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:$dst)]>, XS, VEX_4V;
260 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
261 "movsd\t{$src, $dst|$dst, $src}",
262 [(store FR64:$src, addr:$dst)]>, XD, VEX_4V;
265 // Extract and store.
266 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$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:$src)))]>;
519 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
520 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
521 string asm, Domain d> {
522 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
523 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
524 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
525 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
528 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
529 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
531 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
533 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
534 (ins DstRC:$src1, x86memop:$src), asm, []>;
537 let isAsmParserOnly = 1 in {
538 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
539 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
540 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
541 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
542 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
543 "cvtsi2ss\t{$src, $src1, $dst|$dst, $src1, $src}">, XS,
545 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
546 "cvtsi2sd\t{$src, $src1, $dst|$dst, $src1, $src}">, XD,
550 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
551 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
552 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
553 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
554 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
555 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
556 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
557 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
559 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
560 // and/or XMM operand(s).
561 multiclass sse12_cvt_pint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
562 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
563 string asm, Domain d> {
564 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
565 [(set DstRC:$dst, (Int SrcRC:$src))], d>;
566 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
567 [(set DstRC:$dst, (Int (ld_frag addr:$src)))], d>;
570 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
571 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
573 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
574 [(set DstRC:$dst, (Int SrcRC:$src))]>;
575 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
576 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
579 multiclass sse12_cvt_pint_3addr<bits<8> opc, RegisterClass SrcRC,
580 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
581 PatFrag ld_frag, string asm, Domain d> {
582 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
583 asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))], d>;
584 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst),
585 (ins DstRC:$src1, x86memop:$src2), asm,
586 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))], d>;
589 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
590 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
591 PatFrag ld_frag, string asm> {
592 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
593 asm, [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
594 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
595 (ins DstRC:$src1, x86memop:$src2), asm,
596 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
599 let isAsmParserOnly = 1 in {
600 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
601 f32mem, load, "cvtss2si\t{$src, $dst|$dst, $src}">, XS,
603 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
604 f128mem, load, "cvtsd2si\t{$src, $dst|$dst, $src}">, XD,
607 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
608 f32mem, load, "cvtss2si\t{$src, $dst|$dst, $src}">, XS;
609 defm Int_CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
610 f128mem, load, "cvtsd2si\t{$src, $dst|$dst, $src}">, XD;
613 let Constraints = "$src1 = $dst" in {
614 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
615 int_x86_sse_cvtsi2ss, i32mem, loadi32,
616 "cvtsi2ss\t{$src2, $dst|$dst, $src2}">, XS;
617 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
618 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
619 "cvtsi2ss\t{$src2, $dst|$dst, $src2}">, XD;
622 // Instructions below don't have an AVX form.
623 defm Int_CVTPS2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtps2pi,
624 f64mem, load, "cvtps2pi\t{$src, $dst|$dst, $src}",
625 SSEPackedSingle>, TB;
626 defm Int_CVTPD2PI : sse12_cvt_pint<0x2D, VR128, VR64, int_x86_sse_cvtpd2pi,
627 f128mem, memop, "cvtpd2pi\t{$src, $dst|$dst, $src}",
628 SSEPackedDouble>, TB, OpSize;
629 defm Int_CVTTPS2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttps2pi,
630 f64mem, load, "cvttps2pi\t{$src, $dst|$dst, $src}",
631 SSEPackedSingle>, TB;
632 defm Int_CVTTPD2PI : sse12_cvt_pint<0x2C, VR128, VR64, int_x86_sse_cvttpd2pi,
633 f128mem, memop, "cvttpd2pi\t{$src, $dst|$dst, $src}",
634 SSEPackedDouble>, TB, OpSize;
635 defm Int_CVTPI2PD : sse12_cvt_pint<0x2A, VR64, VR128, int_x86_sse_cvtpi2pd,
636 i64mem, load, "cvtpi2pd\t{$src, $dst|$dst, $src}",
637 SSEPackedDouble>, TB, OpSize;
638 let Constraints = "$src1 = $dst" in {
639 defm Int_CVTPI2PS : sse12_cvt_pint_3addr<0x2A, VR64, VR128,
640 int_x86_sse_cvtpi2ps,
641 i64mem, load, "cvtpi2ps\t{$src2, $dst|$dst, $src2}",
642 SSEPackedSingle>, TB;
647 // Aliases for intrinsics
648 let isAsmParserOnly = 1, Pattern = []<dag> in {
649 defm Int_VCVTTSS2SI : sse12_cvt_sint_3addr<0x2C, VR128, GR32,
650 int_x86_sse_cvttss2si, f32mem, load,
651 "cvttss2si\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS;
652 defm Int_VCVTTSD2SI : sse12_cvt_sint_3addr<0x2C, VR128, GR32,
653 int_x86_sse2_cvttsd2si, f128mem, load,
654 "cvttss2si\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD;
656 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
657 f32mem, load, "cvttss2si\t{$src, $dst|$dst, $src}">,
659 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
660 f128mem, load, "cvttss2si\t{$src, $dst|$dst, $src}">,
663 let isAsmParserOnly = 1, Pattern = []<dag> in {
664 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
665 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
666 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load,
667 "cvtdq2ps\t{$src, $dst|$dst, $src}",
668 SSEPackedSingle>, TB, VEX;
670 let Pattern = []<dag> in {
671 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
672 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
673 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, f128mem, load /*dummy*/,
674 "cvtdq2ps\t{$src, $dst|$dst, $src}",
675 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
680 // Convert scalar double to scalar single
681 let isAsmParserOnly = 1 in {
682 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
683 (ins FR64:$src1, FR64:$src2),
684 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
686 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
687 (ins FR64:$src1, f64mem:$src2),
688 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
689 []>, XD, Requires<[HasAVX, HasSSE2, OptForSize]>, VEX_4V;
691 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
692 "cvtsd2ss\t{$src, $dst|$dst, $src}",
693 [(set FR32:$dst, (fround FR64:$src))]>;
694 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
695 "cvtsd2ss\t{$src, $dst|$dst, $src}",
696 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
697 Requires<[HasSSE2, OptForSize]>;
699 let isAsmParserOnly = 1 in
700 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
701 int_x86_sse2_cvtsd2ss, f64mem, load,
702 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}">,
704 let Constraints = "$src1 = $dst" in
705 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
706 int_x86_sse2_cvtsd2ss, f64mem, load,
707 "cvtsd2ss\t{$src2, $dst|$dst, $src2}">, XS;
709 // Convert scalar single to scalar double
710 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
711 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
712 (ins FR32:$src1, FR32:$src2),
713 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
714 []>, XS, Requires<[HasAVX, HasSSE2]>, VEX_4V;
715 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
716 (ins FR32:$src1, f32mem:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 []>, XS, VEX_4V, Requires<[HasAVX, HasSSE2, OptForSize]>;
720 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
721 "cvtss2sd\t{$src, $dst|$dst, $src}",
722 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
724 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
725 "cvtss2sd\t{$src, $dst|$dst, $src}",
726 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
727 Requires<[HasSSE2, OptForSize]>;
729 let isAsmParserOnly = 1 in {
730 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
731 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
732 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
733 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
734 VR128:$src2))]>, XS, VEX_4V,
735 Requires<[HasAVX, HasSSE2]>;
736 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
737 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
738 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
739 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
740 (load addr:$src2)))]>, XS, VEX_4V,
741 Requires<[HasAVX, HasSSE2]>;
743 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
744 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
745 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
746 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
747 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
750 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
751 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
752 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
753 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
754 (load addr:$src2)))]>, XS,
758 def : Pat<(extloadf32 addr:$src),
759 (CVTSS2SDrr (MOVSSrm addr:$src))>,
760 Requires<[HasSSE2, OptForSpeed]>;
762 // Convert doubleword to packed single/double fp
763 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
764 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
765 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
766 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
767 TB, VEX, Requires<[HasAVX, HasSSE2]>;
768 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
769 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
770 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
771 (bitconvert (memopv2i64 addr:$src))))]>,
772 TB, VEX, Requires<[HasAVX, HasSSE2]>;
774 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
775 "cvtdq2ps\t{$src, $dst|$dst, $src}",
776 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
777 TB, Requires<[HasSSE2]>;
778 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
779 "cvtdq2ps\t{$src, $dst|$dst, $src}",
780 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
781 (bitconvert (memopv2i64 addr:$src))))]>,
782 TB, Requires<[HasSSE2]>;
784 // FIXME: why the non-intrinsic version is described as SSE3?
785 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
786 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
787 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
788 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
789 XS, VEX, Requires<[HasAVX, HasSSE2]>;
790 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
791 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
792 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
793 (bitconvert (memopv2i64 addr:$src))))]>,
794 XS, VEX, Requires<[HasAVX, HasSSE2]>;
796 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
797 "cvtdq2pd\t{$src, $dst|$dst, $src}",
798 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
799 XS, Requires<[HasSSE2]>;
800 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
801 "cvtdq2pd\t{$src, $dst|$dst, $src}",
802 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
803 (bitconvert (memopv2i64 addr:$src))))]>,
804 XS, Requires<[HasSSE2]>;
806 // Convert packed single/double fp to doubleword
807 let isAsmParserOnly = 1 in {
808 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
809 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
810 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
811 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
813 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
814 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
815 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
816 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
818 let isAsmParserOnly = 1 in {
819 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
820 "cvtps2dq\t{$src, $dst|$dst, $src}",
821 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
823 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
825 "cvtps2dq\t{$src, $dst|$dst, $src}",
826 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
827 (memop addr:$src)))]>, VEX;
829 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
830 "cvtps2dq\t{$src, $dst|$dst, $src}",
831 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
832 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
833 "cvtps2dq\t{$src, $dst|$dst, $src}",
834 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
835 (memop addr:$src)))]>;
837 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
838 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
839 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
840 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
841 XD, VEX, Requires<[HasAVX, HasSSE2]>;
842 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
843 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
844 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
845 (memop addr:$src)))]>,
846 XD, VEX, Requires<[HasAVX, HasSSE2]>;
848 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
849 "cvtpd2dq\t{$src, $dst|$dst, $src}",
850 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
851 XD, Requires<[HasSSE2]>;
852 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
853 "cvtpd2dq\t{$src, $dst|$dst, $src}",
854 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
855 (memop addr:$src)))]>,
856 XD, Requires<[HasSSE2]>;
859 // Convert with truncation packed single/double fp to doubleword
860 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
861 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
862 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
863 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
864 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
866 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
867 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
868 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
869 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
872 let isAsmParserOnly = 1 in {
873 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
874 "vcvttps2dq\t{$src, $dst|$dst, $src}",
876 (int_x86_sse2_cvttps2dq VR128:$src))]>,
877 XS, VEX, Requires<[HasAVX, HasSSE2]>;
878 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
879 "vcvttps2dq\t{$src, $dst|$dst, $src}",
880 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
881 (memop addr:$src)))]>,
882 XS, VEX, Requires<[HasAVX, HasSSE2]>;
884 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
885 "cvttps2dq\t{$src, $dst|$dst, $src}",
887 (int_x86_sse2_cvttps2dq VR128:$src))]>,
888 XS, Requires<[HasSSE2]>;
889 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
890 "cvttps2dq\t{$src, $dst|$dst, $src}",
891 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
892 (memop addr:$src)))]>,
893 XS, Requires<[HasSSE2]>;
895 let isAsmParserOnly = 1 in {
896 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
898 "cvttpd2dq\t{$src, $dst|$dst, $src}",
899 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
901 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
903 "cvttpd2dq\t{$src, $dst|$dst, $src}",
904 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
905 (memop addr:$src)))]>, VEX;
907 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
908 "cvttpd2dq\t{$src, $dst|$dst, $src}",
909 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
910 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
911 "cvttpd2dq\t{$src, $dst|$dst, $src}",
912 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
913 (memop addr:$src)))]>;
915 // Convert packed single to packed double
916 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
917 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
918 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX,
920 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
921 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX,
924 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
925 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
926 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
927 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
929 let isAsmParserOnly = 1 in {
930 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
931 "cvtps2pd\t{$src, $dst|$dst, $src}",
932 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
933 VEX, Requires<[HasAVX, HasSSE2]>;
934 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
935 "cvtps2pd\t{$src, $dst|$dst, $src}",
936 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
937 (load addr:$src)))]>,
938 VEX, Requires<[HasAVX, HasSSE2]>;
940 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
941 "cvtps2pd\t{$src, $dst|$dst, $src}",
942 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
943 TB, Requires<[HasSSE2]>;
944 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
945 "cvtps2pd\t{$src, $dst|$dst, $src}",
946 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
947 (load addr:$src)))]>,
948 TB, Requires<[HasSSE2]>;
950 // Convert packed double to packed single
951 let isAsmParserOnly = 1 in {
952 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
953 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
954 // FIXME: the memory form of this instruction should described using
955 // use extra asm syntax
957 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
958 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
959 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
960 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
963 let isAsmParserOnly = 1 in {
964 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
965 "cvtpd2ps\t{$src, $dst|$dst, $src}",
966 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
967 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
969 "cvtpd2ps\t{$src, $dst|$dst, $src}",
970 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
971 (memop addr:$src)))]>;
973 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
974 "cvtpd2ps\t{$src, $dst|$dst, $src}",
975 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
976 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
977 "cvtpd2ps\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
979 (memop addr:$src)))]>;
981 //===----------------------------------------------------------------------===//
982 // SSE 1 & 2 - Compare Instructions
983 //===----------------------------------------------------------------------===//
985 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
986 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
987 string asm, string asm_alt> {
988 def rr : SIi8<0xC2, MRMSrcReg,
989 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
992 def rm : SIi8<0xC2, MRMSrcMem,
993 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
995 // Accept explicit immediate argument form instead of comparison code.
996 let isAsmParserOnly = 1 in {
997 def rr_alt : SIi8<0xC2, MRMSrcReg,
998 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1001 def rm_alt : SIi8<0xC2, MRMSrcMem,
1002 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1007 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1008 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1009 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1010 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1012 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1013 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1014 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1018 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1019 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1020 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1021 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1022 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1023 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1024 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1027 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1028 Intrinsic Int, string asm> {
1029 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1030 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1031 [(set VR128:$dst, (Int VR128:$src1,
1032 VR128:$src, imm:$cc))]>;
1033 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1034 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1035 [(set VR128:$dst, (Int VR128:$src1,
1036 (load addr:$src), imm:$cc))]>;
1039 // Aliases to match intrinsics which expect XMM operand(s).
1040 let isAsmParserOnly = 1 in {
1041 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1042 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1044 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1045 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1048 let Constraints = "$src1 = $dst" in {
1049 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1050 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1051 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1052 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1056 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1057 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1058 ValueType vt, X86MemOperand x86memop,
1059 PatFrag ld_frag, string OpcodeStr, Domain d> {
1060 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1061 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1062 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1063 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1064 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1065 [(set EFLAGS, (OpNode (vt RC:$src1),
1066 (ld_frag addr:$src2)))], d>;
1069 let Defs = [EFLAGS] in {
1070 let isAsmParserOnly = 1 in {
1071 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1072 "ucomiss", SSEPackedSingle>, VEX;
1073 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1074 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1075 let Pattern = []<dag> in {
1076 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1077 "comiss", SSEPackedSingle>, VEX;
1078 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1079 "comisd", SSEPackedDouble>, OpSize, VEX;
1082 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1083 load, "ucomiss", SSEPackedSingle>, VEX;
1084 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1085 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1087 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1088 load, "comiss", SSEPackedSingle>, VEX;
1089 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1090 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1092 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1093 "ucomiss", SSEPackedSingle>, TB;
1094 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1095 "ucomisd", SSEPackedDouble>, TB, OpSize;
1097 let Pattern = []<dag> in {
1098 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1099 "comiss", SSEPackedSingle>, TB;
1100 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1101 "comisd", SSEPackedDouble>, TB, OpSize;
1104 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1105 load, "ucomiss", SSEPackedSingle>, TB;
1106 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1107 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1109 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1110 "comiss", SSEPackedSingle>, TB;
1111 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1112 "comisd", SSEPackedDouble>, TB, OpSize;
1113 } // Defs = [EFLAGS]
1115 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1116 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1117 Intrinsic Int, string asm, string asm_alt,
1119 def rri : PIi8<0xC2, MRMSrcReg,
1120 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1121 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1122 def rmi : PIi8<0xC2, MRMSrcMem,
1123 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1124 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1125 // Accept explicit immediate argument form instead of comparison code.
1126 let isAsmParserOnly = 1 in {
1127 def rri_alt : PIi8<0xC2, MRMSrcReg,
1128 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1130 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1131 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1136 let isAsmParserOnly = 1 in {
1137 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1138 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1139 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1140 SSEPackedSingle>, VEX_4V;
1141 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1142 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1143 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1144 SSEPackedDouble>, OpSize, VEX_4V;
1146 let Constraints = "$src1 = $dst" in {
1147 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1148 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1149 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1150 SSEPackedSingle>, TB;
1151 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1152 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1153 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1154 SSEPackedDouble>, TB, OpSize;
1157 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1158 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1159 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1160 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1161 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1162 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1163 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1164 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1166 //===----------------------------------------------------------------------===//
1167 // SSE 1 & 2 - Shuffle Instructions
1168 //===----------------------------------------------------------------------===//
1170 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1171 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1172 ValueType vt, string asm, PatFrag mem_frag,
1173 Domain d, bit IsConvertibleToThreeAddress = 0> {
1174 def rmi : PIi8<0xC6, MRMSrcMem, (outs VR128:$dst),
1175 (ins VR128:$src1, f128mem:$src2, i8imm:$src3), asm,
1176 [(set VR128:$dst, (vt (shufp:$src3
1177 VR128:$src1, (mem_frag addr:$src2))))], d>;
1178 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1179 def rri : PIi8<0xC6, MRMSrcReg, (outs VR128:$dst),
1180 (ins VR128:$src1, VR128:$src2, i8imm:$src3), asm,
1182 (vt (shufp:$src3 VR128:$src1, VR128:$src2)))], d>;
1185 let isAsmParserOnly = 1 in {
1186 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1187 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1188 memopv4f32, SSEPackedSingle>, VEX_4V;
1189 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1190 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1191 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1194 let Constraints = "$src1 = $dst" in {
1195 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1196 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1197 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1199 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1200 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1201 memopv2f64, SSEPackedDouble>, TB, OpSize;
1204 //===----------------------------------------------------------------------===//
1205 // SSE 1 & 2 - Unpack Instructions
1206 //===----------------------------------------------------------------------===//
1208 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1209 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1210 PatFrag mem_frag, RegisterClass RC,
1211 X86MemOperand x86memop, string asm,
1213 def rr : PI<opc, MRMSrcReg,
1214 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1216 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1217 def rm : PI<opc, MRMSrcMem,
1218 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1220 (vt (OpNode RC:$src1,
1221 (mem_frag addr:$src2))))], d>;
1224 let AddedComplexity = 10 in {
1225 let isAsmParserOnly = 1 in {
1226 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1227 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1228 SSEPackedSingle>, VEX_4V;
1229 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1230 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 SSEPackedDouble>, OpSize, VEX_4V;
1232 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1233 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1234 SSEPackedSingle>, VEX_4V;
1235 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1236 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1237 SSEPackedDouble>, OpSize, VEX_4V;
1239 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1240 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1241 SSEPackedSingle>, VEX_4V;
1242 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1243 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1244 SSEPackedDouble>, OpSize, VEX_4V;
1245 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1246 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1247 SSEPackedSingle>, VEX_4V;
1248 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1249 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1250 SSEPackedDouble>, OpSize, VEX_4V;
1253 let Constraints = "$src1 = $dst" in {
1254 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1255 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1256 SSEPackedSingle>, TB;
1257 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1258 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1259 SSEPackedDouble>, TB, OpSize;
1260 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1261 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1262 SSEPackedSingle>, TB;
1263 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1264 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1265 SSEPackedDouble>, TB, OpSize;
1266 } // Constraints = "$src1 = $dst"
1267 } // AddedComplexity
1269 //===----------------------------------------------------------------------===//
1270 // SSE 1 & 2 - Extract Floating-Point Sign mask
1271 //===----------------------------------------------------------------------===//
1273 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1274 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1276 def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1277 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1278 [(set GR32:$dst, (Int RC:$src))], d>;
1282 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1283 SSEPackedSingle>, TB;
1284 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1285 SSEPackedDouble>, TB, OpSize;
1287 let isAsmParserOnly = 1 in {
1288 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1289 "movmskps", SSEPackedSingle>, VEX;
1290 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1291 "movmskpd", SSEPackedDouble>, OpSize,
1293 // FIXME: merge with multiclass above when the intrinsics come.
1294 def VMOVMSKPSYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1295 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1296 def VMOVMSKPDYrr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
1297 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1301 //===----------------------------------------------------------------------===//
1302 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1303 //===----------------------------------------------------------------------===//
1305 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1306 // names that start with 'Fs'.
1308 // Alias instructions that map fld0 to pxor for sse.
1309 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1310 canFoldAsLoad = 1 in {
1311 // FIXME: Set encoding to pseudo!
1312 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1313 [(set FR32:$dst, fp32imm0)]>,
1314 Requires<[HasSSE1]>, TB, OpSize;
1315 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1316 [(set FR64:$dst, fpimm0)]>,
1317 Requires<[HasSSE2]>, TB, OpSize;
1320 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1321 // bits are disregarded.
1322 let neverHasSideEffects = 1 in {
1323 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1324 "movaps\t{$src, $dst|$dst, $src}", []>;
1325 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1326 "movapd\t{$src, $dst|$dst, $src}", []>;
1329 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1330 // bits are disregarded.
1331 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1332 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1333 "movaps\t{$src, $dst|$dst, $src}",
1334 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1335 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1336 "movapd\t{$src, $dst|$dst, $src}",
1337 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1340 //===----------------------------------------------------------------------===//
1341 // SSE 1 & 2 - Logical Instructions
1342 //===----------------------------------------------------------------------===//
1344 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1346 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1348 let isAsmParserOnly = 1 in {
1349 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1350 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1352 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1353 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1356 let Constraints = "$src1 = $dst" in {
1357 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1358 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1360 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1361 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1365 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1366 let mayLoad = 0 in {
1367 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1368 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1369 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1372 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1373 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1375 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1377 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1378 SDNode OpNode, int HasPat = 0,
1379 list<list<dag>> Pattern = []> {
1380 let isAsmParserOnly = 1 in {
1381 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1382 !strconcat(OpcodeStr, "ps"), f128mem,
1383 !if(HasPat, Pattern[0], // rr
1384 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1386 !if(HasPat, Pattern[2], // rm
1387 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1388 (memopv2i64 addr:$src2)))]), 0>,
1391 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1392 !strconcat(OpcodeStr, "pd"), f128mem,
1393 !if(HasPat, Pattern[1], // rr
1394 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1397 !if(HasPat, Pattern[3], // rm
1398 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1399 (memopv2i64 addr:$src2)))]), 0>,
1402 let Constraints = "$src1 = $dst" in {
1403 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1404 !strconcat(OpcodeStr, "ps"), f128mem,
1405 !if(HasPat, Pattern[0], // rr
1406 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1408 !if(HasPat, Pattern[2], // rm
1409 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1410 (memopv2i64 addr:$src2)))])>, TB;
1412 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1413 !strconcat(OpcodeStr, "pd"), f128mem,
1414 !if(HasPat, Pattern[1], // rr
1415 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1418 !if(HasPat, Pattern[3], // rm
1419 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1420 (memopv2i64 addr:$src2)))])>,
1425 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1426 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1427 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1428 let isCommutable = 0 in
1429 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1431 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1432 (bc_v2i64 (v4i32 immAllOnesV))),
1435 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1436 (bc_v2i64 (v2f64 VR128:$src2))))],
1438 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1439 (bc_v2i64 (v4i32 immAllOnesV))),
1440 (memopv2i64 addr:$src2))))],
1442 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1443 (memopv2i64 addr:$src2)))]]>;
1445 //===----------------------------------------------------------------------===//
1446 // SSE 1 & 2 - Arithmetic Instructions
1447 //===----------------------------------------------------------------------===//
1449 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1452 /// In addition, we also have a special variant of the scalar form here to
1453 /// represent the associated intrinsic operation. This form is unlike the
1454 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1455 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1457 /// These three forms can each be reg+reg or reg+mem.
1459 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1461 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1462 OpNode, FR32, f32mem, Is2Addr>, XS;
1463 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1464 OpNode, FR64, f64mem, Is2Addr>, XD;
1467 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1469 let mayLoad = 0 in {
1470 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1471 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1472 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1473 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1477 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1479 let mayLoad = 0 in {
1480 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1481 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1482 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1483 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1487 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1489 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1490 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1491 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1492 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1495 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1497 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1498 !strconcat(OpcodeStr, "ps"), "", "_ps", f128mem, memopv4f32,
1499 SSEPackedSingle, Is2Addr>, TB;
1501 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1502 !strconcat(OpcodeStr, "pd"), "2", "_pd", f128mem, memopv2f64,
1503 SSEPackedDouble, Is2Addr>, TB, OpSize;
1506 // Binary Arithmetic instructions
1507 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1508 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1509 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1510 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1511 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1512 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1513 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1515 let isCommutable = 0 in {
1516 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1517 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1518 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1519 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1520 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1521 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1522 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1523 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1524 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>, VEX_4V;
1525 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1526 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1527 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1531 let Constraints = "$src1 = $dst" in {
1532 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1533 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1534 basic_sse12_fp_binop_s_int<0x58, "add">;
1535 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1536 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1537 basic_sse12_fp_binop_s_int<0x59, "mul">;
1539 let isCommutable = 0 in {
1540 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1541 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1542 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1543 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1544 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1545 basic_sse12_fp_binop_s_int<0x5E, "div">;
1546 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1547 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1548 basic_sse12_fp_binop_s_int<0x5F, "max">,
1549 basic_sse12_fp_binop_p_int<0x5F, "max">;
1550 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1551 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1552 basic_sse12_fp_binop_s_int<0x5D, "min">,
1553 basic_sse12_fp_binop_p_int<0x5D, "min">;
1558 /// In addition, we also have a special variant of the scalar form here to
1559 /// represent the associated intrinsic operation. This form is unlike the
1560 /// plain scalar form, in that it takes an entire vector (instead of a
1561 /// scalar) and leaves the top elements undefined.
1563 /// And, we have a special variant form for a full-vector intrinsic form.
1565 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1566 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1567 SDNode OpNode, Intrinsic F32Int> {
1568 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1569 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1570 [(set FR32:$dst, (OpNode FR32:$src))]>;
1571 // For scalar unary operations, fold a load into the operation
1572 // only in OptForSize mode. It eliminates an instruction, but it also
1573 // eliminates a whole-register clobber (the load), so it introduces a
1574 // partial register update condition.
1575 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1576 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1577 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1578 Requires<[HasSSE1, OptForSize]>;
1579 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1580 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1581 [(set VR128:$dst, (F32Int VR128:$src))]>;
1582 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1583 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1584 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1587 /// sse1_fp_unop_p - SSE1 unops in scalar form.
1588 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr,
1589 SDNode OpNode, Intrinsic V4F32Int> {
1590 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1591 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1592 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1593 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1594 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1595 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1596 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1597 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1598 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1599 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1600 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1601 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1604 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1605 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1606 SDNode OpNode, Intrinsic F32Int> {
1607 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1608 !strconcat(!strconcat("v", OpcodeStr),
1609 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1610 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1611 !strconcat(!strconcat("v", OpcodeStr),
1612 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1613 []>, XS, Requires<[HasAVX, HasSSE1, OptForSize]>;
1614 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst),
1615 (ins VR128:$src1, VR128:$src2),
1616 !strconcat(!strconcat("v", OpcodeStr),
1617 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1618 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
1619 (ins VR128:$src1, ssmem:$src2),
1620 !strconcat(!strconcat("v", OpcodeStr),
1621 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1624 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1625 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1626 SDNode OpNode, Intrinsic F64Int> {
1627 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1628 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1629 [(set FR64:$dst, (OpNode FR64:$src))]>;
1630 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1631 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1632 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1633 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1634 Requires<[HasSSE2, OptForSize]>;
1635 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1636 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1637 [(set VR128:$dst, (F64Int VR128:$src))]>;
1638 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1639 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1640 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1643 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1644 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1645 SDNode OpNode, Intrinsic V2F64Int> {
1646 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1647 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1648 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1649 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1650 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1651 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1652 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1653 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1654 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1655 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1656 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1657 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1660 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1661 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1662 SDNode OpNode, Intrinsic F64Int> {
1663 def SDr : VSDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1664 !strconcat(OpcodeStr,
1665 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1666 def SDm : VSDI<opc, MRMSrcMem, (outs FR64:$dst),
1667 (ins FR64:$src1, f64mem:$src2),
1668 !strconcat(OpcodeStr,
1669 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1670 def SDr_Int : VSDI<opc, MRMSrcReg, (outs VR128:$dst),
1671 (ins VR128:$src1, VR128:$src2),
1672 !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1674 def SDm_Int : VSDI<opc, MRMSrcMem, (outs VR128:$dst),
1675 (ins VR128:$src1, sdmem:$src2),
1676 !strconcat(OpcodeStr, "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1680 let isAsmParserOnly = 1 in {
1682 let Predicates = [HasAVX, HasSSE2] in {
1683 defm VSQRT : sse2_fp_unop_s_avx<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1686 defm VSQRT : sse2_fp_unop_p<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_pd>, VEX;
1689 let Predicates = [HasAVX, HasSSE1] in {
1690 defm VSQRT : sse1_fp_unop_s_avx<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1692 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ps>, VEX;
1693 // Reciprocal approximations. Note that these typically require refinement
1694 // in order to obtain suitable precision.
1695 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "rsqrt", X86frsqrt,
1696 int_x86_sse_rsqrt_ss>, VEX_4V;
1697 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt, int_x86_sse_rsqrt_ps>,
1699 defm VRCP : sse1_fp_unop_s_avx<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1701 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ps>,
1707 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1708 sse1_fp_unop_p<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ps>,
1709 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1710 sse2_fp_unop_p<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_pd>;
1712 // Reciprocal approximations. Note that these typically require refinement
1713 // in order to obtain suitable precision.
1714 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1715 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ps>;
1716 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1717 sse1_fp_unop_p<0x53, "rcp", X86frcp, int_x86_sse_rcp_ps>;
1719 // There is no f64 version of the reciprocal approximation instructions.
1721 //===----------------------------------------------------------------------===//
1722 // SSE 1 & 2 - Non-temporal stores
1723 //===----------------------------------------------------------------------===//
1725 let isAsmParserOnly = 1 in {
1726 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1727 (ins i128mem:$dst, VR128:$src),
1728 "movntps\t{$src, $dst|$dst, $src}",
1729 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1730 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1731 (ins i128mem:$dst, VR128:$src),
1732 "movntpd\t{$src, $dst|$dst, $src}",
1733 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1735 let ExeDomain = SSEPackedInt in
1736 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1737 (ins f128mem:$dst, VR128:$src),
1738 "movntdq\t{$src, $dst|$dst, $src}",
1739 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1741 let AddedComplexity = 400 in { // Prefer non-temporal versions
1742 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1743 (ins f128mem:$dst, VR128:$src),
1744 "movntps\t{$src, $dst|$dst, $src}",
1745 [(alignednontemporalstore (v4f32 VR128:$src),
1747 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1748 (ins f128mem:$dst, VR128:$src),
1749 "movntpd\t{$src, $dst|$dst, $src}",
1750 [(alignednontemporalstore (v2f64 VR128:$src),
1752 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1753 (ins f128mem:$dst, VR128:$src),
1754 "movntdq\t{$src, $dst|$dst, $src}",
1755 [(alignednontemporalstore (v2f64 VR128:$src),
1757 let ExeDomain = SSEPackedInt in
1758 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1759 (ins f128mem:$dst, VR128:$src),
1760 "movntdq\t{$src, $dst|$dst, $src}",
1761 [(alignednontemporalstore (v4f32 VR128:$src),
1764 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1765 (ins f256mem:$dst, VR256:$src),
1766 "movntps\t{$src, $dst|$dst, $src}",
1767 [(alignednontemporalstore (v8f32 VR256:$src),
1769 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1770 (ins f256mem:$dst, VR256:$src),
1771 "movntpd\t{$src, $dst|$dst, $src}",
1772 [(alignednontemporalstore (v4f64 VR256:$src),
1774 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1775 (ins f256mem:$dst, VR256:$src),
1776 "movntdq\t{$src, $dst|$dst, $src}",
1777 [(alignednontemporalstore (v4f64 VR256:$src),
1779 let ExeDomain = SSEPackedInt in
1780 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1781 (ins f256mem:$dst, VR256:$src),
1782 "movntdq\t{$src, $dst|$dst, $src}",
1783 [(alignednontemporalstore (v8f32 VR256:$src),
1788 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1789 "movntps\t{$src, $dst|$dst, $src}",
1790 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
1791 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1792 "movntpd\t{$src, $dst|$dst, $src}",
1793 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
1795 let ExeDomain = SSEPackedInt in
1796 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1797 "movntdq\t{$src, $dst|$dst, $src}",
1798 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
1800 let AddedComplexity = 400 in { // Prefer non-temporal versions
1801 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1802 "movntps\t{$src, $dst|$dst, $src}",
1803 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1804 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1805 "movntpd\t{$src, $dst|$dst, $src}",
1806 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1808 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1809 "movntdq\t{$src, $dst|$dst, $src}",
1810 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1812 let ExeDomain = SSEPackedInt in
1813 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1814 "movntdq\t{$src, $dst|$dst, $src}",
1815 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1817 // There is no AVX form for instructions below this point
1818 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1819 "movnti\t{$src, $dst|$dst, $src}",
1820 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
1821 TB, Requires<[HasSSE2]>;
1823 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
1824 "movnti\t{$src, $dst|$dst, $src}",
1825 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
1826 TB, Requires<[HasSSE2]>;
1829 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1830 "movnti\t{$src, $dst|$dst, $src}",
1831 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
1832 TB, Requires<[HasSSE2]>;
1834 //===----------------------------------------------------------------------===//
1835 // SSE 1 & 2 - Misc Instructions (No AVX form)
1836 //===----------------------------------------------------------------------===//
1838 // Prefetch intrinsic.
1839 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
1840 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
1841 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
1842 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
1843 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
1844 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
1845 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
1846 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
1848 // Load, store, and memory fence
1849 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
1850 TB, Requires<[HasSSE1]>;
1852 // Alias instructions that map zero vector to pxor / xorp* for sse.
1853 // We set canFoldAsLoad because this can be converted to a constant-pool
1854 // load of an all-zeros value if folding it would be beneficial.
1855 // FIXME: Change encoding to pseudo!
1856 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
1857 isCodeGenOnly = 1 in {
1858 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
1859 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
1860 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
1861 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
1862 let ExeDomain = SSEPackedInt in
1863 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
1864 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
1867 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
1868 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
1869 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
1871 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
1872 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1874 //===----------------------------------------------------------------------===//
1875 // SSE 1 & 2 - Load/Store XCSR register
1876 //===----------------------------------------------------------------------===//
1878 let isAsmParserOnly = 1 in {
1879 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
1880 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
1881 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
1882 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
1885 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
1886 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
1887 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
1888 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
1890 //===---------------------------------------------------------------------===//
1891 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
1892 //===---------------------------------------------------------------------===//
1893 let ExeDomain = SSEPackedInt in { // SSE integer instructions
1895 let isAsmParserOnly = 1 in {
1896 let neverHasSideEffects = 1 in
1897 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1898 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
1899 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1900 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
1902 let canFoldAsLoad = 1, mayLoad = 1 in {
1903 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1904 "movdqa\t{$src, $dst|$dst, $src}",
1905 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>,
1907 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1908 "vmovdqu\t{$src, $dst|$dst, $src}",
1909 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
1910 XS, VEX, Requires<[HasAVX, HasSSE2]>;
1913 let mayStore = 1 in {
1914 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
1915 (ins i128mem:$dst, VR128:$src),
1916 "movdqa\t{$src, $dst|$dst, $src}",
1917 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>, VEX;
1918 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1919 "vmovdqu\t{$src, $dst|$dst, $src}",
1920 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
1921 XS, VEX, Requires<[HasAVX, HasSSE2]>;
1925 let neverHasSideEffects = 1 in
1926 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1927 "movdqa\t{$src, $dst|$dst, $src}", []>;
1929 let canFoldAsLoad = 1, mayLoad = 1 in {
1930 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1931 "movdqa\t{$src, $dst|$dst, $src}",
1932 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
1933 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1934 "movdqu\t{$src, $dst|$dst, $src}",
1935 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
1936 XS, Requires<[HasSSE2]>;
1939 let mayStore = 1 in {
1940 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1941 "movdqa\t{$src, $dst|$dst, $src}",
1942 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
1943 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1944 "movdqu\t{$src, $dst|$dst, $src}",
1945 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
1946 XS, Requires<[HasSSE2]>;
1949 // Intrinsic forms of MOVDQU load and store
1950 let isAsmParserOnly = 1 in {
1951 let canFoldAsLoad = 1 in
1952 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1953 "vmovdqu\t{$src, $dst|$dst, $src}",
1954 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
1955 XS, VEX, Requires<[HasAVX, HasSSE2]>;
1956 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1957 "vmovdqu\t{$src, $dst|$dst, $src}",
1958 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
1959 XS, VEX, Requires<[HasAVX, HasSSE2]>;
1962 let canFoldAsLoad = 1 in
1963 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1964 "movdqu\t{$src, $dst|$dst, $src}",
1965 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
1966 XS, Requires<[HasSSE2]>;
1967 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1968 "movdqu\t{$src, $dst|$dst, $src}",
1969 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
1970 XS, Requires<[HasSSE2]>;
1972 } // ExeDomain = SSEPackedInt
1974 //===---------------------------------------------------------------------===//
1975 // SSE2 - Packed Integer Arithmetic Instructions
1976 //===---------------------------------------------------------------------===//
1978 let ExeDomain = SSEPackedInt in { // SSE integer instructions
1980 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
1981 bit IsCommutable = 0, bit Is2Addr = 1> {
1982 let isCommutable = IsCommutable in
1983 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
1984 (ins VR128:$src1, VR128:$src2),
1986 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
1987 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1988 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
1989 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
1990 (ins VR128:$src1, i128mem:$src2),
1992 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
1993 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1994 [(set VR128:$dst, (IntId VR128:$src1,
1995 (bitconvert (memopv2i64 addr:$src2))))]>;
1998 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
1999 string OpcodeStr, Intrinsic IntId,
2000 Intrinsic IntId2, bit Is2Addr = 1> {
2001 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2002 (ins VR128:$src1, VR128:$src2),
2004 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2005 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2006 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2007 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2008 (ins VR128:$src1, i128mem:$src2),
2010 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2011 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2012 [(set VR128:$dst, (IntId VR128:$src1,
2013 (bitconvert (memopv2i64 addr:$src2))))]>;
2014 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2015 (ins VR128:$src1, i32i8imm:$src2),
2017 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2018 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2019 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2022 /// PDI_binop_rm - Simple SSE2 binary operator.
2023 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2024 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2025 let isCommutable = IsCommutable in
2026 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2027 (ins VR128:$src1, VR128:$src2),
2029 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2030 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2031 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2032 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2033 (ins VR128:$src1, i128mem:$src2),
2035 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2036 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2037 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2038 (bitconvert (memopv2i64 addr:$src2)))))]>;
2041 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2043 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2044 /// to collapse (bitconvert VT to VT) into its operand.
2046 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2047 bit IsCommutable = 0, bit Is2Addr = 1> {
2048 let isCommutable = IsCommutable in
2049 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2050 (ins VR128:$src1, VR128:$src2),
2052 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2053 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2054 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2055 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2056 (ins VR128:$src1, i128mem:$src2),
2058 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2059 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2060 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2063 } // ExeDomain = SSEPackedInt
2065 // 128-bit Integer Arithmetic
2067 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2068 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2069 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2070 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2071 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2072 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2073 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2074 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2075 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2076 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2079 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2081 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2083 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2085 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2087 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2089 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2091 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2093 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2095 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2097 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2099 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2101 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2103 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2105 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2107 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2109 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2111 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2113 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2115 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2119 let Constraints = "$src1 = $dst" in {
2120 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2121 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2122 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2123 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2124 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2125 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2126 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2127 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2128 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2131 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2132 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2133 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2134 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2135 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2136 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2137 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2138 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2139 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2140 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2141 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2142 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2143 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2144 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2145 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2146 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2147 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2148 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2149 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2151 } // Constraints = "$src1 = $dst"
2153 //===---------------------------------------------------------------------===//
2154 // SSE2 - Packed Integer Logical Instructions
2155 //===---------------------------------------------------------------------===//
2157 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2158 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2159 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2161 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2162 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2164 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2165 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2168 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2169 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2171 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2172 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2174 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2175 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2178 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2179 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2181 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2182 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2185 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2186 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2187 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2189 let ExeDomain = SSEPackedInt in {
2190 let neverHasSideEffects = 1 in {
2191 // 128-bit logical shifts.
2192 def VPSLLDQri : PDIi8<0x73, MRM7r,
2193 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2194 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2196 def VPSRLDQri : PDIi8<0x73, MRM3r,
2197 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2198 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2200 // PSRADQri doesn't exist in SSE[1-3].
2202 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2203 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2204 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2205 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2206 VR128:$src2)))]>, VEX_4V;
2208 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2209 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2210 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2211 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2212 (memopv2i64 addr:$src2))))]>,
2217 let Constraints = "$src1 = $dst" in {
2218 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2219 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2220 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2221 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2222 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2223 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2225 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2226 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2227 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2228 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2229 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2230 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2232 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2233 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2234 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2235 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2237 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2238 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2239 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2241 let ExeDomain = SSEPackedInt in {
2242 let neverHasSideEffects = 1 in {
2243 // 128-bit logical shifts.
2244 def PSLLDQri : PDIi8<0x73, MRM7r,
2245 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2246 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2247 def PSRLDQri : PDIi8<0x73, MRM3r,
2248 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2249 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2250 // PSRADQri doesn't exist in SSE[1-3].
2252 def PANDNrr : PDI<0xDF, MRMSrcReg,
2253 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2254 "pandn\t{$src2, $dst|$dst, $src2}",
2255 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2258 def PANDNrm : PDI<0xDF, MRMSrcMem,
2259 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2260 "pandn\t{$src2, $dst|$dst, $src2}",
2261 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2262 (memopv2i64 addr:$src2))))]>;
2264 } // Constraints = "$src1 = $dst"
2266 let Predicates = [HasSSE2] in {
2267 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2268 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2269 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2270 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2271 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2272 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2273 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2274 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2275 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2276 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2278 // Shift up / down and insert zero's.
2279 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2280 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2281 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2282 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2285 //===---------------------------------------------------------------------===//
2286 // SSE2 - Packed Integer Comparison Instructions
2287 //===---------------------------------------------------------------------===//
2289 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2290 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2292 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2294 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2296 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2298 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2300 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2304 let Constraints = "$src1 = $dst" in {
2305 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2306 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2307 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2308 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2309 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2310 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2311 } // Constraints = "$src1 = $dst"
2313 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2314 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2315 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2316 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2317 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2318 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2319 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2320 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2321 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2322 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2323 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2324 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2326 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2327 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2328 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2329 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2330 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2331 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2332 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2333 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2334 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2335 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2336 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2337 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2339 //===---------------------------------------------------------------------===//
2340 // SSE2 - Packed Integer Pack Instructions
2341 //===---------------------------------------------------------------------===//
2343 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2344 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2346 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2348 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2352 let Constraints = "$src1 = $dst" in {
2353 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2354 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2355 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2356 } // Constraints = "$src1 = $dst"
2358 //===---------------------------------------------------------------------===//
2359 // SSE2 - Packed Integer Shuffle Instructions
2360 //===---------------------------------------------------------------------===//
2362 let ExeDomain = SSEPackedInt in {
2363 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2365 def ri : Ii8<0x70, MRMSrcReg,
2366 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2367 !strconcat(OpcodeStr,
2368 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2369 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2371 def mi : Ii8<0x70, MRMSrcMem,
2372 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2373 !strconcat(OpcodeStr,
2374 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2375 [(set VR128:$dst, (vt (pshuf_frag:$src2
2376 (bc_frag (memopv2i64 addr:$src1)),
2379 } // ExeDomain = SSEPackedInt
2381 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2382 let AddedComplexity = 5 in
2383 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2386 // SSE2 with ImmT == Imm8 and XS prefix.
2387 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2390 // SSE2 with ImmT == Imm8 and XD prefix.
2391 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2395 let Predicates = [HasSSE2] in {
2396 let AddedComplexity = 5 in
2397 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2399 // SSE2 with ImmT == Imm8 and XS prefix.
2400 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2402 // SSE2 with ImmT == Imm8 and XD prefix.
2403 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2406 //===---------------------------------------------------------------------===//
2407 // SSE2 - Packed Integer Unpack Instructions
2408 //===---------------------------------------------------------------------===//
2410 let ExeDomain = SSEPackedInt in {
2411 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2412 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2413 def rr : PDI<opc, MRMSrcReg,
2414 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2416 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2417 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2418 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2419 def rm : PDI<opc, MRMSrcMem,
2420 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2422 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2423 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2424 [(set VR128:$dst, (unp_frag VR128:$src1,
2425 (bc_frag (memopv2i64
2429 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in {
2430 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2432 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2434 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2437 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2438 /// knew to collapse (bitconvert VT to VT) into its operand.
2439 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2440 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2441 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2443 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2444 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2445 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2446 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2448 (v2i64 (unpckl VR128:$src1,
2449 (memopv2i64 addr:$src2))))]>, VEX_4V;
2451 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2453 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2455 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2458 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2459 /// knew to collapse (bitconvert VT to VT) into its operand.
2460 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2461 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2462 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2464 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2465 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2466 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2467 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 (v2i64 (unpckh VR128:$src1,
2470 (memopv2i64 addr:$src2))))]>, VEX_4V;
2473 let Constraints = "$src1 = $dst" in {
2474 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2475 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2476 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2478 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2479 /// knew to collapse (bitconvert VT to VT) into its operand.
2480 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2481 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2482 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2484 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2485 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2486 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2487 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2489 (v2i64 (unpckl VR128:$src1,
2490 (memopv2i64 addr:$src2))))]>;
2492 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2493 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2494 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2496 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2497 /// knew to collapse (bitconvert VT to VT) into its operand.
2498 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2499 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2500 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2502 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2503 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2504 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2505 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2507 (v2i64 (unpckh VR128:$src1,
2508 (memopv2i64 addr:$src2))))]>;
2511 } // ExeDomain = SSEPackedInt
2513 //===---------------------------------------------------------------------===//
2514 // SSE2 - Packed Integer Extract and Insert
2515 //===---------------------------------------------------------------------===//
2517 let ExeDomain = SSEPackedInt in {
2518 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2519 def rri : Ii8<0xC4, MRMSrcReg,
2520 (outs VR128:$dst), (ins VR128:$src1,
2521 GR32:$src2, i32i8imm:$src3),
2523 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2524 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2526 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2527 def rmi : Ii8<0xC4, MRMSrcMem,
2528 (outs VR128:$dst), (ins VR128:$src1,
2529 i16mem:$src2, i32i8imm:$src3),
2531 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2532 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2534 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2539 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in
2540 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2541 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2542 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2543 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2544 imm:$src2))]>, OpSize, VEX;
2545 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2546 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2547 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2548 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2552 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE2] in
2553 defm PINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2555 let Constraints = "$src1 = $dst" in
2556 defm VPINSRW : sse2_pinsrw, TB, OpSize;
2558 } // ExeDomain = SSEPackedInt
2560 //===---------------------------------------------------------------------===//
2561 // SSE2 - Packed Mask Creation
2562 //===---------------------------------------------------------------------===//
2564 let ExeDomain = SSEPackedInt in {
2566 let isAsmParserOnly = 1 in
2567 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2568 "pmovmskb\t{$src, $dst|$dst, $src}",
2569 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2570 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2571 "pmovmskb\t{$src, $dst|$dst, $src}",
2572 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2574 } // ExeDomain = SSEPackedInt
2576 //===---------------------------------------------------------------------===//
2577 // SSE2 - Conditional Store
2578 //===---------------------------------------------------------------------===//
2580 let ExeDomain = SSEPackedInt in {
2582 let isAsmParserOnly = 1 in {
2584 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2585 (ins VR128:$src, VR128:$mask),
2586 "maskmovdqu\t{$mask, $src|$src, $mask}",
2587 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2589 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2590 (ins VR128:$src, VR128:$mask),
2591 "maskmovdqu\t{$mask, $src|$src, $mask}",
2592 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2596 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2597 "maskmovdqu\t{$mask, $src|$src, $mask}",
2598 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2600 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2601 "maskmovdqu\t{$mask, $src|$src, $mask}",
2602 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2604 } // ExeDomain = SSEPackedInt
2606 //===---------------------------------------------------------------------===//
2607 // SSE2 - Move Doubleword
2608 //===---------------------------------------------------------------------===//
2610 // Move Int Doubleword to Packed Double Int
2611 let isAsmParserOnly = 1 in {
2612 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2613 "movd\t{$src, $dst|$dst, $src}",
2615 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2616 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2617 "movd\t{$src, $dst|$dst, $src}",
2619 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2622 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2623 "movd\t{$src, $dst|$dst, $src}",
2625 (v4i32 (scalar_to_vector GR32:$src)))]>;
2626 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2627 "movd\t{$src, $dst|$dst, $src}",
2629 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2632 // Move Int Doubleword to Single Scalar
2633 let isAsmParserOnly = 1 in {
2634 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2635 "movd\t{$src, $dst|$dst, $src}",
2636 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2638 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2639 "movd\t{$src, $dst|$dst, $src}",
2640 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2643 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2644 "movd\t{$src, $dst|$dst, $src}",
2645 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2647 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2648 "movd\t{$src, $dst|$dst, $src}",
2649 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2651 // Move Packed Doubleword Int to Packed Double Int
2652 let isAsmParserOnly = 1 in {
2653 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2654 "movd\t{$src, $dst|$dst, $src}",
2655 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2657 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2658 (ins i32mem:$dst, VR128:$src),
2659 "movd\t{$src, $dst|$dst, $src}",
2660 [(store (i32 (vector_extract (v4i32 VR128:$src),
2661 (iPTR 0))), addr:$dst)]>, VEX;
2663 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2664 "movd\t{$src, $dst|$dst, $src}",
2665 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2667 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2668 "movd\t{$src, $dst|$dst, $src}",
2669 [(store (i32 (vector_extract (v4i32 VR128:$src),
2670 (iPTR 0))), addr:$dst)]>;
2672 // Move Scalar Single to Double Int
2673 let isAsmParserOnly = 1 in {
2674 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2675 "movd\t{$src, $dst|$dst, $src}",
2676 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2677 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2678 "movd\t{$src, $dst|$dst, $src}",
2679 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2681 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2682 "movd\t{$src, $dst|$dst, $src}",
2683 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2684 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2685 "movd\t{$src, $dst|$dst, $src}",
2686 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2688 // movd / movq to XMM register zero-extends
2689 let AddedComplexity = 15, isAsmParserOnly = 1 in {
2690 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2691 "movd\t{$src, $dst|$dst, $src}",
2692 [(set VR128:$dst, (v4i32 (X86vzmovl
2693 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2695 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2696 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2697 [(set VR128:$dst, (v2i64 (X86vzmovl
2698 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2701 let AddedComplexity = 15 in {
2702 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2703 "movd\t{$src, $dst|$dst, $src}",
2704 [(set VR128:$dst, (v4i32 (X86vzmovl
2705 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2706 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2707 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2708 [(set VR128:$dst, (v2i64 (X86vzmovl
2709 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2712 let AddedComplexity = 20 in {
2713 let isAsmParserOnly = 1 in
2714 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2715 "movd\t{$src, $dst|$dst, $src}",
2717 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2718 (loadi32 addr:$src))))))]>,
2720 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2721 "movd\t{$src, $dst|$dst, $src}",
2723 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2724 (loadi32 addr:$src))))))]>;
2726 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2727 (MOVZDI2PDIrm addr:$src)>;
2728 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2729 (MOVZDI2PDIrm addr:$src)>;
2730 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2731 (MOVZDI2PDIrm addr:$src)>;
2734 //===---------------------------------------------------------------------===//
2735 // SSE2 - Move Quadword
2736 //===---------------------------------------------------------------------===//
2738 // Move Quadword Int to Packed Quadword Int
2739 let isAsmParserOnly = 1 in
2740 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2741 "vmovq\t{$src, $dst|$dst, $src}",
2743 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2744 VEX, Requires<[HasAVX, HasSSE2]>;
2745 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2746 "movq\t{$src, $dst|$dst, $src}",
2748 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2749 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
2751 // Move Packed Quadword Int to Quadword Int
2752 let isAsmParserOnly = 1 in
2753 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2754 "movq\t{$src, $dst|$dst, $src}",
2755 [(store (i64 (vector_extract (v2i64 VR128:$src),
2756 (iPTR 0))), addr:$dst)]>, VEX;
2757 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2758 "movq\t{$src, $dst|$dst, $src}",
2759 [(store (i64 (vector_extract (v2i64 VR128:$src),
2760 (iPTR 0))), addr:$dst)]>;
2762 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
2763 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
2765 // Store / copy lower 64-bits of a XMM register.
2766 let isAsmParserOnly = 1 in
2767 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2768 "movq\t{$src, $dst|$dst, $src}",
2769 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
2770 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2771 "movq\t{$src, $dst|$dst, $src}",
2772 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
2774 let AddedComplexity = 20, isAsmParserOnly = 1 in
2775 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2776 "vmovq\t{$src, $dst|$dst, $src}",
2778 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
2779 (loadi64 addr:$src))))))]>,
2780 XS, VEX, Requires<[HasAVX, HasSSE2]>;
2782 let AddedComplexity = 20 in {
2783 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2784 "movq\t{$src, $dst|$dst, $src}",
2786 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
2787 (loadi64 addr:$src))))))]>,
2788 XS, Requires<[HasSSE2]>;
2790 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
2791 (MOVZQI2PQIrm addr:$src)>;
2792 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
2793 (MOVZQI2PQIrm addr:$src)>;
2794 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
2797 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
2798 // IA32 document. movq xmm1, xmm2 does clear the high bits.
2799 let isAsmParserOnly = 1, AddedComplexity = 15 in
2800 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2801 "vmovq\t{$src, $dst|$dst, $src}",
2802 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
2803 XS, VEX, Requires<[HasAVX, HasSSE2]>;
2804 let AddedComplexity = 15 in
2805 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2806 "movq\t{$src, $dst|$dst, $src}",
2807 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
2808 XS, Requires<[HasSSE2]>;
2810 let AddedComplexity = 20, isAsmParserOnly = 1 in
2811 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2812 "vmovq\t{$src, $dst|$dst, $src}",
2813 [(set VR128:$dst, (v2i64 (X86vzmovl
2814 (loadv2i64 addr:$src))))]>,
2815 XS, VEX, Requires<[HasAVX, HasSSE2]>;
2816 let AddedComplexity = 20 in {
2817 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2818 "movq\t{$src, $dst|$dst, $src}",
2819 [(set VR128:$dst, (v2i64 (X86vzmovl
2820 (loadv2i64 addr:$src))))]>,
2821 XS, Requires<[HasSSE2]>;
2823 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
2824 (MOVZPQILo2PQIrm addr:$src)>;
2827 // Instructions to match in the assembler
2828 let isAsmParserOnly = 1 in {
2829 // This instructions is in fact an alias to movd with 64 bit dst
2830 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2831 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
2832 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2833 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
2836 // Instructions for the disassembler
2837 // xr = XMM register
2840 let isAsmParserOnly = 1 in
2841 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2842 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
2843 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2844 "movq\t{$src, $dst|$dst, $src}", []>, XS;
2846 //===---------------------------------------------------------------------===//
2847 // SSE2 - Misc Instructions
2848 //===---------------------------------------------------------------------===//
2851 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
2852 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
2853 TB, Requires<[HasSSE2]>;
2855 // Load, store, and memory fence
2856 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
2857 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
2858 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
2859 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
2861 // Pause. This "instruction" is encoded as "rep; nop", so even though it
2862 // was introduced with SSE2, it's backward compatible.
2863 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
2865 //TODO: custom lower this so as to never even generate the noop
2866 def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
2868 def : Pat<(membarrier (i8 0), (i8 0), (i8 0), (i8 1), (i8 1)), (SFENCE)>;
2869 def : Pat<(membarrier (i8 1), (i8 0), (i8 0), (i8 0), (i8 1)), (LFENCE)>;
2870 def : Pat<(membarrier (i8 imm), (i8 imm), (i8 imm), (i8 imm),
2873 // Alias instructions that map zero vector to pxor / xorp* for sse.
2874 // We set canFoldAsLoad because this can be converted to a constant-pool
2875 // load of an all-ones value if folding it would be beneficial.
2876 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2877 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
2878 // FIXME: Change encoding to pseudo.
2879 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
2880 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
2882 //===---------------------------------------------------------------------===//
2883 // SSE3 - Conversion Instructions
2884 //===---------------------------------------------------------------------===//
2886 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in {
2887 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2888 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
2889 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2890 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2891 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2892 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
2895 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2896 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
2897 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2898 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
2899 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2900 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
2901 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2902 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
2904 //===---------------------------------------------------------------------===//
2905 // SSE3 - Move Instructions
2906 //===---------------------------------------------------------------------===//
2908 // Replicate Single FP
2909 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
2910 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2911 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
2912 [(set VR128:$dst, (v4f32 (rep_frag
2913 VR128:$src, (undef))))]>;
2914 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2915 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
2916 [(set VR128:$dst, (rep_frag
2917 (memopv4f32 addr:$src), (undef)))]>;
2920 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in {
2921 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
2922 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
2924 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
2925 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
2927 // Replicate Double FP
2928 multiclass sse3_replicate_dfp<string OpcodeStr> {
2929 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2930 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
2931 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
2932 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
2933 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
2935 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
2939 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in
2940 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
2941 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
2943 // Move Unaligned Integer
2944 let isAsmParserOnly = 1 in
2945 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2946 "vlddqu\t{$src, $dst|$dst, $src}",
2947 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
2948 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2949 "lddqu\t{$src, $dst|$dst, $src}",
2950 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
2952 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
2954 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
2956 // Several Move patterns
2957 let AddedComplexity = 5 in {
2958 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
2959 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
2960 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
2961 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
2962 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
2963 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
2964 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
2965 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
2968 // vector_shuffle v1, <undef> <1, 1, 3, 3>
2969 let AddedComplexity = 15 in
2970 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
2971 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
2972 let AddedComplexity = 20 in
2973 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
2974 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
2976 // vector_shuffle v1, <undef> <0, 0, 2, 2>
2977 let AddedComplexity = 15 in
2978 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
2979 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
2980 let AddedComplexity = 20 in
2981 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
2982 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
2984 //===---------------------------------------------------------------------===//
2985 // SSE3 - Arithmetic
2986 //===---------------------------------------------------------------------===//
2988 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, bit Is2Addr = 1> {
2989 def rr : I<0xD0, MRMSrcReg,
2990 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2992 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2993 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2994 [(set VR128:$dst, (Int VR128:$src1,
2996 def rm : I<0xD0, MRMSrcMem,
2997 (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
2999 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3000 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3001 [(set VR128:$dst, (Int VR128:$src1,
3002 (memop addr:$src2)))]>;
3006 let isAsmParserOnly = 1, Predicates = [HasSSE3, HasAVX],
3007 ExeDomain = SSEPackedDouble in {
3008 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", 0>, XD,
3010 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", 0>, OpSize,
3013 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3014 ExeDomain = SSEPackedDouble in {
3015 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps">, XD;
3016 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd">, TB, OpSize;
3019 //===---------------------------------------------------------------------===//
3020 // SSE3 Instructions
3021 //===---------------------------------------------------------------------===//
3024 class S3D_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId, bit Is2Addr = 1>
3025 : S3DI<o, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3027 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3029 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, VR128:$src2)))]>;
3030 class S3D_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId, bit Is2Addr = 1>
3031 : S3DI<o, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
3033 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3034 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3035 [(set VR128:$dst, (v4f32 (IntId VR128:$src1, (memop addr:$src2))))]>;
3036 class S3_Intrr<bits<8> o, string OpcodeStr, Intrinsic IntId, bit Is2Addr = 1>
3037 : S3I<o, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3039 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3040 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3041 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, VR128:$src2)))]>;
3042 class S3_Intrm<bits<8> o, string OpcodeStr, Intrinsic IntId, bit Is2Addr = 1>
3043 : S3I<o, MRMSrcMem, (outs VR128:$dst), (ins VR128:$src1, f128mem:$src2),
3045 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3046 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3047 [(set VR128:$dst, (v2f64 (IntId VR128:$src1, (memopv2f64 addr:$src2))))]>;
3049 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in {
3050 def VHADDPSrr : S3D_Intrr<0x7C, "vhaddps", int_x86_sse3_hadd_ps, 0>, VEX_4V;
3051 def VHADDPSrm : S3D_Intrm<0x7C, "vhaddps", int_x86_sse3_hadd_ps, 0>, VEX_4V;
3052 def VHADDPDrr : S3_Intrr <0x7C, "vhaddpd", int_x86_sse3_hadd_pd, 0>, VEX_4V;
3053 def VHADDPDrm : S3_Intrm <0x7C, "vhaddpd", int_x86_sse3_hadd_pd, 0>, VEX_4V;
3054 def VHSUBPSrr : S3D_Intrr<0x7D, "vhsubps", int_x86_sse3_hsub_ps, 0>, VEX_4V;
3055 def VHSUBPSrm : S3D_Intrm<0x7D, "vhsubps", int_x86_sse3_hsub_ps, 0>, VEX_4V;
3056 def VHSUBPDrr : S3_Intrr <0x7D, "vhsubpd", int_x86_sse3_hsub_pd, 0>, VEX_4V;
3057 def VHSUBPDrm : S3_Intrm <0x7D, "vhsubpd", int_x86_sse3_hsub_pd, 0>, VEX_4V;
3060 let Constraints = "$src1 = $dst" in {
3061 def HADDPSrr : S3D_Intrr<0x7C, "haddps", int_x86_sse3_hadd_ps>;
3062 def HADDPSrm : S3D_Intrm<0x7C, "haddps", int_x86_sse3_hadd_ps>;
3063 def HADDPDrr : S3_Intrr <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
3064 def HADDPDrm : S3_Intrm <0x7C, "haddpd", int_x86_sse3_hadd_pd>;
3065 def HSUBPSrr : S3D_Intrr<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
3066 def HSUBPSrm : S3D_Intrm<0x7D, "hsubps", int_x86_sse3_hsub_ps>;
3067 def HSUBPDrr : S3_Intrr <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
3068 def HSUBPDrm : S3_Intrm <0x7D, "hsubpd", int_x86_sse3_hsub_pd>;
3071 //===---------------------------------------------------------------------===//
3072 // SSSE3 - Packed Absolute Instructions
3073 //===---------------------------------------------------------------------===//
3075 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3076 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3077 PatFrag mem_frag64, PatFrag mem_frag128,
3078 Intrinsic IntId64, Intrinsic IntId128> {
3079 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
3080 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3081 [(set VR64:$dst, (IntId64 VR64:$src))]>;
3083 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
3084 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3086 (IntId64 (bitconvert (mem_frag64 addr:$src))))]>;
3088 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3090 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3091 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3094 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3096 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3099 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3102 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in {
3103 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv8i8, memopv16i8,
3104 int_x86_ssse3_pabs_b,
3105 int_x86_ssse3_pabs_b_128>, VEX;
3106 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv4i16, memopv8i16,
3107 int_x86_ssse3_pabs_w,
3108 int_x86_ssse3_pabs_w_128>, VEX;
3109 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv2i32, memopv4i32,
3110 int_x86_ssse3_pabs_d,
3111 int_x86_ssse3_pabs_d_128>, VEX;
3114 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv8i8, memopv16i8,
3115 int_x86_ssse3_pabs_b,
3116 int_x86_ssse3_pabs_b_128>;
3117 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv4i16, memopv8i16,
3118 int_x86_ssse3_pabs_w,
3119 int_x86_ssse3_pabs_w_128>;
3120 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv2i32, memopv4i32,
3121 int_x86_ssse3_pabs_d,
3122 int_x86_ssse3_pabs_d_128>;
3124 //===---------------------------------------------------------------------===//
3125 // SSSE3 - Packed Binary Operator Instructions
3126 //===---------------------------------------------------------------------===//
3128 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3129 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3130 PatFrag mem_frag64, PatFrag mem_frag128,
3131 Intrinsic IntId64, Intrinsic IntId128,
3133 let isCommutable = 1 in
3134 def rr64 : SS38I<opc, MRMSrcReg, (outs VR64:$dst),
3135 (ins VR64:$src1, VR64:$src2),
3137 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3138 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3139 [(set VR64:$dst, (IntId64 VR64:$src1, VR64:$src2))]>;
3140 def rm64 : SS38I<opc, MRMSrcMem, (outs VR64:$dst),
3141 (ins VR64:$src1, i64mem:$src2),
3143 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3144 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3146 (IntId64 VR64:$src1,
3147 (bitconvert (memopv8i8 addr:$src2))))]>;
3149 let isCommutable = 1 in
3150 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3151 (ins VR128:$src1, VR128:$src2),
3153 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3154 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3155 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3157 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3158 (ins VR128:$src1, i128mem:$src2),
3160 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3161 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3163 (IntId128 VR128:$src1,
3164 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3167 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in {
3168 let isCommutable = 0 in {
3169 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv4i16, memopv8i16,
3170 int_x86_ssse3_phadd_w,
3171 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3172 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv2i32, memopv4i32,
3173 int_x86_ssse3_phadd_d,
3174 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3175 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv4i16, memopv8i16,
3176 int_x86_ssse3_phadd_sw,
3177 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3178 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv4i16, memopv8i16,
3179 int_x86_ssse3_phsub_w,
3180 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3181 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv2i32, memopv4i32,
3182 int_x86_ssse3_phsub_d,
3183 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3184 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv4i16, memopv8i16,
3185 int_x86_ssse3_phsub_sw,
3186 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3187 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv8i8, memopv16i8,
3188 int_x86_ssse3_pmadd_ub_sw,
3189 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3190 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv8i8, memopv16i8,
3191 int_x86_ssse3_pshuf_b,
3192 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3193 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv8i8, memopv16i8,
3194 int_x86_ssse3_psign_b,
3195 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3196 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv4i16, memopv8i16,
3197 int_x86_ssse3_psign_w,
3198 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3199 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv2i32, memopv4i32,
3200 int_x86_ssse3_psign_d,
3201 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3203 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv4i16, memopv8i16,
3204 int_x86_ssse3_pmul_hr_sw,
3205 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3208 // None of these have i8 immediate fields.
3209 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3210 let isCommutable = 0 in {
3211 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv4i16, memopv8i16,
3212 int_x86_ssse3_phadd_w,
3213 int_x86_ssse3_phadd_w_128>;
3214 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv2i32, memopv4i32,
3215 int_x86_ssse3_phadd_d,
3216 int_x86_ssse3_phadd_d_128>;
3217 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv4i16, memopv8i16,
3218 int_x86_ssse3_phadd_sw,
3219 int_x86_ssse3_phadd_sw_128>;
3220 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv4i16, memopv8i16,
3221 int_x86_ssse3_phsub_w,
3222 int_x86_ssse3_phsub_w_128>;
3223 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv2i32, memopv4i32,
3224 int_x86_ssse3_phsub_d,
3225 int_x86_ssse3_phsub_d_128>;
3226 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv4i16, memopv8i16,
3227 int_x86_ssse3_phsub_sw,
3228 int_x86_ssse3_phsub_sw_128>;
3229 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv8i8, memopv16i8,
3230 int_x86_ssse3_pmadd_ub_sw,
3231 int_x86_ssse3_pmadd_ub_sw_128>;
3232 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv8i8, memopv16i8,
3233 int_x86_ssse3_pshuf_b,
3234 int_x86_ssse3_pshuf_b_128>;
3235 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv8i8, memopv16i8,
3236 int_x86_ssse3_psign_b,
3237 int_x86_ssse3_psign_b_128>;
3238 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv4i16, memopv8i16,
3239 int_x86_ssse3_psign_w,
3240 int_x86_ssse3_psign_w_128>;
3241 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv2i32, memopv4i32,
3242 int_x86_ssse3_psign_d,
3243 int_x86_ssse3_psign_d_128>;
3245 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv4i16, memopv8i16,
3246 int_x86_ssse3_pmul_hr_sw,
3247 int_x86_ssse3_pmul_hr_sw_128>;
3250 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3251 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3252 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3253 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3255 //===---------------------------------------------------------------------===//
3256 // SSSE3 - Packed Align Instruction Patterns
3257 //===---------------------------------------------------------------------===//
3259 multiclass sse3_palign<string asm, bit Is2Addr = 1> {
3260 def R64rr : SS3AI<0x0F, MRMSrcReg, (outs VR64:$dst),
3261 (ins VR64:$src1, VR64:$src2, i8imm:$src3),
3263 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3265 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3267 def R64rm : SS3AI<0x0F, MRMSrcMem, (outs VR64:$dst),
3268 (ins VR64:$src1, i64mem:$src2, i8imm:$src3),
3270 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3272 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3275 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3276 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3278 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3280 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3282 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3283 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3285 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3287 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3291 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE3] in
3292 defm VPALIGN : sse3_palign<"vpalignr", 0>, VEX_4V;
3293 let Constraints = "$src1 = $dst" in
3294 defm PALIGN : sse3_palign<"palignr">;
3296 let AddedComplexity = 5 in {
3298 def : Pat<(v1i64 (palign:$src3 VR64:$src1, VR64:$src2)),
3299 (PALIGNR64rr VR64:$src2, VR64:$src1,
3300 (SHUFFLE_get_palign_imm VR64:$src3))>,
3301 Requires<[HasSSSE3]>;
3302 def : Pat<(v2i32 (palign:$src3 VR64:$src1, VR64:$src2)),
3303 (PALIGNR64rr VR64:$src2, VR64:$src1,
3304 (SHUFFLE_get_palign_imm VR64:$src3))>,
3305 Requires<[HasSSSE3]>;
3306 def : Pat<(v4i16 (palign:$src3 VR64:$src1, VR64:$src2)),
3307 (PALIGNR64rr VR64:$src2, VR64:$src1,
3308 (SHUFFLE_get_palign_imm VR64:$src3))>,
3309 Requires<[HasSSSE3]>;
3310 def : Pat<(v8i8 (palign:$src3 VR64:$src1, VR64:$src2)),
3311 (PALIGNR64rr VR64:$src2, VR64:$src1,
3312 (SHUFFLE_get_palign_imm VR64:$src3))>,
3313 Requires<[HasSSSE3]>;
3315 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3316 (PALIGNR128rr VR128:$src2, VR128:$src1,
3317 (SHUFFLE_get_palign_imm VR128:$src3))>,
3318 Requires<[HasSSSE3]>;
3319 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3320 (PALIGNR128rr VR128:$src2, VR128:$src1,
3321 (SHUFFLE_get_palign_imm VR128:$src3))>,
3322 Requires<[HasSSSE3]>;
3323 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3324 (PALIGNR128rr VR128:$src2, VR128:$src1,
3325 (SHUFFLE_get_palign_imm VR128:$src3))>,
3326 Requires<[HasSSSE3]>;
3327 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3328 (PALIGNR128rr VR128:$src2, VR128:$src1,
3329 (SHUFFLE_get_palign_imm VR128:$src3))>,
3330 Requires<[HasSSSE3]>;
3333 //===---------------------------------------------------------------------===//
3334 // SSSE3 Misc Instructions
3335 //===---------------------------------------------------------------------===//
3337 // Thread synchronization
3338 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3339 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3340 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3341 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3343 //===---------------------------------------------------------------------===//
3344 // Non-Instruction Patterns
3345 //===---------------------------------------------------------------------===//
3347 // extload f32 -> f64. This matches load+fextend because we have a hack in
3348 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3350 // Since these loads aren't folded into the fextend, we have to match it
3352 let Predicates = [HasSSE2] in
3353 def : Pat<(fextend (loadf32 addr:$src)),
3354 (CVTSS2SDrm addr:$src)>;
3357 let Predicates = [HasSSE2] in {
3358 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3359 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3360 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3361 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3362 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3363 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3364 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3365 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3366 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3367 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3368 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3369 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3370 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3371 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3372 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3373 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3374 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3375 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3376 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3377 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3378 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3379 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3380 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3381 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3382 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3383 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3384 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3385 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3386 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3387 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3390 // Move scalar to XMM zero-extended
3391 // movd to XMM register zero-extends
3392 let AddedComplexity = 15 in {
3393 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3394 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3395 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3396 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3397 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3398 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3399 (MOVSSrr (v4f32 (V_SET0PS)),
3400 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3401 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3402 (MOVSSrr (v4i32 (V_SET0PI)),
3403 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3406 // Splat v2f64 / v2i64
3407 let AddedComplexity = 10 in {
3408 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3409 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3410 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3411 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3412 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3413 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3414 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3415 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3418 // Special unary SHUFPSrri case.
3419 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3420 (SHUFPSrri VR128:$src1, VR128:$src1,
3421 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3422 let AddedComplexity = 5 in
3423 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3424 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3425 Requires<[HasSSE2]>;
3426 // Special unary SHUFPDrri case.
3427 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3428 (SHUFPDrri VR128:$src1, VR128:$src1,
3429 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3430 Requires<[HasSSE2]>;
3431 // Special unary SHUFPDrri case.
3432 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3433 (SHUFPDrri VR128:$src1, VR128:$src1,
3434 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3435 Requires<[HasSSE2]>;
3436 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3437 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3438 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3439 Requires<[HasSSE2]>;
3441 // Special binary v4i32 shuffle cases with SHUFPS.
3442 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3443 (SHUFPSrri VR128:$src1, VR128:$src2,
3444 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3445 Requires<[HasSSE2]>;
3446 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3447 (SHUFPSrmi VR128:$src1, addr:$src2,
3448 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3449 Requires<[HasSSE2]>;
3450 // Special binary v2i64 shuffle cases using SHUFPDrri.
3451 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3452 (SHUFPDrri VR128:$src1, VR128:$src2,
3453 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3454 Requires<[HasSSE2]>;
3456 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3457 let AddedComplexity = 15 in {
3458 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3459 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3460 Requires<[OptForSpeed, HasSSE2]>;
3461 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3462 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3463 Requires<[OptForSpeed, HasSSE2]>;
3465 let AddedComplexity = 10 in {
3466 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3467 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3468 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3469 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3470 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3471 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3472 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3473 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3476 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3477 let AddedComplexity = 15 in {
3478 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3479 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3480 Requires<[OptForSpeed, HasSSE2]>;
3481 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3482 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3483 Requires<[OptForSpeed, HasSSE2]>;
3485 let AddedComplexity = 10 in {
3486 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3487 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3488 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3489 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3490 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3491 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3492 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3493 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3496 let AddedComplexity = 20 in {
3497 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3498 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3499 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3501 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3502 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3503 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3505 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3506 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3507 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3508 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3509 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3512 let AddedComplexity = 20 in {
3513 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3514 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3515 (MOVLPSrm VR128:$src1, addr:$src2)>;
3516 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3517 (MOVLPDrm VR128:$src1, addr:$src2)>;
3518 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3519 (MOVLPSrm VR128:$src1, addr:$src2)>;
3520 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3521 (MOVLPDrm VR128:$src1, addr:$src2)>;
3524 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3525 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3526 (MOVLPSmr addr:$src1, VR128:$src2)>;
3527 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3528 (MOVLPDmr addr:$src1, VR128:$src2)>;
3529 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3531 (MOVLPSmr addr:$src1, VR128:$src2)>;
3532 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3533 (MOVLPDmr addr:$src1, VR128:$src2)>;
3535 let AddedComplexity = 15 in {
3536 // Setting the lowest element in the vector.
3537 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3538 (MOVSSrr (v4i32 VR128:$src1),
3539 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3540 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3541 (MOVSDrr (v2i64 VR128:$src1),
3542 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3544 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3545 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3546 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3547 Requires<[HasSSE2]>;
3548 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3549 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3550 Requires<[HasSSE2]>;
3553 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3554 // fall back to this for SSE1)
3555 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3556 (SHUFPSrri VR128:$src2, VR128:$src1,
3557 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3559 // Set lowest element and zero upper elements.
3560 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3561 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3563 // Some special case pandn patterns.
3564 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3566 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3567 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3569 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3570 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3572 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3574 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3575 (memop addr:$src2))),
3576 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3577 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3578 (memop addr:$src2))),
3579 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3580 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3581 (memop addr:$src2))),
3582 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3584 // vector -> vector casts
3585 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3586 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3587 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3588 (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3589 def : Pat<(v2f64 (sint_to_fp (v2i32 VR64:$src))),
3590 (Int_CVTPI2PDrr VR64:$src)>, Requires<[HasSSE2]>;
3591 def : Pat<(v2i32 (fp_to_sint (v2f64 VR128:$src))),
3592 (Int_CVTTPD2PIrr VR128:$src)>, Requires<[HasSSE2]>;
3594 // Use movaps / movups for SSE integer load / store (one byte shorter).
3595 def : Pat<(alignedloadv4i32 addr:$src),
3596 (MOVAPSrm addr:$src)>;
3597 def : Pat<(loadv4i32 addr:$src),
3598 (MOVUPSrm addr:$src)>;
3599 def : Pat<(alignedloadv2i64 addr:$src),
3600 (MOVAPSrm addr:$src)>;
3601 def : Pat<(loadv2i64 addr:$src),
3602 (MOVUPSrm addr:$src)>;
3604 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3605 (MOVAPSmr addr:$dst, VR128:$src)>;
3606 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3607 (MOVAPSmr addr:$dst, VR128:$src)>;
3608 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3609 (MOVAPSmr addr:$dst, VR128:$src)>;
3610 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3611 (MOVAPSmr addr:$dst, VR128:$src)>;
3612 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3613 (MOVUPSmr addr:$dst, VR128:$src)>;
3614 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3615 (MOVUPSmr addr:$dst, VR128:$src)>;
3616 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3617 (MOVUPSmr addr:$dst, VR128:$src)>;
3618 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3619 (MOVUPSmr addr:$dst, VR128:$src)>;
3621 //===----------------------------------------------------------------------===//
3622 // SSE4.1 - Packed Move with Sign/Zero Extend
3623 //===----------------------------------------------------------------------===//
3625 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3626 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3627 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3628 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3630 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3631 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3633 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3637 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
3638 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3640 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3642 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3644 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3646 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3648 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3652 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3653 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3654 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3655 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3656 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3657 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3659 // Common patterns involving scalar load.
3660 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3661 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3662 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3663 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3665 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3666 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3667 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3668 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3670 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3671 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3672 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3673 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3675 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3676 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3677 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3678 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3680 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3681 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3682 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3683 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3685 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3686 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3687 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3688 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3691 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3692 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3693 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3694 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3696 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3699 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
3703 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
3704 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
3706 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
3708 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
3710 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
3714 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
3715 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
3716 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
3717 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
3719 // Common patterns involving scalar load
3720 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
3721 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
3722 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
3723 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
3725 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
3726 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
3727 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
3728 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
3731 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3732 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3733 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3734 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3736 // Expecting a i16 load any extended to i32 value.
3737 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
3738 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3739 [(set VR128:$dst, (IntId (bitconvert
3740 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
3744 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
3745 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
3747 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
3750 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
3751 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
3753 // Common patterns involving scalar load
3754 def : Pat<(int_x86_sse41_pmovsxbq
3755 (bitconvert (v4i32 (X86vzmovl
3756 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
3757 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
3759 def : Pat<(int_x86_sse41_pmovzxbq
3760 (bitconvert (v4i32 (X86vzmovl
3761 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
3762 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
3764 //===----------------------------------------------------------------------===//
3765 // SSE4.1 - Extract Instructions
3766 //===----------------------------------------------------------------------===//
3768 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
3769 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
3770 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
3771 (ins VR128:$src1, i32i8imm:$src2),
3772 !strconcat(OpcodeStr,
3773 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3774 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
3776 def mr : SS4AIi8<opc, MRMDestMem, (outs),
3777 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
3778 !strconcat(OpcodeStr,
3779 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3782 // There's an AssertZext in the way of writing the store pattern
3783 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
3786 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3787 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
3789 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
3792 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
3793 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
3794 def mr : SS4AIi8<opc, MRMDestMem, (outs),
3795 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
3796 !strconcat(OpcodeStr,
3797 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3800 // There's an AssertZext in the way of writing the store pattern
3801 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
3804 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3805 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
3807 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
3810 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
3811 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
3812 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
3813 (ins VR128:$src1, i32i8imm:$src2),
3814 !strconcat(OpcodeStr,
3815 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3817 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
3818 def mr : SS4AIi8<opc, MRMDestMem, (outs),
3819 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
3820 !strconcat(OpcodeStr,
3821 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3822 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
3823 addr:$dst)]>, OpSize;
3826 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3827 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
3829 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
3831 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
3832 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
3833 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
3834 (ins VR128:$src1, i32i8imm:$src2),
3835 !strconcat(OpcodeStr,
3836 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3838 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
3839 def mr : SS4AIi8<opc, MRMDestMem, (outs),
3840 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
3841 !strconcat(OpcodeStr,
3842 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3843 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
3844 addr:$dst)]>, OpSize, REX_W;
3847 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3848 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
3850 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
3852 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
3854 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
3855 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
3856 (ins VR128:$src1, i32i8imm:$src2),
3857 !strconcat(OpcodeStr,
3858 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3860 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
3862 def mr : SS4AIi8<opc, MRMDestMem, (outs),
3863 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
3864 !strconcat(OpcodeStr,
3865 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3866 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
3867 addr:$dst)]>, OpSize;
3870 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3871 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
3872 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
3874 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
3875 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
3878 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
3879 Requires<[HasSSE41]>;
3881 //===----------------------------------------------------------------------===//
3882 // SSE4.1 - Insert Instructions
3883 //===----------------------------------------------------------------------===//
3885 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
3886 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
3887 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
3889 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3891 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3893 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
3894 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
3895 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
3897 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3899 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3901 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
3902 imm:$src3))]>, OpSize;
3905 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3906 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
3907 let Constraints = "$src1 = $dst" in
3908 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
3910 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
3911 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
3912 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
3914 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3916 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3918 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
3920 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
3921 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
3923 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3925 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3927 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
3928 imm:$src3)))]>, OpSize;
3931 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3932 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
3933 let Constraints = "$src1 = $dst" in
3934 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
3936 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
3937 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
3938 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3940 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3942 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3944 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
3946 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
3947 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
3949 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3951 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3953 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
3954 imm:$src3)))]>, OpSize;
3957 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3958 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
3959 let Constraints = "$src1 = $dst" in
3960 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
3962 // insertps has a few different modes, there's the first two here below which
3963 // are optimized inserts that won't zero arbitrary elements in the destination
3964 // vector. The next one matches the intrinsic and could zero arbitrary elements
3965 // in the target vector.
3966 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
3967 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
3968 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
3970 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3972 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3974 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
3976 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
3977 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
3979 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3981 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3983 (X86insrtps VR128:$src1,
3984 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
3985 imm:$src3))]>, OpSize;
3988 let Constraints = "$src1 = $dst" in
3989 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
3990 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
3991 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
3993 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
3994 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>;
3996 //===----------------------------------------------------------------------===//
3997 // SSE4.1 - Round Instructions
3998 //===----------------------------------------------------------------------===//
4000 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd,
4003 Intrinsic V2F64Int> {
4004 // Intrinsic operation, reg.
4005 // Vector intrinsic operation, reg
4006 def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
4007 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4008 !strconcat(OpcodeStr,
4009 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4010 [(set VR128:$dst, (V4F32Int VR128:$src1, imm:$src2))]>,
4013 // Vector intrinsic operation, mem
4014 def PSm_Int : Ii8<opcps, MRMSrcMem,
4015 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4016 !strconcat(OpcodeStr,
4017 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4019 (V4F32Int (memopv4f32 addr:$src1),imm:$src2))]>,
4021 Requires<[HasSSE41]>;
4023 // Vector intrinsic operation, reg
4024 def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
4025 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4026 !strconcat(OpcodeStr,
4027 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4028 [(set VR128:$dst, (V2F64Int VR128:$src1, imm:$src2))]>,
4031 // Vector intrinsic operation, mem
4032 def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
4033 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4034 !strconcat(OpcodeStr,
4035 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4037 (V2F64Int (memopv2f64 addr:$src1),imm:$src2))]>,
4041 multiclass sse41_fp_unop_rm_avx<bits<8> opcps, bits<8> opcpd,
4043 // Intrinsic operation, reg.
4044 // Vector intrinsic operation, reg
4045 def PSr : SS4AIi8<opcps, MRMSrcReg,
4046 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4047 !strconcat(OpcodeStr,
4048 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4051 // Vector intrinsic operation, mem
4052 def PSm : Ii8<opcps, MRMSrcMem,
4053 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4054 !strconcat(OpcodeStr,
4055 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4056 []>, TA, OpSize, Requires<[HasSSE41]>;
4058 // Vector intrinsic operation, reg
4059 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4060 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
4061 !strconcat(OpcodeStr,
4062 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4065 // Vector intrinsic operation, mem
4066 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4067 (outs VR128:$dst), (ins f128mem:$src1, i32i8imm:$src2),
4068 !strconcat(OpcodeStr,
4069 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4073 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4076 Intrinsic F64Int, bit Is2Addr = 1> {
4077 // Intrinsic operation, reg.
4078 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
4079 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4081 !strconcat(OpcodeStr,
4082 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4083 !strconcat(OpcodeStr,
4084 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4085 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4088 // Intrinsic operation, mem.
4089 def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
4090 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4092 !strconcat(OpcodeStr,
4093 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4094 !strconcat(OpcodeStr,
4095 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4097 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4100 // Intrinsic operation, reg.
4101 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
4102 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4104 !strconcat(OpcodeStr,
4105 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4106 !strconcat(OpcodeStr,
4107 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4108 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4111 // Intrinsic operation, mem.
4112 def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
4113 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4115 !strconcat(OpcodeStr,
4116 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4117 !strconcat(OpcodeStr,
4118 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4120 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4124 multiclass sse41_fp_binop_rm_avx<bits<8> opcss, bits<8> opcsd,
4126 // Intrinsic operation, reg.
4127 def SSr : SS4AIi8<opcss, MRMSrcReg,
4128 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4129 !strconcat(OpcodeStr,
4130 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4133 // Intrinsic operation, mem.
4134 def SSm : SS4AIi8<opcss, MRMSrcMem,
4135 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4136 !strconcat(OpcodeStr,
4137 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4140 // Intrinsic operation, reg.
4141 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4142 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4143 !strconcat(OpcodeStr,
4144 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4147 // Intrinsic operation, mem.
4148 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4149 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4150 !strconcat(OpcodeStr,
4151 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4155 // FP round - roundss, roundps, roundsd, roundpd
4156 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
4158 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround",
4159 int_x86_sse41_round_ps, int_x86_sse41_round_pd>,
4161 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4162 int_x86_sse41_round_ss, int_x86_sse41_round_sd,
4164 // Instructions for the assembler
4165 defm VROUND : sse41_fp_unop_rm_avx<0x08, 0x09, "vround">, VEX;
4166 defm VROUND : sse41_fp_binop_rm_avx<0x0A, 0x0B, "vround">, VEX_4V;
4169 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round",
4170 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4171 let Constraints = "$src1 = $dst" in
4172 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4173 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4175 //===----------------------------------------------------------------------===//
4176 // SSE4.1 - Misc Instructions
4177 //===----------------------------------------------------------------------===//
4179 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4180 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4181 Intrinsic IntId128> {
4182 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4184 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4185 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4186 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4188 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4191 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4194 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
4195 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4196 int_x86_sse41_phminposuw>, VEX;
4197 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4198 int_x86_sse41_phminposuw>;
4200 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4201 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4202 Intrinsic IntId128, bit Is2Addr = 1> {
4203 let isCommutable = 1 in
4204 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4205 (ins VR128:$src1, VR128:$src2),
4207 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4208 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4209 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4210 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4211 (ins VR128:$src1, i128mem:$src2),
4213 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4214 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4216 (IntId128 VR128:$src1,
4217 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4220 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
4221 let isCommutable = 0 in
4222 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4224 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4226 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4228 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4230 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4232 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4234 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4236 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4238 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4240 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4242 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4246 let Constraints = "$src1 = $dst" in {
4247 let isCommutable = 0 in
4248 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4249 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4250 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4251 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4252 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4253 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4254 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4255 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4256 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4257 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4258 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4261 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4262 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4263 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4264 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4266 /// SS48I_binop_rm - Simple SSE41 binary operator.
4267 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4268 ValueType OpVT, bit Is2Addr = 1> {
4269 let isCommutable = 1 in
4270 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4271 (ins VR128:$src1, VR128:$src2),
4273 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4274 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4275 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4277 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4278 (ins VR128:$src1, i128mem:$src2),
4280 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4281 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4282 [(set VR128:$dst, (OpNode VR128:$src1,
4283 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4287 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
4288 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4289 let Constraints = "$src1 = $dst" in
4290 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4292 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4293 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4294 Intrinsic IntId128, bit Is2Addr = 1> {
4295 let isCommutable = 1 in
4296 def rri : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4297 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4299 !strconcat(OpcodeStr,
4300 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4301 !strconcat(OpcodeStr,
4302 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4304 (IntId128 VR128:$src1, VR128:$src2, imm:$src3))]>,
4306 def rmi : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4307 (ins VR128:$src1, i128mem:$src2, i32i8imm:$src3),
4309 !strconcat(OpcodeStr,
4310 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4311 !strconcat(OpcodeStr,
4312 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4314 (IntId128 VR128:$src1,
4315 (bitconvert (memopv16i8 addr:$src2)), imm:$src3))]>,
4319 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
4320 let isCommutable = 0 in {
4321 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4323 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4325 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4327 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4330 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4332 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4336 let Constraints = "$src1 = $dst" in {
4337 let isCommutable = 0 in {
4338 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps>;
4339 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd>;
4340 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw>;
4341 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw>;
4343 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps>;
4344 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd>;
4347 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4348 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
4349 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr> {
4350 def rr : I<opc, MRMSrcReg, (outs VR128:$dst),
4351 (ins VR128:$src1, VR128:$src2, VR128:$src3),
4352 !strconcat(OpcodeStr,
4353 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4354 [], SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4356 def rm : I<opc, MRMSrcMem, (outs VR128:$dst),
4357 (ins VR128:$src1, i128mem:$src2, VR128:$src3),
4358 !strconcat(OpcodeStr,
4359 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4360 [], SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4364 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd">;
4365 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps">;
4366 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb">;
4368 /// SS41I_ternary_int - SSE 4.1 ternary operator
4369 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4370 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4371 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4372 (ins VR128:$src1, VR128:$src2),
4373 !strconcat(OpcodeStr,
4374 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4375 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4378 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4379 (ins VR128:$src1, i128mem:$src2),
4380 !strconcat(OpcodeStr,
4381 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4384 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4388 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4389 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4390 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4392 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4393 // the intel intrinsic that corresponds to this.
4394 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in {
4395 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4396 "vptest\t{$src2, $src1|$src1, $src2}",
4397 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4399 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4400 "vptest\t{$src2, $src1|$src1, $src2}",
4401 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4405 let Defs = [EFLAGS] in {
4406 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4407 "ptest \t{$src2, $src1|$src1, $src2}",
4408 [(set EFLAGS, (X86ptest VR128:$src1, VR128:$src2))]>,
4410 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, i128mem:$src2),
4411 "ptest \t{$src2, $src1|$src1, $src2}",
4412 [(set EFLAGS, (X86ptest VR128:$src1, (load addr:$src2)))]>,
4416 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE41] in
4417 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4418 "vmovntdqa\t{$src, $dst|$dst, $src}",
4419 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4421 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4422 "movntdqa\t{$src, $dst|$dst, $src}",
4423 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4426 //===----------------------------------------------------------------------===//
4427 // SSE4.2 - Compare Instructions
4428 //===----------------------------------------------------------------------===//
4430 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4431 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4432 Intrinsic IntId128, bit Is2Addr = 1> {
4433 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4434 (ins VR128:$src1, VR128:$src2),
4436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4438 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4440 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4441 (ins VR128:$src1, i128mem:$src2),
4443 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4444 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4446 (IntId128 VR128:$src1,
4447 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4450 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in
4451 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4453 let Constraints = "$src1 = $dst" in
4454 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4456 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4457 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4458 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4459 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4461 //===----------------------------------------------------------------------===//
4462 // SSE4.2 - String/text Processing Instructions
4463 //===----------------------------------------------------------------------===//
4465 // Packed Compare Implicit Length Strings, Return Mask
4466 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4467 def PCMPISTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
4468 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4469 "#PCMPISTRM128rr PSEUDO!",
4470 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4471 imm:$src3))]>, OpSize;
4472 def PCMPISTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
4473 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4474 "#PCMPISTRM128rm PSEUDO!",
4475 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4476 VR128:$src1, (load addr:$src2), imm:$src3))]>, OpSize;
4479 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4480 Predicates = [HasAVX, HasSSE42] in {
4481 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4482 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4483 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4484 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4485 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4486 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4489 let Defs = [XMM0, EFLAGS] in {
4490 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4491 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4492 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4493 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4494 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4495 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4498 // Packed Compare Explicit Length Strings, Return Mask
4499 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4500 def PCMPESTRM128REG : SS42AI<0, Pseudo, (outs VR128:$dst),
4501 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4502 "#PCMPESTRM128rr PSEUDO!",
4504 (int_x86_sse42_pcmpestrm128
4505 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>, OpSize;
4507 def PCMPESTRM128MEM : SS42AI<0, Pseudo, (outs VR128:$dst),
4508 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4509 "#PCMPESTRM128rm PSEUDO!",
4510 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4511 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>,
4515 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42],
4516 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4517 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4518 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4519 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4520 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4521 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4522 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4525 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4526 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4527 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4528 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4529 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4530 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4531 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4534 // Packed Compare Implicit Length Strings, Return Index
4535 let Defs = [ECX, EFLAGS] in {
4536 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4537 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4538 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4539 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4540 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4541 (implicit EFLAGS)]>, OpSize;
4542 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4543 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4544 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4545 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4546 (implicit EFLAGS)]>, OpSize;
4550 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in {
4551 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4553 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4555 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4557 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4559 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4561 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4565 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4566 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4567 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4568 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4569 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4570 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4572 // Packed Compare Explicit Length Strings, Return Index
4573 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4574 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4575 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4576 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4577 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4578 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4579 (implicit EFLAGS)]>, OpSize;
4580 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4581 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4582 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4584 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4585 (implicit EFLAGS)]>, OpSize;
4589 let isAsmParserOnly = 1, Predicates = [HasAVX, HasSSE42] in {
4590 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4592 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
4594 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
4596 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
4598 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
4600 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
4604 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
4605 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
4606 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
4607 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
4608 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
4609 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
4611 //===----------------------------------------------------------------------===//
4612 // SSE4.2 - CRC Instructions
4613 //===----------------------------------------------------------------------===//
4615 // No CRC instructions have AVX equivalents
4617 // crc intrinsic instruction
4618 // This set of instructions are only rm, the only difference is the size
4620 let Constraints = "$src1 = $dst" in {
4621 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
4622 (ins GR32:$src1, i8mem:$src2),
4623 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4625 (int_x86_sse42_crc32_8 GR32:$src1,
4626 (load addr:$src2)))]>;
4627 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
4628 (ins GR32:$src1, GR8:$src2),
4629 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4631 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
4632 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4633 (ins GR32:$src1, i16mem:$src2),
4634 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4636 (int_x86_sse42_crc32_16 GR32:$src1,
4637 (load addr:$src2)))]>,
4639 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4640 (ins GR32:$src1, GR16:$src2),
4641 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4643 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
4645 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4646 (ins GR32:$src1, i32mem:$src2),
4647 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4649 (int_x86_sse42_crc32_32 GR32:$src1,
4650 (load addr:$src2)))]>;
4651 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4652 (ins GR32:$src1, GR32:$src2),
4653 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4655 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
4656 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
4657 (ins GR64:$src1, i8mem:$src2),
4658 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4660 (int_x86_sse42_crc64_8 GR64:$src1,
4661 (load addr:$src2)))]>,
4663 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
4664 (ins GR64:$src1, GR8:$src2),
4665 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4667 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
4669 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
4670 (ins GR64:$src1, i64mem:$src2),
4671 "crc32{q} \t{$src2, $src1|$src1, $src2}",
4673 (int_x86_sse42_crc64_64 GR64:$src1,
4674 (load addr:$src2)))]>,
4676 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
4677 (ins GR64:$src1, GR64:$src2),
4678 "crc32{q} \t{$src2, $src1|$src1, $src2}",
4680 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
4684 //===----------------------------------------------------------------------===//
4685 // AES-NI Instructions
4686 //===----------------------------------------------------------------------===//
4688 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
4689 Intrinsic IntId128, bit Is2Addr = 1> {
4690 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
4691 (ins VR128:$src1, VR128:$src2),
4693 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4694 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4695 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4697 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
4698 (ins VR128:$src1, i128mem:$src2),
4700 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4701 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4703 (IntId128 VR128:$src1,
4704 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4707 // Perform One Round of an AES Encryption/Decryption Flow
4708 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
4709 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
4710 int_x86_aesni_aesenc, 0>, VEX_4V;
4711 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
4712 int_x86_aesni_aesenclast, 0>, VEX_4V;
4713 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
4714 int_x86_aesni_aesdec, 0>, VEX_4V;
4715 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
4716 int_x86_aesni_aesdeclast, 0>, VEX_4V;
4719 let Constraints = "$src1 = $dst" in {
4720 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
4721 int_x86_aesni_aesenc>;
4722 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
4723 int_x86_aesni_aesenclast>;
4724 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
4725 int_x86_aesni_aesdec>;
4726 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
4727 int_x86_aesni_aesdeclast>;
4730 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
4731 (AESENCrr VR128:$src1, VR128:$src2)>;
4732 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
4733 (AESENCrm VR128:$src1, addr:$src2)>;
4734 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
4735 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
4736 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
4737 (AESENCLASTrm VR128:$src1, addr:$src2)>;
4738 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
4739 (AESDECrr VR128:$src1, VR128:$src2)>;
4740 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
4741 (AESDECrm VR128:$src1, addr:$src2)>;
4742 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
4743 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
4744 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
4745 (AESDECLASTrm VR128:$src1, addr:$src2)>;
4747 // Perform the AES InvMixColumn Transformation
4748 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
4749 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
4751 "vaesimc\t{$src1, $dst|$dst, $src1}",
4753 (int_x86_aesni_aesimc VR128:$src1))]>,
4755 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
4756 (ins i128mem:$src1),
4757 "vaesimc\t{$src1, $dst|$dst, $src1}",
4759 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
4762 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
4764 "aesimc\t{$src1, $dst|$dst, $src1}",
4766 (int_x86_aesni_aesimc VR128:$src1))]>,
4768 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
4769 (ins i128mem:$src1),
4770 "aesimc\t{$src1, $dst|$dst, $src1}",
4772 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
4775 // AES Round Key Generation Assist
4776 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
4777 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
4778 (ins VR128:$src1, i8imm:$src2),
4779 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4781 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
4783 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
4784 (ins i128mem:$src1, i8imm:$src2),
4785 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4787 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
4791 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
4792 (ins VR128:$src1, i8imm:$src2),
4793 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4795 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
4797 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
4798 (ins i128mem:$src1, i8imm:$src2),
4799 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4801 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),