let isAsmParserOnly = 1 in {
860 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
861 "cvtps2dq\t{$src, $dst|$dst, $src}",
862 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
864 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
866 "cvtps2dq\t{$src, $dst|$dst, $src}",
867 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
868 (memop addr:$src)))]>, VEX;
870 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
871 "cvtps2dq\t{$src, $dst|$dst, $src}",
872 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
873 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
874 "cvtps2dq\t{$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
876 (memop addr:$src)))]>;
878 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
879 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
880 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
881 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
882 XD, VEX, Requires<[HasAVX]>;
883 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
884 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
885 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
886 (memop addr:$src)))]>,
887 XD, VEX, Requires<[HasAVX]>;
889 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
890 "cvtpd2dq\t{$src, $dst|$dst, $src}",
891 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
892 XD, Requires<[HasSSE2]>;
893 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
894 "cvtpd2dq\t{$src, $dst|$dst, $src}",
895 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
896 (memop addr:$src)))]>,
897 XD, Requires<[HasSSE2]>;
900 // Convert with truncation packed single/double fp to doubleword
901 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
902 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
903 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
904 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
905 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
906 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
907 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
908 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
909 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
911 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
912 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
913 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
914 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
917 let isAsmParserOnly = 1 in {
918 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
919 "vcvttps2dq\t{$src, $dst|$dst, $src}",
921 (int_x86_sse2_cvttps2dq VR128:$src))]>,
922 XS, VEX, Requires<[HasAVX]>;
923 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
924 "vcvttps2dq\t{$src, $dst|$dst, $src}",
925 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
926 (memop addr:$src)))]>,
927 XS, VEX, Requires<[HasAVX]>;
929 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
930 "cvttps2dq\t{$src, $dst|$dst, $src}",
932 (int_x86_sse2_cvttps2dq VR128:$src))]>,
933 XS, Requires<[HasSSE2]>;
934 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
935 "cvttps2dq\t{$src, $dst|$dst, $src}",
936 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
937 (memop addr:$src)))]>,
938 XS, Requires<[HasSSE2]>;
940 let isAsmParserOnly = 1 in {
941 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
943 "cvttpd2dq\t{$src, $dst|$dst, $src}",
944 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
946 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
948 "cvttpd2dq\t{$src, $dst|$dst, $src}",
949 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
950 (memop addr:$src)))]>, VEX;
952 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
953 "cvttpd2dq\t{$src, $dst|$dst, $src}",
954 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
955 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
956 "cvttpd2dq\t{$src, $dst|$dst, $src}",
957 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
958 (memop addr:$src)))]>;
960 let isAsmParserOnly = 1 in {
961 // The assembler can recognize rr 256-bit instructions by seeing a ymm
962 // register, but the same isn't true when using memory operands instead.
963 // Provide other assembly rr and rm forms to address this explicitly.
964 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
965 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
966 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
967 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
970 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
971 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
972 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
973 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
976 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
977 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
978 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
979 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
982 // Convert packed single to packed double
983 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
984 // SSE2 instructions without OpSize prefix
985 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
986 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
987 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
988 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
989 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
990 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
991 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
992 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
994 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
995 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
996 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
997 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
999 let isAsmParserOnly = 1 in {
1000 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1001 "cvtps2pd\t{$src, $dst|$dst, $src}",
1002 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1003 VEX, Requires<[HasAVX]>;
1004 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1005 "cvtps2pd\t{$src, $dst|$dst, $src}",
1006 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1007 (load addr:$src)))]>,
1008 VEX, Requires<[HasAVX]>;
1010 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1011 "cvtps2pd\t{$src, $dst|$dst, $src}",
1012 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1013 TB, Requires<[HasSSE2]>;
1014 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1015 "cvtps2pd\t{$src, $dst|$dst, $src}",
1016 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1017 (load addr:$src)))]>,
1018 TB, Requires<[HasSSE2]>;
1020 // Convert packed double to packed single
1021 let isAsmParserOnly = 1 in {
1022 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1023 // register, but the same isn't true when using memory operands instead.
1024 // Provide other assembly rr and rm forms to address this explicitly.
1025 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1026 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1027 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1028 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1031 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1032 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1033 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1034 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1037 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1038 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1039 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1040 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1042 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1043 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1044 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1045 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1048 let isAsmParserOnly = 1 in {
1049 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1050 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1051 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1052 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1054 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1055 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1056 (memop addr:$src)))]>;
1058 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1059 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1060 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1061 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1062 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1063 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1064 (memop addr:$src)))]>;
1066 //===----------------------------------------------------------------------===//
1067 // SSE 1 & 2 - Compare Instructions
1068 //===----------------------------------------------------------------------===//
1070 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1071 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1072 string asm, string asm_alt> {
1073 def rr : SIi8<0xC2, MRMSrcReg,
1074 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1077 def rm : SIi8<0xC2, MRMSrcMem,
1078 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1080 // Accept explicit immediate argument form instead of comparison code.
1081 let isAsmParserOnly = 1 in {
1082 def rr_alt : SIi8<0xC2, MRMSrcReg,
1083 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1086 def rm_alt : SIi8<0xC2, MRMSrcMem,
1087 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1092 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1093 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1094 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1095 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1097 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1098 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1099 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1103 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1104 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1105 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1106 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1107 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1108 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1109 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1112 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1113 Intrinsic Int, string asm> {
1114 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1115 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1116 [(set VR128:$dst, (Int VR128:$src1,
1117 VR128:$src, imm:$cc))]>;
1118 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1119 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1120 [(set VR128:$dst, (Int VR128:$src1,
1121 (load addr:$src), imm:$cc))]>;
1124 // Aliases to match intrinsics which expect XMM operand(s).
1125 let isAsmParserOnly = 1 in {
1126 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1127 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1129 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1130 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1133 let Constraints = "$src1 = $dst" in {
1134 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1135 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1136 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1137 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1141 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1142 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1143 ValueType vt, X86MemOperand x86memop,
1144 PatFrag ld_frag, string OpcodeStr, Domain d> {
1145 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1146 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1147 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1148 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1149 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1150 [(set EFLAGS, (OpNode (vt RC:$src1),
1151 (ld_frag addr:$src2)))], d>;
1154 let Defs = [EFLAGS] in {
1155 let isAsmParserOnly = 1 in {
1156 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1157 "ucomiss", SSEPackedSingle>, VEX;
1158 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1159 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1160 let Pattern = []<dag> in {
1161 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1162 "comiss", SSEPackedSingle>, VEX;
1163 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1164 "comisd", SSEPackedDouble>, OpSize, VEX;
1167 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1168 load, "ucomiss", SSEPackedSingle>, VEX;
1169 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1170 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1172 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1173 load, "comiss", SSEPackedSingle>, VEX;
1174 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1175 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1177 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1178 "ucomiss", SSEPackedSingle>, TB;
1179 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1180 "ucomisd", SSEPackedDouble>, TB, OpSize;
1182 let Pattern = []<dag> in {
1183 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1184 "comiss", SSEPackedSingle>, TB;
1185 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1186 "comisd", SSEPackedDouble>, TB, OpSize;
1189 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1190 load, "ucomiss", SSEPackedSingle>, TB;
1191 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1192 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1194 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1195 "comiss", SSEPackedSingle>, TB;
1196 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1197 "comisd", SSEPackedDouble>, TB, OpSize;
1198 } // Defs = [EFLAGS]
1200 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1201 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1202 Intrinsic Int, string asm, string asm_alt,
1204 def rri : PIi8<0xC2, MRMSrcReg,
1205 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1206 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1207 def rmi : PIi8<0xC2, MRMSrcMem,
1208 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1209 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1210 // Accept explicit immediate argument form instead of comparison code.
1211 let isAsmParserOnly = 1 in {
1212 def rri_alt : PIi8<0xC2, MRMSrcReg,
1213 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1215 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1216 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1221 let isAsmParserOnly = 1 in {
1222 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1223 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1224 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1225 SSEPackedSingle>, VEX_4V;
1226 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1227 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1228 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1229 SSEPackedDouble>, OpSize, VEX_4V;
1230 let Pattern = []<dag> in {
1231 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_sse_cmp_ps,
1232 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1233 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1234 SSEPackedSingle>, VEX_4V;
1235 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_sse2_cmp_pd,
1236 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1237 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1238 SSEPackedDouble>, OpSize, VEX_4V;
1241 let Constraints = "$src1 = $dst" in {
1242 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1243 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1244 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1245 SSEPackedSingle>, TB;
1246 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1247 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1248 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1249 SSEPackedDouble>, TB, OpSize;
1252 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1253 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1254 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1255 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1256 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1257 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1258 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1259 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1261 //===----------------------------------------------------------------------===//
1262 // SSE 1 & 2 - Shuffle Instructions
1263 //===----------------------------------------------------------------------===//
1265 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1266 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1267 ValueType vt, string asm, PatFrag mem_frag,
1268 Domain d, bit IsConvertibleToThreeAddress = 0> {
1269 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1270 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1271 [(set RC:$dst, (vt (shufp:$src3
1272 RC:$src1, (mem_frag addr:$src2))))], d>;
1273 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1274 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1275 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1277 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1280 let isAsmParserOnly = 1 in {
1281 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1282 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1283 memopv4f32, SSEPackedSingle>, VEX_4V;
1284 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1285 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1286 memopv8f32, SSEPackedSingle>, VEX_4V;
1287 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1288 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1289 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1290 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1291 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1292 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1295 let Constraints = "$src1 = $dst" in {
1296 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1297 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1298 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1300 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1301 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1302 memopv2f64, SSEPackedDouble>, TB, OpSize;
1305 //===----------------------------------------------------------------------===//
1306 // SSE 1 & 2 - Unpack Instructions
1307 //===----------------------------------------------------------------------===//
1309 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1310 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1311 PatFrag mem_frag, RegisterClass RC,
1312 X86MemOperand x86memop, string asm,
1314 def rr : PI<opc, MRMSrcReg,
1315 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1317 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1318 def rm : PI<opc, MRMSrcMem,
1319 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1321 (vt (OpNode RC:$src1,
1322 (mem_frag addr:$src2))))], d>;
1325 let AddedComplexity = 10 in {
1326 let isAsmParserOnly = 1 in {
1327 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1328 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1329 SSEPackedSingle>, VEX_4V;
1330 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1331 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1332 SSEPackedDouble>, OpSize, VEX_4V;
1333 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1334 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1335 SSEPackedSingle>, VEX_4V;
1336 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1337 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1338 SSEPackedDouble>, OpSize, VEX_4V;
1340 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1341 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1342 SSEPackedSingle>, VEX_4V;
1343 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1344 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1345 SSEPackedDouble>, OpSize, VEX_4V;
1346 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1347 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1348 SSEPackedSingle>, VEX_4V;
1349 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1350 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1351 SSEPackedDouble>, OpSize, VEX_4V;
1354 let Constraints = "$src1 = $dst" in {
1355 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1356 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1357 SSEPackedSingle>, TB;
1358 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1359 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1360 SSEPackedDouble>, TB, OpSize;
1361 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1362 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1363 SSEPackedSingle>, TB;
1364 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1365 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1366 SSEPackedDouble>, TB, OpSize;
1367 } // Constraints = "$src1 = $dst"
1368 } // AddedComplexity
1370 //===----------------------------------------------------------------------===//
1371 // SSE 1 & 2 - Extract Floating-Point Sign mask
1372 //===----------------------------------------------------------------------===//
1374 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1375 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1377 def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1378 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1379 [(set GR32:$dst, (Int RC:$src))], d>;
1383 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1384 SSEPackedSingle>, TB;
1385 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1386 SSEPackedDouble>, TB, OpSize;
1388 let isAsmParserOnly = 1 in {
1389 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1390 "movmskps", SSEPackedSingle>, VEX;
1391 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1392 "movmskpd", SSEPackedDouble>, OpSize,
1394 // FIXME: merge with multiclass above when the intrinsics come.
1395 def VMOVMSKPSYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1396 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1397 def VMOVMSKPDYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1398 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1402 //===----------------------------------------------------------------------===//
1403 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1404 //===----------------------------------------------------------------------===//
1406 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1407 // names that start with 'Fs'.
1409 // Alias instructions that map fld0 to pxor for sse.
1410 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1411 canFoldAsLoad = 1 in {
1412 // FIXME: Set encoding to pseudo!
1413 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1414 [(set FR32:$dst, fp32imm0)]>,
1415 Requires<[HasSSE1]>, TB, OpSize;
1416 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1417 [(set FR64:$dst, fpimm0)]>,
1418 Requires<[HasSSE2]>, TB, OpSize;
1421 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1422 // bits are disregarded.
1423 let neverHasSideEffects = 1 in {
1424 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1425 "movaps\t{$src, $dst|$dst, $src}", []>;
1426 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1427 "movapd\t{$src, $dst|$dst, $src}", []>;
1430 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1431 // bits are disregarded.
1432 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1433 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1434 "movaps\t{$src, $dst|$dst, $src}",
1435 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1436 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1437 "movapd\t{$src, $dst|$dst, $src}",
1438 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1441 //===----------------------------------------------------------------------===//
1442 // SSE 1 & 2 - Logical Instructions
1443 //===----------------------------------------------------------------------===//
1445 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1447 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1449 let isAsmParserOnly = 1 in {
1450 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1451 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1453 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1454 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1457 let Constraints = "$src1 = $dst" in {
1458 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1459 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1461 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1462 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1466 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1467 let mayLoad = 0 in {
1468 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1469 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1470 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1473 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1474 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1476 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1478 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1479 SDNode OpNode, int HasPat = 0,
1480 list<list<dag>> Pattern = []> {
1481 let isAsmParserOnly = 1, Pattern = []<dag> in {
1482 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1483 !strconcat(OpcodeStr, "ps"), f128mem,
1484 !if(HasPat, Pattern[0], // rr
1485 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1487 !if(HasPat, Pattern[2], // rm
1488 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1489 (memopv2i64 addr:$src2)))]), 0>,
1492 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1493 !strconcat(OpcodeStr, "pd"), f128mem,
1494 !if(HasPat, Pattern[1], // rr
1495 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1498 !if(HasPat, Pattern[3], // rm
1499 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1500 (memopv2i64 addr:$src2)))]), 0>,
1503 let Constraints = "$src1 = $dst" in {
1504 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1505 !strconcat(OpcodeStr, "ps"), f128mem,
1506 !if(HasPat, Pattern[0], // rr
1507 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1509 !if(HasPat, Pattern[2], // rm
1510 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1511 (memopv2i64 addr:$src2)))])>, TB;
1513 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1514 !strconcat(OpcodeStr, "pd"), f128mem,
1515 !if(HasPat, Pattern[1], // rr
1516 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1519 !if(HasPat, Pattern[3], // rm
1520 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1521 (memopv2i64 addr:$src2)))])>,
1526 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1528 let isAsmParserOnly = 1 in {
1529 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1530 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1531 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1533 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1534 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1538 // AVX 256-bit packed logical ops forms
1539 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1540 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1541 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1542 let isCommutable = 0 in
1543 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1545 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1546 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1547 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1548 let isCommutable = 0 in
1549 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1551 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1552 (bc_v2i64 (v4i32 immAllOnesV))),
1555 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1556 (bc_v2i64 (v2f64 VR128:$src2))))],
1558 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1559 (bc_v2i64 (v4i32 immAllOnesV))),
1560 (memopv2i64 addr:$src2))))],
1562 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1563 (memopv2i64 addr:$src2)))]]>;
1565 //===----------------------------------------------------------------------===//
1566 // SSE 1 & 2 - Arithmetic Instructions
1567 //===----------------------------------------------------------------------===//
1569 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1572 /// In addition, we also have a special variant of the scalar form here to
1573 /// represent the associated intrinsic operation. This form is unlike the
1574 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1575 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1577 /// These three forms can each be reg+reg or reg+mem.
1579 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1581 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1582 OpNode, FR32, f32mem, Is2Addr>, XS;
1583 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1584 OpNode, FR64, f64mem, Is2Addr>, XD;
1587 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1589 let mayLoad = 0 in {
1590 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1591 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1592 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1593 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1597 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1599 let mayLoad = 0 in {
1600 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1601 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1602 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1603 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1607 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1609 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1610 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1611 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1612 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1615 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1617 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1618 !strconcat(OpcodeStr, "ps"), "", "_ps", f128mem, memopv4f32,
1619 SSEPackedSingle, Is2Addr>, TB;
1621 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1622 !strconcat(OpcodeStr, "pd"), "2", "_pd", f128mem, memopv2f64,
1623 SSEPackedDouble, Is2Addr>, TB, OpSize;
1626 // Binary Arithmetic instructions
1627 let isAsmParserOnly = 1 in {
1628 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1629 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1630 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1631 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1632 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1633 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1635 let isCommutable = 0 in {
1636 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1637 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1638 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1639 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1640 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1641 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1642 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1643 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1644 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>, VEX_4V;
1645 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1646 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1647 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1651 let Constraints = "$src1 = $dst" in {
1652 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1653 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1654 basic_sse12_fp_binop_s_int<0x58, "add">;
1655 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1656 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1657 basic_sse12_fp_binop_s_int<0x59, "mul">;
1659 let isCommutable = 0 in {
1660 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1661 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1662 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1663 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1664 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1665 basic_sse12_fp_binop_s_int<0x5E, "div">;
1666 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1667 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1668 basic_sse12_fp_binop_s_int<0x5F, "max">,
1669 basic_sse12_fp_binop_p_int<0x5F, "max">;
1670 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1671 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1672 basic_sse12_fp_binop_s_int<0x5D, "min">,
1673 basic_sse12_fp_binop_p_int<0x5D, "min">;
1678 /// In addition, we also have a special variant of the scalar form here to
1679 /// represent the associated intrinsic operation. This form is unlike the
1680 /// plain scalar form, in that it takes an entire vector (instead of a
1681 /// scalar) and leaves the top elements undefined.
1683 /// And, we have a special variant form for a full-vector intrinsic form.
1685 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1686 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1687 SDNode OpNode, Intrinsic F32Int> {
1688 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1689 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1690 [(set FR32:$dst, (OpNode FR32:$src))]>;
1691 // For scalar unary operations, fold a load into the operation
1692 // only in OptForSize mode. It eliminates an instruction, but it also
1693 // eliminates a whole-register clobber (the load), so it introduces a
1694 // partial register update condition.
1695 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1696 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1697 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1698 Requires<[HasSSE1, OptForSize]>;
1699 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1700 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1701 [(set VR128:$dst, (F32Int VR128:$src))]>;
1702 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1703 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1704 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1707 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1708 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1709 SDNode OpNode, Intrinsic F32Int> {
1710 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1711 !strconcat(!strconcat("v", OpcodeStr),
1712 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1713 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1714 !strconcat(!strconcat("v", OpcodeStr),
1715 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1716 []>, XS, Requires<[HasAVX, OptForSize]>;
1717 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
1718 (ins VR128:$src1, VR128:$src2),
1719 !strconcat(!strconcat("v", OpcodeStr),
1720 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1721 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
1722 (ins VR128:$src1, ssmem:$src2),
1723 !strconcat(!strconcat("v", OpcodeStr),
1724 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1727 /// sse1_fp_unop_p - SSE1 unops in packed form.
1728 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1729 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1730 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1731 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1732 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1733 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1734 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1737 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1738 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1739 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1740 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1741 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1742 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1743 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1744 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1747 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1748 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1749 Intrinsic V4F32Int> {
1750 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1751 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1752 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1753 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1754 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1755 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1759 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1760 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1761 SDNode OpNode, Intrinsic F64Int> {
1762 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1763 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1764 [(set FR64:$dst, (OpNode FR64:$src))]>;
1765 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1766 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1767 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1768 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1769 Requires<[HasSSE2, OptForSize]>;
1770 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1771 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1772 [(set VR128:$dst, (F64Int VR128:$src))]>;
1773 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1774 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1775 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1778 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1779 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1780 SDNode OpNode, Intrinsic F64Int> {
1781 def SDr : VSDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1782 !strconcat(OpcodeStr,
1783 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1784 def SDm : VSDI<opc, MRMSrcMem, (outs FR64:$dst),
1785 (ins FR64:$src1, f64mem:$src2),
1786 !strconcat(OpcodeStr,
1787 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1788 def SDr_Int : VSDI<opc, MRMSrcReg, (outs VR128:$dst),
1789 (ins VR128:$src1, VR128:$src2),
1790 !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1792 def SDm_Int : VSDI<opc, MRMSrcMem, (outs VR128:$dst),
1793 (ins VR128:$src1, sdmem:$src2),
1794 !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1798 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1799 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1801 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1802 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1803 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1804 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1805 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1806 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1809 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1810 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1811 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1812 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1813 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1814 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1815 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1816 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1819 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1820 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1821 Intrinsic V2F64Int> {
1822 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1823 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1824 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1825 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1826 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1827 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1830 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1832 defm VSQRT : sse1_fp_unop_s_avx<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1833 sse2_fp_unop_s_avx<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1836 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1837 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1838 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1839 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1842 // Reciprocal approximations. Note that these typically require refinement
1843 // in order to obtain suitable precision.
1844 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "rsqrt", X86frsqrt,
1845 int_x86_sse_rsqrt_ss>, VEX_4V;
1846 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1847 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>, VEX;
1849 defm VRCP : sse1_fp_unop_s_avx<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1851 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1852 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>, VEX;
1856 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1857 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1858 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1859 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1860 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1861 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1863 // Reciprocal approximations. Note that these typically require refinement
1864 // in order to obtain suitable precision.
1865 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1866 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1867 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1868 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1869 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1870 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1872 // There is no f64 version of the reciprocal approximation instructions.
1874 //===----------------------------------------------------------------------===//
1875 // SSE 1 & 2 - Non-temporal stores
1876 //===----------------------------------------------------------------------===//
1878 let isAsmParserOnly = 1 in {
1879 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1880 (ins i128mem:$dst, VR128:$src),
1881 "movntps\t{$src, $dst|$dst, $src}",
1882 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1883 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1884 (ins i128mem:$dst, VR128:$src),
1885 "movntpd\t{$src, $dst|$dst, $src}",
1886 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1888 let ExeDomain = SSEPackedInt in
1889 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1890 (ins f128mem:$dst, VR128:$src),
1891 "movntdq\t{$src, $dst|$dst, $src}",
1892 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1894 let AddedComplexity = 400 in { // Prefer non-temporal versions
1895 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1896 (ins f128mem:$dst, VR128:$src),
1897 "movntps\t{$src, $dst|$dst, $src}",
1898 [(alignednontemporalstore (v4f32 VR128:$src),
1900 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1901 (ins f128mem:$dst, VR128:$src),
1902 "movntpd\t{$src, $dst|$dst, $src}",
1903 [(alignednontemporalstore (v2f64 VR128:$src),
1905 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1906 (ins f128mem:$dst, VR128:$src),
1907 "movntdq\t{$src, $dst|$dst, $src}",
1908 [(alignednontemporalstore (v2f64 VR128:$src),
1910 let ExeDomain = SSEPackedInt in
1911 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1912 (ins f128mem:$dst, VR128:$src),
1913 "movntdq\t{$src, $dst|$dst, $src}",
1914 [(alignednontemporalstore (v4f32 VR128:$src),
1917 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1918 (ins f256mem:$dst, VR256:$src),
1919 "movntps\t{$src, $dst|$dst, $src}",
1920 [(alignednontemporalstore (v8f32 VR256:$src),
1922 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1923 (ins f256mem:$dst, VR256:$src),
1924 "movntpd\t{$src, $dst|$dst, $src}",
1925 [(alignednontemporalstore (v4f64 VR256:$src),
1927 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1928 (ins f256mem:$dst, VR256:$src),
1929 "movntdq\t{$src, $dst|$dst, $src}",
1930 [(alignednontemporalstore (v4f64 VR256:$src),
1932 let ExeDomain = SSEPackedInt in
1933 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1934 (ins f256mem:$dst, VR256:$src),
1935 "movntdq\t{$src, $dst|$dst, $src}",
1936 [(alignednontemporalstore (v8f32 VR256:$src),
1941 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1942 "movntps\t{$src, $dst|$dst, $src}",
1943 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
1944 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1945 "movntpd\t{$src, $dst|$dst, $src}",
1946 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
1948 let ExeDomain = SSEPackedInt in
1949 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1950 "movntdq\t{$src, $dst|$dst, $src}",
1951 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
1953 let AddedComplexity = 400 in { // Prefer non-temporal versions
1954 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1955 "movntps\t{$src, $dst|$dst, $src}",
1956 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1957 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1958 "movntpd\t{$src, $dst|$dst, $src}",
1959 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1961 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1962 "movntdq\t{$src, $dst|$dst, $src}",
1963 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1965 let ExeDomain = SSEPackedInt in
1966 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1967 "movntdq\t{$src, $dst|$dst, $src}",
1968 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1970 // There is no AVX form for instructions below this point
1971 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1972 "movnti\t{$src, $dst|$dst, $src}",
1973 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
1974 TB, Requires<[HasSSE2]>;
1976 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
1977 "movnti\t{$src, $dst|$dst, $src}",
1978 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
1979 TB, Requires<[HasSSE2]>;
1982 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1983 "movnti\t{$src, $dst|$dst, $src}",
1984 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
1985 TB, Requires<[HasSSE2]>;
1987 //===----------------------------------------------------------------------===//
1988 // SSE 1 & 2 - Misc Instructions (No AVX form)
1989 //===----------------------------------------------------------------------===//
1991 // Prefetch intrinsic.
1992 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
1993 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
1994 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
1995 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
1996 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
1997 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
1998 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
1999 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2001 // Load, store, and memory fence
2002 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2003 TB, Requires<[HasSSE1]>;
2005 // Alias instructions that map zero vector to pxor / xorp* for sse.
2006 // We set canFoldAsLoad because this can be converted to a constant-pool
2007 // load of an all-zeros value if folding it would be beneficial.
2008 // FIXME: Change encoding to pseudo!
2009 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2010 isCodeGenOnly = 1 in {
2011 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2012 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2013 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2014 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2015 let ExeDomain = SSEPackedInt in
2016 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2017 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2020 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2021 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2022 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2024 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2025 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2027 //===----------------------------------------------------------------------===//
2028 // SSE 1 & 2 - Load/Store XCSR register
2029 //===----------------------------------------------------------------------===//
2031 let isAsmParserOnly = 1 in {
2032 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2033 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2034 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2035 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2038 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2039 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2040 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2041 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2043 //===---------------------------------------------------------------------===//
2044 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2045 //===---------------------------------------------------------------------===//
2047 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2049 let isAsmParserOnly = 1 in {
2050 let neverHasSideEffects = 1 in {
2051 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2052 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2053 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2054 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2056 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2057 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2058 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2059 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2061 let canFoldAsLoad = 1, mayLoad = 1 in {
2062 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2063 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2064 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2065 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2066 let Predicates = [HasAVX] in {
2067 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2068 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2069 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2070 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2074 let mayStore = 1 in {
2075 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2076 (ins i128mem:$dst, VR128:$src),
2077 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2078 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2079 (ins i256mem:$dst, VR256:$src),
2080 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2081 let Predicates = [HasAVX] in {
2082 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2083 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2084 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2085 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2090 let neverHasSideEffects = 1 in
2091 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2092 "movdqa\t{$src, $dst|$dst, $src}", []>;
2094 let canFoldAsLoad = 1, mayLoad = 1 in {
2095 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2096 "movdqa\t{$src, $dst|$dst, $src}",
2097 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2098 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2099 "movdqu\t{$src, $dst|$dst, $src}",
2100 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2101 XS, Requires<[HasSSE2]>;
2104 let mayStore = 1 in {
2105 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2106 "movdqa\t{$src, $dst|$dst, $src}",
2107 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2108 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2109 "movdqu\t{$src, $dst|$dst, $src}",
2110 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2111 XS, Requires<[HasSSE2]>;
2114 // Intrinsic forms of MOVDQU load and store
2115 let isAsmParserOnly = 1 in {
2116 let canFoldAsLoad = 1 in
2117 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2118 "vmovdqu\t{$src, $dst|$dst, $src}",
2119 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2120 XS, VEX, Requires<[HasAVX]>;
2121 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2122 "vmovdqu\t{$src, $dst|$dst, $src}",
2123 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2124 XS, VEX, Requires<[HasAVX]>;
2127 let canFoldAsLoad = 1 in
2128 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2129 "movdqu\t{$src, $dst|$dst, $src}",
2130 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2131 XS, Requires<[HasSSE2]>;
2132 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2133 "movdqu\t{$src, $dst|$dst, $src}",
2134 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2135 XS, Requires<[HasSSE2]>;
2137 } // ExeDomain = SSEPackedInt
2139 //===---------------------------------------------------------------------===//
2140 // SSE2 - Packed Integer Arithmetic Instructions
2141 //===---------------------------------------------------------------------===//
2143 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2145 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2146 bit IsCommutable = 0, bit Is2Addr = 1> {
2147 let isCommutable = IsCommutable in
2148 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2149 (ins VR128:$src1, VR128:$src2),
2151 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2152 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2153 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2154 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2155 (ins VR128:$src1, i128mem:$src2),
2157 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2158 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2159 [(set VR128:$dst, (IntId VR128:$src1,
2160 (bitconvert (memopv2i64 addr:$src2))))]>;
2163 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2164 string OpcodeStr, Intrinsic IntId,
2165 Intrinsic IntId2, bit Is2Addr = 1> {
2166 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2167 (ins VR128:$src1, VR128:$src2),
2169 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2170 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2171 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2172 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2173 (ins VR128:$src1, i128mem:$src2),
2175 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2176 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2177 [(set VR128:$dst, (IntId VR128:$src1,
2178 (bitconvert (memopv2i64 addr:$src2))))]>;
2179 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2180 (ins VR128:$src1, i32i8imm:$src2),
2182 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2183 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2184 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2187 /// PDI_binop_rm - Simple SSE2 binary operator.
2188 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2189 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2190 let isCommutable = IsCommutable in
2191 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2192 (ins VR128:$src1, VR128:$src2),
2194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2196 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2197 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2198 (ins VR128:$src1, i128mem:$src2),
2200 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2201 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2202 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2203 (bitconvert (memopv2i64 addr:$src2)))))]>;
2206 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2208 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2209 /// to collapse (bitconvert VT to VT) into its operand.
2211 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2212 bit IsCommutable = 0, bit Is2Addr = 1> {
2213 let isCommutable = IsCommutable in
2214 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2215 (ins VR128:$src1, VR128:$src2),
2217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2219 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2220 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2221 (ins VR128:$src1, i128mem:$src2),
2223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2225 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2228 } // ExeDomain = SSEPackedInt
2230 // 128-bit Integer Arithmetic
2232 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2233 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2234 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2235 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2236 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2237 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2238 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2239 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2240 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2241 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2244 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2246 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2248 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2250 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2252 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2254 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2256 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2258 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2260 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2262 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2264 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2266 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2268 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2270 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2272 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2274 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2276 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2278 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2280 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2284 let Constraints = "$src1 = $dst" in {
2285 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2286 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2287 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2288 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2289 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2290 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2291 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2292 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2293 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2296 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2297 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2298 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2299 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2300 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2301 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2302 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2303 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2304 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2305 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2306 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2307 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2308 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2309 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2310 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2311 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2312 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2313 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2314 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2316 } // Constraints = "$src1 = $dst"
2318 //===---------------------------------------------------------------------===//
2319 // SSE2 - Packed Integer Logical Instructions
2320 //===---------------------------------------------------------------------===//
2322 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2323 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2324 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2326 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2327 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2329 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2330 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2333 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2334 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2336 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2337 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2339 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2340 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2343 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2344 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2346 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2347 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2350 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2351 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2352 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2354 let ExeDomain = SSEPackedInt in {
2355 let neverHasSideEffects = 1 in {
2356 // 128-bit logical shifts.
2357 def VPSLLDQri : PDIi8<0x73, MRM7r,
2358 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2359 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2361 def VPSRLDQri : PDIi8<0x73, MRM3r,
2362 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2363 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2365 // PSRADQri doesn't exist in SSE[1-3].
2367 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2368 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2369 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2370 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2371 VR128:$src2)))]>, VEX_4V;
2373 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2374 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2375 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2376 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2377 (memopv2i64 addr:$src2))))]>,
2382 let Constraints = "$src1 = $dst" in {
2383 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2384 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2385 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2386 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2387 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2388 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2390 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2391 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2392 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2393 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2394 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2395 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2397 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2398 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2399 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2400 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2402 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2403 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2404 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2406 let ExeDomain = SSEPackedInt in {
2407 let neverHasSideEffects = 1 in {
2408 // 128-bit logical shifts.
2409 def PSLLDQri : PDIi8<0x73, MRM7r,
2410 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2411 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2412 def PSRLDQri : PDIi8<0x73, MRM3r,
2413 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2414 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2415 // PSRADQri doesn't exist in SSE[1-3].
2417 def PANDNrr : PDI<0xDF, MRMSrcReg,
2418 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2419 "pandn\t{$src2, $dst|$dst, $src2}",
2420 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2423 def PANDNrm : PDI<0xDF, MRMSrcMem,
2424 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2425 "pandn\t{$src2, $dst|$dst, $src2}",
2426 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2427 (memopv2i64 addr:$src2))))]>;
2429 } // Constraints = "$src1 = $dst"
2431 let Predicates = [HasSSE2] in {
2432 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2433 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2434 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2435 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2436 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2437 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2438 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2439 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2440 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2441 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2443 // Shift up / down and insert zero's.
2444 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2445 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2446 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2447 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2450 //===---------------------------------------------------------------------===//
2451 // SSE2 - Packed Integer Comparison Instructions
2452 //===---------------------------------------------------------------------===//
2454 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2455 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2457 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2459 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2461 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2463 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2465 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2469 let Constraints = "$src1 = $dst" in {
2470 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2471 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2472 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2473 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2474 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2475 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2476 } // Constraints = "$src1 = $dst"
2478 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2479 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2480 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2481 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2482 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2483 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2484 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2485 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2486 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2487 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2488 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2489 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2491 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2492 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2493 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2494 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2495 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2496 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2497 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2498 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2499 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2500 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2501 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2502 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2504 //===---------------------------------------------------------------------===//
2505 // SSE2 - Packed Integer Pack Instructions
2506 //===---------------------------------------------------------------------===//
2508 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2509 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2511 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2513 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2517 let Constraints = "$src1 = $dst" in {
2518 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2519 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2520 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2521 } // Constraints = "$src1 = $dst"
2523 //===---------------------------------------------------------------------===//
2524 // SSE2 - Packed Integer Shuffle Instructions
2525 //===---------------------------------------------------------------------===//
2527 let ExeDomain = SSEPackedInt in {
2528 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2530 def ri : Ii8<0x70, MRMSrcReg,
2531 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2532 !strconcat(OpcodeStr,
2533 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2534 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2536 def mi : Ii8<0x70, MRMSrcMem,
2537 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2538 !strconcat(OpcodeStr,
2539 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2540 [(set VR128:$dst, (vt (pshuf_frag:$src2
2541 (bc_frag (memopv2i64 addr:$src1)),
2544 } // ExeDomain = SSEPackedInt
2546 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2547 let AddedComplexity = 5 in
2548 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2551 // SSE2 with ImmT == Imm8 and XS prefix.
2552 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2555 // SSE2 with ImmT == Imm8 and XD prefix.
2556 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2560 let Predicates = [HasSSE2] in {
2561 let AddedComplexity = 5 in
2562 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2564 // SSE2 with ImmT == Imm8 and XS prefix.
2565 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2567 // SSE2 with ImmT == Imm8 and XD prefix.
2568 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2571 //===---------------------------------------------------------------------===//
2572 // SSE2 - Packed Integer Unpack Instructions
2573 //===---------------------------------------------------------------------===//
2575 let ExeDomain = SSEPackedInt in {
2576 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2577 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2578 def rr : PDI<opc, MRMSrcReg,
2579 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2581 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2582 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2583 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2584 def rm : PDI<opc, MRMSrcMem,
2585 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2587 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2588 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2589 [(set VR128:$dst, (unp_frag VR128:$src1,
2590 (bc_frag (memopv2i64
2594 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2595 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2597 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2599 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2602 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2603 /// knew to collapse (bitconvert VT to VT) into its operand.
2604 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2605 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2606 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2608 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2609 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2610 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2611 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2613 (v2i64 (unpckl VR128:$src1,
2614 (memopv2i64 addr:$src2))))]>, VEX_4V;
2616 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2618 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2620 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2623 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2624 /// knew to collapse (bitconvert VT to VT) into its operand.
2625 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2626 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2627 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2629 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2630 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2631 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2632 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2634 (v2i64 (unpckh VR128:$src1,
2635 (memopv2i64 addr:$src2))))]>, VEX_4V;
2638 let Constraints = "$src1 = $dst" in {
2639 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2640 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2641 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2643 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2644 /// knew to collapse (bitconvert VT to VT) into its operand.
2645 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2646 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2647 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2649 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2650 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2651 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2652 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2654 (v2i64 (unpckl VR128:$src1,
2655 (memopv2i64 addr:$src2))))]>;
2657 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2658 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2659 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2661 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2662 /// knew to collapse (bitconvert VT to VT) into its operand.
2663 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2664 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2665 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2667 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2668 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2669 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2670 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2672 (v2i64 (unpckh VR128:$src1,
2673 (memopv2i64 addr:$src2))))]>;
2676 } // ExeDomain = SSEPackedInt
2678 //===---------------------------------------------------------------------===//
2679 // SSE2 - Packed Integer Extract and Insert
2680 //===---------------------------------------------------------------------===//
2682 let ExeDomain = SSEPackedInt in {
2683 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2684 def rri : Ii8<0xC4, MRMSrcReg,
2685 (outs VR128:$dst), (ins VR128:$src1,
2686 GR32:$src2, i32i8imm:$src3),
2688 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2689 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2691 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2692 def rmi : Ii8<0xC4, MRMSrcMem,
2693 (outs VR128:$dst), (ins VR128:$src1,
2694 i16mem:$src2, i32i8imm:$src3),
2696 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2697 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2699 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2704 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2705 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2706 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2707 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2708 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2709 imm:$src2))]>, OpSize, VEX;
2710 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2711 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2712 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2713 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2717 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2718 defm PINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2720 let Constraints = "$src1 = $dst" in
2721 defm VPINSRW : sse2_pinsrw, TB, OpSize;
2723 } // ExeDomain = SSEPackedInt
2725 //===---------------------------------------------------------------------===//
2726 // SSE2 - Packed Mask Creation
2727 //===---------------------------------------------------------------------===//
2729 let ExeDomain = SSEPackedInt in {
2731 let isAsmParserOnly = 1 in
2732 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2733 "pmovmskb\t{$src, $dst|$dst, $src}",
2734 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2735 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2736 "pmovmskb\t{$src, $dst|$dst, $src}",
2737 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2739 } // ExeDomain = SSEPackedInt
2741 //===---------------------------------------------------------------------===//
2742 // SSE2 - Conditional Store
2743 //===---------------------------------------------------------------------===//
2745 let ExeDomain = SSEPackedInt in {
2747 let isAsmParserOnly = 1 in {
2749 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2750 (ins VR128:$src, VR128:$mask),
2751 "maskmovdqu\t{$mask, $src|$src, $mask}",
2752 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2754 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2755 (ins VR128:$src, VR128:$mask),
2756 "maskmovdqu\t{$mask, $src|$src, $mask}",
2757 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2761 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2762 "maskmovdqu\t{$mask, $src|$src, $mask}",
2763 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2765 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2766 "maskmovdqu\t{$mask, $src|$src, $mask}",
2767 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2769 } // ExeDomain = SSEPackedInt
2771 //===---------------------------------------------------------------------===//
2772 // SSE2 - Move Doubleword
2773 //===---------------------------------------------------------------------===//
2775 // Move Int Doubleword to Packed Double Int
2776 let isAsmParserOnly = 1 in {
2777 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2778 "movd\t{$src, $dst|$dst, $src}",
2780 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2781 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2782 "movd\t{$src, $dst|$dst, $src}",
2784 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2787 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2788 "movd\t{$src, $dst|$dst, $src}",
2790 (v4i32 (scalar_to_vector GR32:$src)))]>;
2791 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2792 "movd\t{$src, $dst|$dst, $src}",
2794 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2797 // Move Int Doubleword to Single Scalar
2798 let isAsmParserOnly = 1 in {
2799 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2800 "movd\t{$src, $dst|$dst, $src}",
2801 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2803 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2804 "movd\t{$src, $dst|$dst, $src}",
2805 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2808 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2809 "movd\t{$src, $dst|$dst, $src}",
2810 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2812 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2813 "movd\t{$src, $dst|$dst, $src}",
2814 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2816 // Move Packed Doubleword Int to Packed Double Int
2817 let isAsmParserOnly = 1 in {
2818 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2819 "movd\t{$src, $dst|$dst, $src}",
2820 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2822 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2823 (ins i32mem:$dst, VR128:$src),
2824 "movd\t{$src, $dst|$dst, $src}",
2825 [(store (i32 (vector_extract (v4i32 VR128:$src),
2826 (iPTR 0))), addr:$dst)]>, VEX;
2828 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2829 "movd\t{$src, $dst|$dst, $src}",
2830 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2832 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2833 "movd\t{$src, $dst|$dst, $src}",
2834 [(store (i32 (vector_extract (v4i32 VR128:$src),
2835 (iPTR 0))), addr:$dst)]>;
2837 // Move Scalar Single to Double Int
2838 let isAsmParserOnly = 1 in {
2839 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2840 "movd\t{$src, $dst|$dst, $src}",
2841 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2842 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2843 "movd\t{$src, $dst|$dst, $src}",
2844 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2846 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2847 "movd\t{$src, $dst|$dst, $src}",
2848 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2849 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2850 "movd\t{$src, $dst|$dst, $src}",
2851 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2853 // movd / movq to XMM register zero-extends
2854 let AddedComplexity = 15, isAsmParserOnly = 1 in {
2855 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2856 "movd\t{$src, $dst|$dst, $src}",
2857 [(set VR128:$dst, (v4i32 (X86vzmovl
2858 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2860 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2861 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2862 [(set VR128:$dst, (v2i64 (X86vzmovl
2863 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2866 let AddedComplexity = 15 in {
2867 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2868 "movd\t{$src, $dst|$dst, $src}",
2869 [(set VR128:$dst, (v4i32 (X86vzmovl
2870 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2871 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2872 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2873 [(set VR128:$dst, (v2i64 (X86vzmovl
2874 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2877 let AddedComplexity = 20 in {
2878 let isAsmParserOnly = 1 in
2879 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2880 "movd\t{$src, $dst|$dst, $src}",
2882 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2883 (loadi32 addr:$src))))))]>,
2885 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2886 "movd\t{$src, $dst|$dst, $src}",
2888 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2889 (loadi32 addr:$src))))))]>;
2891 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2892 (MOVZDI2PDIrm addr:$src)>;
2893 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2894 (MOVZDI2PDIrm addr:$src)>;
2895 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2896 (MOVZDI2PDIrm addr:$src)>;
2899 //===---------------------------------------------------------------------===//
2900 // SSE2 - Move Quadword
2901 //===---------------------------------------------------------------------===//
2903 // Move Quadword Int to Packed Quadword Int
2904 let isAsmParserOnly = 1 in
2905 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2906 "vmovq\t{$src, $dst|$dst, $src}",
2908 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2909 VEX, Requires<[HasAVX]>;
2910 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2911 "movq\t{$src, $dst|$dst, $src}",
2913 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2914 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
2916 // Move Packed Quadword Int to Quadword Int
2917 let isAsmParserOnly = 1 in
2918 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2919 "movq\t{$src, $dst|$dst, $src}",
2920 [(store (i64 (vector_extract (v2i64 VR128:$src),
2921 (iPTR 0))), addr:$dst)]>, VEX;
2922 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2923 "movq\t{$src, $dst|$dst, $src}",
2924 [(store (i64 (vector_extract (v2i64 VR128:$src),
2925 (iPTR 0))), addr:$dst)]>;
2927 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
2928 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
2930 // Store / copy lower 64-bits of a XMM register.
2931 let isAsmParserOnly = 1 in
2932 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2933 "movq\t{$src, $dst|$dst, $src}",
2934 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
2935 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2936 "movq\t{$src, $dst|$dst, $src}",
2937 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
2939 let AddedComplexity = 20, isAsmParserOnly = 1 in
2940 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2941 "vmovq\t{$src, $dst|$dst, $src}",
2943 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
2944 (loadi64 addr:$src))))))]>,
2945 XS, VEX, Requires<[HasAVX]>;
2947 let AddedComplexity = 20 in {
2948 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2949 "movq\t{$src, $dst|$dst, $src}",
2951 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
2952 (loadi64 addr:$src))))))]>,
2953 XS, Requires<[HasSSE2]>;
2955 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
2956 (MOVZQI2PQIrm addr:$src)>;
2957 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
2958 (MOVZQI2PQIrm addr:$src)>;
2959 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
2962 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
2963 // IA32 document. movq xmm1, xmm2 does clear the high bits.
2964 let isAsmParserOnly = 1, AddedComplexity = 15 in
2965 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2966 "vmovq\t{$src, $dst|$dst, $src}",
2967 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
2968 XS, VEX, Requires<[HasAVX]>;
2969 let AddedComplexity = 15 in
2970 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2971 "movq\t{$src, $dst|$dst, $src}",
2972 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
2973 XS, Requires<[HasSSE2]>;
2975 let AddedComplexity = 20, isAsmParserOnly = 1 in
2976 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2977 "vmovq\t{$src, $dst|$dst, $src}",
2978 [(set VR128:$dst, (v2i64 (X86vzmovl
2979 (loadv2i64 addr:$src))))]>,
2980 XS, VEX, Requires<[HasAVX]>;
2981 let AddedComplexity = 20 in {
2982 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2983 "movq\t{$src, $dst|$dst, $src}",
2984 [(set VR128:$dst, (v2i64 (X86vzmovl
2985 (loadv2i64 addr:$src))))]>,
2986 XS, Requires<[HasSSE2]>;
2988 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
2989 (MOVZPQILo2PQIrm addr:$src)>;
2992 // Instructions to match in the assembler
2993 let isAsmParserOnly = 1 in {
2994 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2995 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
2996 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2997 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
2998 // Recognize "movd" with GR64 destination, but encode as a "movq"
2999 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3000 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3003 // Instructions for the disassembler
3004 // xr = XMM register
3007 let isAsmParserOnly = 1 in
3008 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3009 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3010 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3011 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3013 //===---------------------------------------------------------------------===//
3014 // SSE2 - Misc Instructions
3015 //===---------------------------------------------------------------------===//
3018 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3019 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3020 TB, Requires<[HasSSE2]>;
3022 // Load, store, and memory fence
3023 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3024 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3025 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3026 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3028 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3029 // was introduced with SSE2, it's backward compatible.
3030 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3032 //TODO: custom lower this so as to never even generate the noop
3033 def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
3035 def : Pat<(membarrier (i8 0), (i8 0), (i8 0), (i8 1), (i8 1)), (SFENCE)>;
3036 def : Pat<(membarrier (i8 1), (i8 0), (i8 0), (i8 0), (i8 1)), (LFENCE)>;
3037 def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
3040 // Alias instructions that map zero vector to pxor / xorp* for sse.
3041 // We set canFoldAsLoad because this can be converted to a constant-pool
3042 // load of an all-ones value if folding it would be beneficial.
3043 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3044 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3045 // FIXME: Change encoding to pseudo.
3046 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3047 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3049 //===---------------------------------------------------------------------===//
3050 // SSE3 - Conversion Instructions
3051 //===---------------------------------------------------------------------===//
3053 // Convert Packed Double FP to Packed DW Integers
3054 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3055 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3056 // register, but the same isn't true when using memory operands instead.
3057 // Provide other assembly rr and rm forms to address this explicitly.
3058 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3059 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3060 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3061 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3064 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3065 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3066 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3067 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3070 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3071 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3072 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3073 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3076 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3077 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3078 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3079 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3081 // Convert Packed DW Integers to Packed Double FP
3082 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3083 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3084 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3085 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3086 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3087 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3088 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3089 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3090 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3093 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3094 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3095 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3096 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3098 //===---------------------------------------------------------------------===//
3099 // SSE3 - Move Instructions
3100 //===---------------------------------------------------------------------===//
3102 // Replicate Single FP
3103 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3104 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3105 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3106 [(set VR128:$dst, (v4f32 (rep_frag
3107 VR128:$src, (undef))))]>;
3108 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3109 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3110 [(set VR128:$dst, (rep_frag
3111 (memopv4f32 addr:$src), (undef)))]>;
3114 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3116 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3117 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3118 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3119 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3122 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3123 // FIXME: Merge above classes when we have patterns for the ymm version
3124 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3125 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3126 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3127 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3129 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3130 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3132 // Replicate Double FP
3133 multiclass sse3_replicate_dfp<string OpcodeStr> {
3134 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3135 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3136 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3137 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3138 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3140 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3144 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3145 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3146 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3148 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3149 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3153 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3154 // FIXME: Merge above classes when we have patterns for the ymm version
3155 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3156 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3158 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3160 // Move Unaligned Integer
3161 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3162 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3163 "vlddqu\t{$src, $dst|$dst, $src}",
3164 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3165 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3166 "vlddqu\t{$src, $dst|$dst, $src}", []>, VEX;
3168 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3169 "lddqu\t{$src, $dst|$dst, $src}",
3170 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3172 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3174 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3176 // Several Move patterns
3177 let AddedComplexity = 5 in {
3178 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3179 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3180 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3181 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3182 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3183 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3184 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3185 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3188 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3189 let AddedComplexity = 15 in
3190 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3191 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3192 let AddedComplexity = 20 in
3193 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3194 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3196 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3197 let AddedComplexity = 15 in
3198 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3199 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3200 let AddedComplexity = 20 in
3201 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3202 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3204 //===---------------------------------------------------------------------===//
3205 // SSE3 - Arithmetic
3206 //===---------------------------------------------------------------------===//
3208 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3209 X86MemOperand x86memop, bit Is2Addr = 1> {
3210 def rr : I<0xD0, MRMSrcReg,
3211 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3213 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3214 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3215 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3216 def rm : I<0xD0, MRMSrcMem,
3217 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3219 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3220 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3221 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3224 let isAsmParserOnly = 1, Predicates = [HasAVX],
3225 ExeDomain = SSEPackedDouble in {
3226 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3227 f128mem, 0>, XD, VEX_4V;
3228 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3229 f128mem, 0>, OpSize, VEX_4V;
3230 let Pattern = []<dag> in {
3231 defm VADDSUBPSY : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR256,
3232 f256mem, 0>, XD, VEX_4V;
3233 defm VADDSUBPDY : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR256,
3234 f256mem, 0>, OpSize, VEX_4V;
3237 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3238 ExeDomain = SSEPackedDouble in {
3239 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3241 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3242 f128mem>, TB, OpSize;
3245 //===---------------------------------------------------------------------===//
3246 // SSE3 Instructions
3247 //===---------------------------------------------------------------------===//
3250 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3251 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3252 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3254 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3255 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3256 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3258 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3260 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3261 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3262 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3264 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3265 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3266 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3268 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3269 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3270 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3272 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3274 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3275 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3276 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3279 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3280 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3281 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3282 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3283 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3284 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3285 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3286 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3287 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3288 let Pattern = []<dag> in {
3289 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3290 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3291 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3292 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3293 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3294 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3295 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3296 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3300 let Constraints = "$src1 = $dst" in {
3301 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3302 int_x86_sse3_hadd_ps>;
3303 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3304 int_x86_sse3_hadd_pd>;
3305 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3306 int_x86_sse3_hsub_ps>;
3307 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3308 int_x86_sse3_hsub_pd>;
3311 //===---------------------------------------------------------------------===//
3312 // SSSE3 - Packed Absolute Instructions
3313 //===---------------------------------------------------------------------===//
3315 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3316 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3317 PatFrag mem_frag64, PatFrag mem_frag128,
3318 Intrinsic IntId64, Intrinsic IntId128> {
3319 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
3320 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3321 [(set VR64:$dst, (IntId64 VR64:$src))]>;
3323 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
3324 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3326 (IntId64 (bitconvert (mem_frag64 addr:$src))))]>;
3328 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3330 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3331 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3334 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3339 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3342 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3343 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv8i8, memopv16i8,
3344 int_x86_ssse3_pabs_b,
3345 int_x86_ssse3_pabs_b_128>, VEX;
3346 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv4i16, memopv8i16,
3347 int_x86_ssse3_pabs_w,
3348 int_x86_ssse3_pabs_w_128>, VEX;
3349 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv2i32, memopv4i32,
3350 int_x86_ssse3_pabs_d,
3351 int_x86_ssse3_pabs_d_128>, VEX;
3354 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv8i8, memopv16i8,
3355 int_x86_ssse3_pabs_b,
3356 int_x86_ssse3_pabs_b_128>;
3357 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv4i16, memopv8i16,
3358 int_x86_ssse3_pabs_w,
3359 int_x86_ssse3_pabs_w_128>;
3360 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv2i32, memopv4i32,
3361 int_x86_ssse3_pabs_d,
3362 int_x86_ssse3_pabs_d_128>;
3364 //===---------------------------------------------------------------------===//
3365 // SSSE3 - Packed Binary Operator Instructions
3366 //===---------------------------------------------------------------------===//
3368 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3369 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3370 PatFrag mem_frag64, PatFrag mem_frag128,
3371 Intrinsic IntId64, Intrinsic IntId128,
3373 let isCommutable = 1 in
3374 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
3375 (ins VR64:$src1, VR64:$src2),
3377 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3379 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
3380 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
3381 (ins VR64:$src1, i64mem:$src2),
3383 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3384 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3386 (IntId64 VR64:$src1,
3387 (bitconvert (memopv8i8 addr:$src2))))]>;
3389 let isCommutable = 1 in
3390 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3391 (ins VR128:$src1, VR128:$src2),
3393 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3394 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3395 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3397 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3398 (ins VR128:$src1, i128mem:$src2),
3400 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3401 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3403 (IntId128 VR128:$src1,
3404 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3407 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3408 let isCommutable = 0 in {
3409 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv4i16, memopv8i16,
3410 int_x86_ssse3_phadd_w,
3411 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3412 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv2i32, memopv4i32,
3413 int_x86_ssse3_phadd_d,
3414 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3415 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv4i16, memopv8i16,
3416 int_x86_ssse3_phadd_sw,
3417 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3418 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv4i16, memopv8i16,
3419 int_x86_ssse3_phsub_w,
3420 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3421 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv2i32, memopv4i32,
3422 int_x86_ssse3_phsub_d,
3423 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3424 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv4i16, memopv8i16,
3425 int_x86_ssse3_phsub_sw,
3426 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3427 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv8i8, memopv16i8,
3428 int_x86_ssse3_pmadd_ub_sw,
3429 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3430 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv8i8, memopv16i8,
3431 int_x86_ssse3_pshuf_b,
3432 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3433 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv8i8, memopv16i8,
3434 int_x86_ssse3_psign_b,
3435 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3436 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv4i16, memopv8i16,
3437 int_x86_ssse3_psign_w,
3438 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3439 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv2i32, memopv4i32,
3440 int_x86_ssse3_psign_d,
3441 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3443 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv4i16, memopv8i16,
3444 int_x86_ssse3_pmul_hr_sw,
3445 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3448 // None of these have i8 immediate fields.
3449 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3450 let isCommutable = 0 in {
3451 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv4i16, memopv8i16,
3452 int_x86_ssse3_phadd_w,
3453 int_x86_ssse3_phadd_w_128>;
3454 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv2i32, memopv4i32,
3455 int_x86_ssse3_phadd_d,
3456 int_x86_ssse3_phadd_d_128>;
3457 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv4i16, memopv8i16,
3458 int_x86_ssse3_phadd_sw,
3459 int_x86_ssse3_phadd_sw_128>;
3460 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv4i16, memopv8i16,
3461 int_x86_ssse3_phsub_w,
3462 int_x86_ssse3_phsub_w_128>;
3463 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv2i32, memopv4i32,
3464 int_x86_ssse3_phsub_d,
3465 int_x86_ssse3_phsub_d_128>;
3466 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv4i16, memopv8i16,
3467 int_x86_ssse3_phsub_sw,
3468 int_x86_ssse3_phsub_sw_128>;
3469 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv8i8, memopv16i8,
3470 int_x86_ssse3_pmadd_ub_sw,
3471 int_x86_ssse3_pmadd_ub_sw_128>;
3472 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv8i8, memopv16i8,
3473 int_x86_ssse3_pshuf_b,
3474 int_x86_ssse3_pshuf_b_128>;
3475 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv8i8, memopv16i8,
3476 int_x86_ssse3_psign_b,
3477 int_x86_ssse3_psign_b_128>;
3478 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv4i16, memopv8i16,
3479 int_x86_ssse3_psign_w,
3480 int_x86_ssse3_psign_w_128>;
3481 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv2i32, memopv4i32,
3482 int_x86_ssse3_psign_d,
3483 int_x86_ssse3_psign_d_128>;
3485 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv4i16, memopv8i16,
3486 int_x86_ssse3_pmul_hr_sw,
3487 int_x86_ssse3_pmul_hr_sw_128>;
3490 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3491 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3492 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3493 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3495 //===---------------------------------------------------------------------===//
3496 // SSSE3 - Packed Align Instruction Patterns
3497 //===---------------------------------------------------------------------===//
3499 multiclass sse3_palign<string asm, bit Is2Addr = 1> {
3500 def R64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
3501 (ins VR64:$src1, VR64:$src2, i8imm:$src3),
3503 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3505 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3507 def R64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
3508 (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
3510 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3512 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3515 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3516 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3518 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3520 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3522 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3523 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3525 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3527 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3531 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3532 defm VPALIGN : sse3_palign<"vpalignr", 0>, VEX_4V;
3533 let Constraints = "$src1 = $dst" in
3534 defm PALIGN : sse3_palign<"palignr">;
3536 let AddedComplexity = 5 in {
3538 def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
3539 (PALIGNR64rr VR64:$src2, VR64:$src1,
3540 (SHUFFLE_get_palign_imm VR64:$src3))>,
3541 Requires<[HasSSSE3]>;
3542 def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
3543 (PALIGNR64rr VR64:$src2, VR64:$src1,
3544 (SHUFFLE_get_palign_imm VR64:$src3))>,
3545 Requires<[HasSSSE3]>;
3546 def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
3547 (PALIGNR64rr VR64:$src2, VR64:$src1,
3548 (SHUFFLE_get_palign_imm VR64:$src3))>,
3549 Requires<[HasSSSE3]>;
3550 def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
3551 (PALIGNR64rr VR64:$src2, VR64:$src1,
3552 (SHUFFLE_get_palign_imm VR64:$src3))>,
3553 Requires<[HasSSSE3]>;
3555 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3556 (PALIGNR128rr VR128:$src2, VR128:$src1,
3557 (SHUFFLE_get_palign_imm VR128:$src3))>,
3558 Requires<[HasSSSE3]>;
3559 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3560 (PALIGNR128rr VR128:$src2, VR128:$src1,
3561 (SHUFFLE_get_palign_imm VR128:$src3))>,
3562 Requires<[HasSSSE3]>;
3563 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3564 (PALIGNR128rr VR128:$src2, VR128:$src1,
3565 (SHUFFLE_get_palign_imm VR128:$src3))>,
3566 Requires<[HasSSSE3]>;
3567 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3568 (PALIGNR128rr VR128:$src2, VR128:$src1,
3569 (SHUFFLE_get_palign_imm VR128:$src3))>,
3570 Requires<[HasSSSE3]>;
3573 //===---------------------------------------------------------------------===//
3574 // SSSE3 Misc Instructions
3575 //===---------------------------------------------------------------------===//
3577 // Thread synchronization
3578 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3579 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3580 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3581 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3583 //===---------------------------------------------------------------------===//
3584 // Non-Instruction Patterns
3585 //===---------------------------------------------------------------------===//
3587 // extload f32 -> f64. This matches load+fextend because we have a hack in
3588 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3590 // Since these loads aren't folded into the fextend, we have to match it
3592 let Predicates = [HasSSE2] in
3593 def : Pat<(fextend (loadf32 addr:$src)),
3594 (CVTSS2SDrm addr:$src)>;
3597 let Predicates = [HasSSE2] in {
3598 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3599 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3600 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3601 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3602 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3603 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3604 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3605 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3606 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3607 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3608 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3609 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3610 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3611 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3612 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3613 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3614 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3615 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3616 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3617 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3618 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3619 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3620 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3621 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3622 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3623 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3624 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3625 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3626 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3627 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3630 // Move scalar to XMM zero-extended
3631 // movd to XMM register zero-extends
3632 let AddedComplexity = 15 in {
3633 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3634 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3635 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3636 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3637 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3638 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3639 (MOVSSrr (v4f32 (V_SET0PS)),
3640 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3641 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3642 (MOVSSrr (v4i32 (V_SET0PI)),
3643 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3646 // Splat v2f64 / v2i64
3647 let AddedComplexity = 10 in {
3648 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3649 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3650 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3651 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3652 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3653 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3654 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3655 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3658 // Special unary SHUFPSrri case.
3659 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3660 (SHUFPSrri VR128:$src1, VR128:$src1,
3661 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3662 let AddedComplexity = 5 in
3663 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3664 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3665 Requires<[HasSSE2]>;
3666 // Special unary SHUFPDrri case.
3667 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3668 (SHUFPDrri VR128:$src1, VR128:$src1,
3669 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3670 Requires<[HasSSE2]>;
3671 // Special unary SHUFPDrri case.
3672 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3673 (SHUFPDrri VR128:$src1, VR128:$src1,
3674 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3675 Requires<[HasSSE2]>;
3676 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3677 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3678 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3679 Requires<[HasSSE2]>;
3681 // Special binary v4i32 shuffle cases with SHUFPS.
3682 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3683 (SHUFPSrri VR128:$src1, VR128:$src2,
3684 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3685 Requires<[HasSSE2]>;
3686 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3687 (SHUFPSrmi VR128:$src1, addr:$src2,
3688 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3689 Requires<[HasSSE2]>;
3690 // Special binary v2i64 shuffle cases using SHUFPDrri.
3691 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3692 (SHUFPDrri VR128:$src1, VR128:$src2,
3693 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3694 Requires<[HasSSE2]>;
3696 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3697 let AddedComplexity = 15 in {
3698 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3699 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3700 Requires<[OptForSpeed, HasSSE2]>;
3701 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3702 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3703 Requires<[OptForSpeed, HasSSE2]>;
3705 let AddedComplexity = 10 in {
3706 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3707 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3708 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3709 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3710 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3711 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3712 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3713 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3716 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3717 let AddedComplexity = 15 in {
3718 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3719 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3720 Requires<[OptForSpeed, HasSSE2]>;
3721 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3722 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3723 Requires<[OptForSpeed, HasSSE2]>;
3725 let AddedComplexity = 10 in {
3726 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3727 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3728 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3729 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3730 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3731 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3732 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3733 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3736 let AddedComplexity = 20 in {
3737 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3738 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3739 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3741 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3742 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3743 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3745 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3746 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3747 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3748 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3749 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3752 let AddedComplexity = 20 in {
3753 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3754 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3755 (MOVLPSrm VR128:$src1, addr:$src2)>;
3756 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3757 (MOVLPDrm VR128:$src1, addr:$src2)>;
3758 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3759 (MOVLPSrm VR128:$src1, addr:$src2)>;
3760 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3761 (MOVLPDrm VR128:$src1, addr:$src2)>;
3764 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3765 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3766 (MOVLPSmr addr:$src1, VR128:$src2)>;
3767 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3768 (MOVLPDmr addr:$src1, VR128:$src2)>;
3769 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3771 (MOVLPSmr addr:$src1, VR128:$src2)>;
3772 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3773 (MOVLPDmr addr:$src1, VR128:$src2)>;
3775 let AddedComplexity = 15 in {
3776 // Setting the lowest element in the vector.
3777 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3778 (MOVSSrr (v4i32 VR128:$src1),
3779 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3780 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3781 (MOVSDrr (v2i64 VR128:$src1),
3782 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3784 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3785 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3786 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3787 Requires<[HasSSE2]>;
3788 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3789 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3790 Requires<[HasSSE2]>;
3793 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3794 // fall back to this for SSE1)
3795 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3796 (SHUFPSrri VR128:$src2, VR128:$src1,
3797 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3799 // Set lowest element and zero upper elements.
3800 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3801 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3803 // Some special case pandn patterns.
3804 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3806 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3807 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3809 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3810 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3812 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3814 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3815 (memop addr:$src2))),
3816 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3817 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3818 (memop addr:$src2))),
3819 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3820 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3821 (memop addr:$src2))),
3822 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3824 // vector -> vector casts
3825 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3826 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3827 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3828 (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3829 def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
3830 (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
3831 def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
3832 (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
3834 // Use movaps / movups for SSE integer load / store (one byte shorter).
3835 def : Pat<(alignedloadv4i32 addr:$src),
3836 (MOVAPSrm addr:$src)>;
3837 def : Pat<(loadv4i32 addr:$src),
3838 (MOVUPSrm addr:$src)>;
3839 def : Pat<(alignedloadv2i64 addr:$src),
3840 (MOVAPSrm addr:$src)>;
3841 def : Pat<(loadv2i64 addr:$src),
3842 (MOVUPSrm addr:$src)>;
3844 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3845 (MOVAPSmr addr:$dst, VR128:$src)>;
3846 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3847 (MOVAPSmr addr:$dst, VR128:$src)>;
3848 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3849 (MOVAPSmr addr:$dst, VR128:$src)>;
3850 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3851 (MOVAPSmr addr:$dst, VR128:$src)>;
3852 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3853 (MOVUPSmr addr:$dst, VR128:$src)>;
3854 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3855 (MOVUPSmr addr:$dst, VR128:$src)>;
3856 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3857 (MOVUPSmr addr:$dst, VR128:$src)>;
3858 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3859 (MOVUPSmr addr:$dst, VR128:$src)>;
3861 //===----------------------------------------------------------------------===//
3862 // SSE4.1 - Packed Move with Sign/Zero Extend
3863 //===----------------------------------------------------------------------===//
3865 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3866 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3867 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3868 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3870 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3871 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3873 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3877 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3878 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3880 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3882 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3884 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3886 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3888 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3892 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3893 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3894 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3895 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3896 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3897 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3899 // Common patterns involving scalar load.
3900 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3901 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3902 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3903 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3905 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3906 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3907 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3908 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3910 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3911 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3912 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3913 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3915 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3916 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3917 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3918 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3920 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3921 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3922 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3923 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3925 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3926 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3927 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3928 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3931 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3932 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3933 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3934 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3936 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3937 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3939 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
3943 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3944 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
3946 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
3948 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
3950 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
3954 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
3955 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
3956 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
3957 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
3959 // Common patterns involving scalar load
3960 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
3961 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
3962 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
3963 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
3965 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
3966 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
3967 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
3968 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
3971 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3972 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3973 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3974 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3976 // Expecting a i16 load any extended to i32 value.
3977 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
3978 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3979 [(set VR128:$dst, (IntId (bitconvert
3980 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
3984 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3985 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
3987 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
3990 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
3991 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
3993 // Common patterns involving scalar load
3994 def : Pat<(int_x86_sse41_pmovsxbq
3995 (bitconvert (v4i32 (X86vzmovl
3996 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
3997 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
3999 def : Pat<(int_x86_sse41_pmovzxbq
4000 (bitconvert (v4i32 (X86vzmovl
4001 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4002 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4004 //===----------------------------------------------------------------------===//
4005 // SSE4.1 - Extract Instructions
4006 //===----------------------------------------------------------------------===//
4008 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4009 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4010 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4011 (ins VR128:$src1, i32i8imm:$src2),
4012 !strconcat(OpcodeStr,
4013 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4014 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4016 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4017 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4018 !strconcat(OpcodeStr,
4019 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4022 // There's an AssertZext in the way of writing the store pattern
4023 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4026 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4027 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4029 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4032 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4033 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4034 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4035 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4036 !strconcat(OpcodeStr,
4037 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4040 // There's an AssertZext in the way of writing the store pattern
4041 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4044 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4045 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4047 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4050 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4051 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4052 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4053 (ins VR128:$src1, i32i8imm:$src2),
4054 !strconcat(OpcodeStr,
4055 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4057 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4058 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4059 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4060 !strconcat(OpcodeStr,
4061 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4062 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4063 addr:$dst)]>, OpSize;
4066 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4067 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4069 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4071 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4072 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4073 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4074 (ins VR128:$src1, i32i8imm:$src2),
4075 !strconcat(OpcodeStr,
4076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4078 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4079 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4080 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4081 !strconcat(OpcodeStr,
4082 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4083 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4084 addr:$dst)]>, OpSize, REX_W;
4087 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4088 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4090 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4092 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4094 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4095 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4096 (ins VR128:$src1, i32i8imm:$src2),
4097 !strconcat(OpcodeStr,
4098 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4100 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4102 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4103 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4104 !strconcat(OpcodeStr,
4105 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4106 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4107 addr:$dst)]>, OpSize;
4110 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4111 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4112 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4113 (ins VR128:$src1, i32i8imm:$src2),
4114 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4117 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4119 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4120 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4123 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4124 Requires<[HasSSE41]>;
4126 //===----------------------------------------------------------------------===//
4127 // SSE4.1 - Insert Instructions
4128 //===----------------------------------------------------------------------===//
4130 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4131 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4132 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4134 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4136 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4138 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4139 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4140 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4142 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4144 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4146 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4147 imm:$src3))]>, OpSize;
4150 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4151 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4152 let Constraints = "$src1 = $dst" in
4153 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4155 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4156 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4157 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4159 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4161 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4163 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4165 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4166 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4168 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4170 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4172 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4173 imm:$src3)))]>, OpSize;
4176 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4177 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4178 let Constraints = "$src1 = $dst" in
4179 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4181 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4182 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4183 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4185 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4187 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4189 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4191 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4192 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4194 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4196 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4198 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4199 imm:$src3)))]>, OpSize;
4202 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4203 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4204 let Constraints = "$src1 = $dst" in
4205 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4207 // insertps has a few different modes, there's the first two here below which
4208 // are optimized inserts that won't zero arbitrary elements in the destination
4209 // vector. The next one matches the intrinsic and could zero arbitrary elements
4210 // in the target vector.
4211 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4212 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4213 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4215 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4217 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4219 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4221 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4222 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4224 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4226 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4228 (X86insrtps VR128:$src1,
4229 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4230 imm:$src3))]>, OpSize;
4233 let Constraints = "$src1 = $dst" in
4234 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4235 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4236 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4238 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4239 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>;
4241 //===----------------------------------------------------------------------===//
4242 // SSE4.1 - Round Instructions
4243 //===----------------------------------------------------------------------===//
4245 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd,
4248 Intrinsic V2F64Int> {
4249 // Intrinsic operation, reg.
4250 // Vector intrinsic operation, reg
4251 def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
4252 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4253 !strconcat(OpcodeStr,
4254 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4255 [(set VR128:$dst, (V4F32Int VR128:$src1, imm:$src2))]>,
4258 // Vector intrinsic operation, mem
4259 def PSm_Int : Ii8<opcps, MRMSrcMem,
4260 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4261 !strconcat(OpcodeStr,
4262 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4264 (V4F32Int (memopv4f32 addr:$src1),imm:$src2))]>,
4266 Requires<[HasSSE41]>;
4268 // Vector intrinsic operation, reg
4269 def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
4270 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4271 !strconcat(OpcodeStr,
4272 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4273 [(set VR128:$dst, (V2F64Int VR128:$src1, imm:$src2))]>,
4276 // Vector intrinsic operation, mem
4277 def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
4278 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4279 !strconcat(OpcodeStr,
4280 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4282 (V2F64Int (memopv2f64 addr:$src1),imm:$src2))]>,
4286 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4287 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4288 // Intrinsic operation, reg.
4289 // Vector intrinsic operation, reg
4290 def PSr : SS4AIi8<opcps, MRMSrcReg,
4291 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4292 !strconcat(OpcodeStr,
4293 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4296 // Vector intrinsic operation, mem
4297 def PSm : Ii8<opcps, MRMSrcMem,
4298 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4299 !strconcat(OpcodeStr,
4300 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4301 []>, TA, OpSize, Requires<[HasSSE41]>;
4303 // Vector intrinsic operation, reg
4304 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4305 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4306 !strconcat(OpcodeStr,
4307 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4310 // Vector intrinsic operation, mem
4311 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4312 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4313 !strconcat(OpcodeStr,
4314 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4318 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4321 Intrinsic F64Int, bit Is2Addr = 1> {
4322 // Intrinsic operation, reg.
4323 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
4324 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4326 !strconcat(OpcodeStr,
4327 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4328 !strconcat(OpcodeStr,
4329 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4330 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4333 // Intrinsic operation, mem.
4334 def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
4335 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4337 !strconcat(OpcodeStr,
4338 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4339 !strconcat(OpcodeStr,
4340 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4342 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4345 // Intrinsic operation, reg.
4346 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
4347 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4349 !strconcat(OpcodeStr,
4350 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4351 !strconcat(OpcodeStr,
4352 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4353 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4356 // Intrinsic operation, mem.
4357 def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
4358 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4360 !strconcat(OpcodeStr,
4361 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4362 !strconcat(OpcodeStr,
4363 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4365 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4369 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4371 // Intrinsic operation, reg.
4372 def SSr : SS4AIi8<opcss, MRMSrcReg,
4373 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4374 !strconcat(OpcodeStr,
4375 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4378 // Intrinsic operation, mem.
4379 def SSm : SS4AIi8<opcss, MRMSrcMem,
4380 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4381 !strconcat(OpcodeStr,
4382 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4385 // Intrinsic operation, reg.
4386 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4387 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4388 !strconcat(OpcodeStr,
4389 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4392 // Intrinsic operation, mem.
4393 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4394 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4395 !strconcat(OpcodeStr,
4396 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4400 // FP round - roundss, roundps, roundsd, roundpd
4401 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4403 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround",
4404 int_x86_sse41_round_ps, int_x86_sse41_round_pd>,
4406 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4407 int_x86_sse41_round_ss, int_x86_sse41_round_sd,
4409 // Instructions for the assembler
4410 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4412 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4414 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4417 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round",
4418 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4419 let Constraints = "$src1 = $dst" in
4420 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4421 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4423 //===----------------------------------------------------------------------===//
4424 // SSE4.1 - Packed Bit Test
4425 //===----------------------------------------------------------------------===//
4427 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4428 // the intel intrinsic that corresponds to this.
4429 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4430 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4431 "vptest\t{$src2, $src1|$src1, $src2}",
4432 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4434 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4435 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4437 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4438 "vptest\t{$src2, $src1|$src1, $src2}",
4439 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4441 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4442 "vptest\t{$src2, $src1|$src1, $src2}", []>, OpSize, VEX;
4445 let Defs = [EFLAGS] in {
4446 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4447 "ptest \t{$src2, $src1|$src1, $src2}",
4448 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4450 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4451 "ptest \t{$src2, $src1|$src1, $src2}",
4452 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4456 // The bit test instructions below are AVX only
4457 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4458 X86MemOperand x86memop> {
4459 def rr : SS48I<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4460 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4462 def rm : SS48I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4463 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4467 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4468 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem>;
4469 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem>;
4470 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem>;
4471 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem>;
4474 //===----------------------------------------------------------------------===//
4475 // SSE4.1 - Misc Instructions
4476 //===----------------------------------------------------------------------===//
4478 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4479 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4480 Intrinsic IntId128> {
4481 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4483 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4484 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4485 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4487 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4490 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4493 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4494 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4495 int_x86_sse41_phminposuw>, VEX;
4496 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4497 int_x86_sse41_phminposuw>;
4499 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4500 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4501 Intrinsic IntId128, bit Is2Addr = 1> {
4502 let isCommutable = 1 in
4503 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4504 (ins VR128:$src1, VR128:$src2),
4506 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4507 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4508 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4509 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4510 (ins VR128:$src1, i128mem:$src2),
4512 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4513 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4515 (IntId128 VR128:$src1,
4516 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4519 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4520 let isCommutable = 0 in
4521 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4523 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4525 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4527 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4529 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4531 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4533 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4535 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4537 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4539 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4541 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4545 let Constraints = "$src1 = $dst" in {
4546 let isCommutable = 0 in
4547 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4548 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4549 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4550 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4551 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4552 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4553 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4554 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4555 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4556 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4557 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4560 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4561 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4562 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4563 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4565 /// SS48I_binop_rm - Simple SSE41 binary operator.
4566 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4567 ValueType OpVT, bit Is2Addr = 1> {
4568 let isCommutable = 1 in
4569 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4570 (ins VR128:$src1, VR128:$src2),
4572 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4573 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4574 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4576 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4577 (ins VR128:$src1, i128mem:$src2),
4579 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4580 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4581 [(set VR128:$dst, (OpNode VR128:$src1,
4582 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4586 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4587 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4588 let Constraints = "$src1 = $dst" in
4589 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4591 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4592 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4593 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4594 X86MemOperand x86memop, bit Is2Addr = 1> {
4595 let isCommutable = 1 in
4596 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4597 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4599 !strconcat(OpcodeStr,
4600 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4601 !strconcat(OpcodeStr,
4602 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4603 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4605 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4606 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4608 !strconcat(OpcodeStr,
4609 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4610 !strconcat(OpcodeStr,
4611 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4614 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4618 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4619 let isCommutable = 0 in {
4620 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4621 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4622 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4623 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4624 let Pattern = []<dag> in {
4625 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4626 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4627 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4628 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4630 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4631 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4632 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4633 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4635 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4636 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4637 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4638 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4639 let Pattern = []<dag> in
4640 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4641 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4644 let Constraints = "$src1 = $dst" in {
4645 let isCommutable = 0 in {
4646 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4647 VR128, memopv16i8, i128mem>;
4648 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4649 VR128, memopv16i8, i128mem>;
4650 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4651 VR128, memopv16i8, i128mem>;
4652 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4653 VR128, memopv16i8, i128mem>;
4655 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4656 VR128, memopv16i8, i128mem>;
4657 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4658 VR128, memopv16i8, i128mem>;
4661 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4662 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4663 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4664 RegisterClass RC, X86MemOperand x86memop> {
4665 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4666 (ins RC:$src1, RC:$src2, RC:$src3),
4667 !strconcat(OpcodeStr,
4668 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4669 [], SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4671 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4672 (ins RC:$src1, x86memop:$src2, RC:$src3),
4673 !strconcat(OpcodeStr,
4674 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4675 [], SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4679 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem>;
4680 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem>;
4681 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem>;
4682 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem>;
4684 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem>;
4686 /// SS41I_ternary_int - SSE 4.1 ternary operator
4687 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4688 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4689 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4690 (ins VR128:$src1, VR128:$src2),
4691 !strconcat(OpcodeStr,
4692 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4693 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4696 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4697 (ins VR128:$src1, i128mem:$src2),
4698 !strconcat(OpcodeStr,
4699 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4702 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4706 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4707 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4708 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4710 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4711 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4712 "vmovntdqa\t{$src, $dst|$dst, $src}",
4713 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4715 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4716 "movntdqa\t{$src, $dst|$dst, $src}",
4717 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4720 //===----------------------------------------------------------------------===//
4721 // SSE4.2 - Compare Instructions
4722 //===----------------------------------------------------------------------===//
4724 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4725 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4726 Intrinsic IntId128, bit Is2Addr = 1> {
4727 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4728 (ins VR128:$src1, VR128:$src2),
4730 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4731 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4732 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4734 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4735 (ins VR128:$src1, i128mem:$src2),
4737 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4738 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4740 (IntId128 VR128:$src1,
4741 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4744 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4745 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4747 let Constraints = "$src1 = $dst" in
4748 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4750 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4751 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4752 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4753 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4755 //===----------------------------------------------------------------------===//
4756 // SSE4.2 - String/text Processing Instructions
4757 //===----------------------------------------------------------------------===//
4759 // Packed Compare Implicit Length Strings, Return Mask
4760 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4761 def PCMPISTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
4762 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4763 "#PCMPISTRM128rr PSEUDO!",
4764 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4765 imm:$src3))]>, OpSize;
4766 def PCMPISTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
4767 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4768 "#PCMPISTRM128rm PSEUDO!",
4769 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4770 VR128:$src1, (load addr:$src2), imm:$src3))]>, OpSize;
4773 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4774 Predicates = [HasAVX] in {
4775 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4776 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4777 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4778 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4779 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4780 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4783 let Defs = [XMM0, EFLAGS] in {
4784 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4785 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4786 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4787 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4788 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4789 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4792 // Packed Compare Explicit Length Strings, Return Mask
4793 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4794 def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
4795 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4796 "#PCMPESTRM128rr PSEUDO!",
4798 (int_x86_sse42_pcmpestrm128
4799 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>, OpSize;
4801 def PCMPESTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
4802 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4803 "#PCMPESTRM128rm PSEUDO!",
4804 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4805 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
4809 let isAsmParserOnly = 1, Predicates = [HasAVX],
4810 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4811 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4812 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4813 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4814 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4815 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4816 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4819 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4820 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4821 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4822 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4823 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4824 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4825 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4828 // Packed Compare Implicit Length Strings, Return Index
4829 let Defs = [ECX, EFLAGS] in {
4830 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4831 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4832 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4833 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4834 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4835 (implicit EFLAGS)]>, OpSize;
4836 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4837 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4838 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4839 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4840 (implicit EFLAGS)]>, OpSize;
4844 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4845 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4847 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4849 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4851 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4853 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4855 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4859 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4860 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4861 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4862 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4863 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4864 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4866 // Packed Compare Explicit Length Strings, Return Index
4867 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4868 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4869 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4870 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4871 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4872 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4873 (implicit EFLAGS)]>, OpSize;
4874 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4875 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4876 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4878 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4879 (implicit EFLAGS)]>, OpSize;
4883 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4884 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4886 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
4888 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
4890 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
4892 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
4894 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
4898 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
4899 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
4900 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
4901 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
4902 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
4903 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
4905 //===----------------------------------------------------------------------===//
4906 // SSE4.2 - CRC Instructions
4907 //===----------------------------------------------------------------------===//
4909 // No CRC instructions have AVX equivalents
4911 // crc intrinsic instruction
4912 // This set of instructions are only rm, the only difference is the size
4914 let Constraints = "$src1 = $dst" in {
4915 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
4916 (ins GR32:$src1, i8mem:$src2),
4917 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4919 (int_x86_sse42_crc32_8 GR32:$src1,
4920 (load addr:$src2)))]>;
4921 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
4922 (ins GR32:$src1, GR8:$src2),
4923 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4925 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
4926 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4927 (ins GR32:$src1, i16mem:$src2),
4928 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4930 (int_x86_sse42_crc32_16 GR32:$src1,
4931 (load addr:$src2)))]>,
4933 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4934 (ins GR32:$src1, GR16:$src2),
4935 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4937 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
4939 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4940 (ins GR32:$src1, i32mem:$src2),
4941 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4943 (int_x86_sse42_crc32_32 GR32:$src1,
4944 (load addr:$src2)))]>;
4945 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4946 (ins GR32:$src1, GR32:$src2),
4947 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4949 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
4950 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
4951 (ins GR64:$src1, i8mem:$src2),
4952 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4954 (int_x86_sse42_crc64_8 GR64:$src1,
4955 (load addr:$src2)))]>,
4957 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
4958 (ins GR64:$src1, GR8:$src2),
4959 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4961 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
4963 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
4964 (ins GR64:$src1, i64mem:$src2),
4965 "crc32{q} \t{$src2, $src1|$src1, $src2}",
4967 (int_x86_sse42_crc64_64 GR64:$src1,
4968 (load addr:$src2)))]>,
4970 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
4971 (ins GR64:$src1, GR64:$src2),
4972 "crc32{q} \t{$src2, $src1|$src1, $src2}",
4974 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
4978 //===----------------------------------------------------------------------===//
4979 // AES-NI Instructions
4980 //===----------------------------------------------------------------------===//
4982 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
4983 Intrinsic IntId128, bit Is2Addr = 1> {
4984 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
4985 (ins VR128:$src1, VR128:$src2),
4987 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4988 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4989 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4991 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
4992 (ins VR128:$src1, i128mem:$src2),
4994 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4995 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4997 (IntId128 VR128:$src1,
4998 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5001 // Perform One Round of an AES Encryption/Decryption Flow
5002 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5003 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5004 int_x86_aesni_aesenc, 0>, VEX_4V;
5005 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5006 int_x86_aesni_aesenclast, 0>, VEX_4V;
5007 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5008 int_x86_aesni_aesdec, 0>, VEX_4V;
5009 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5010 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5013 let Constraints = "$src1 = $dst" in {
5014 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5015 int_x86_aesni_aesenc>;
5016 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5017 int_x86_aesni_aesenclast>;
5018 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5019 int_x86_aesni_aesdec>;
5020 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5021 int_x86_aesni_aesdeclast>;
5024 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5025 (AESENCrr VR128:$src1, VR128:$src2)>;
5026 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5027 (AESENCrm VR128:$src1, addr:$src2)>;
5028 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5029 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5030 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5031 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5032 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5033 (AESDECrr VR128:$src1, VR128:$src2)>;
5034 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5035 (AESDECrm VR128:$src1, addr:$src2)>;
5036 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5037 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5038 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5039 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5041 // Perform the AES InvMixColumn Transformation
5042 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5043 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5045 "vaesimc\t{$src1, $dst|$dst, $src1}",
5047 (int_x86_aesni_aesimc VR128:$src1))]>,
5049 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5050 (ins i128mem:$src1),
5051 "vaesimc\t{$src1, $dst|$dst, $src1}",
5053 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5056 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5058 "aesimc\t{$src1, $dst|$dst, $src1}",
5060 (int_x86_aesni_aesimc VR128:$src1))]>,
5062 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5063 (ins i128mem:$src1),
5064 "aesimc\t{$src1, $dst|$dst, $src1}",
5066 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5069 // AES Round Key Generation Assist
5070 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5071 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5072 (ins VR128:$src1, i8imm:$src2),
5073 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5075 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5077 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5078 (ins i128mem:$src1, i8imm:$src2),
5079 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5081 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5085 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5086 (ins VR128:$src1, i8imm:$src2),
5087 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5089 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5091 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5092 (ins i128mem:$src1, i8imm:$src2),
5093 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5095 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5099 //===----------------------------------------------------------------------===//
5101 //===----------------------------------------------------------------------===//
5103 let isAsmParserOnly = 1 in {
5105 // Load from memory and broadcast to all elements of the destination operand
5106 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5107 X86MemOperand x86memop> :
5108 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5109 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>, VEX;
5111 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem>;
5112 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem>;
5113 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem>;
5114 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem>;
5116 // Insert packed floating-point values
5117 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5118 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5119 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5121 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5122 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5123 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5126 // Extract packed floating-point values
5127 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5128 (ins VR256:$src1, i8imm:$src2),
5129 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5131 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5132 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5133 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5136 // Conditional SIMD Packed Loads and Stores
5137 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr> {
5138 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5139 (ins VR128:$src1, f128mem:$src2),
5140 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5142 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5143 (ins VR256:$src1, f256mem:$src2),
5144 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5146 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5147 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5148 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5150 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5151 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5152 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5156 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps">;
5157 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd">;
5159 // Permute Floating-Point Values
5160 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5161 RegisterClass RC, X86MemOperand x86memop> {
5162 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5163 (ins RC:$src1, RC:$src2),
5164 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5166 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5167 (ins RC:$src1, x86memop:$src2),
5168 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5170 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5171 (ins RC:$src1, i8imm:$src2),
5172 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5174 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5175 (ins x86memop:$src1, i8imm:$src2),
5176 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5180 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem>;
5181 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem>;
5182 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem>;
5183 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem>;
5185 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5186 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5187 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5189 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5190 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5191 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5194 // Zero All YMM registers
5195 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall", []>, VEX, VEX_L;
5197 // Zero Upper bits of YMM registers
5198 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper", []>, VEX;
5200 } // isAsmParserOnly