1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE scalar FP Instructions
19 //===----------------------------------------------------------------------===//
21 // CMOV* - Used to implement the SSE SELECT DAG operation. Expanded after
22 // instruction selection into a branch sequence.
23 let Uses = [EFLAGS], usesCustomInserter = 1 in {
24 def CMOV_FR32 : I<0, Pseudo,
25 (outs FR32:$dst), (ins FR32:$t, FR32:$f, i8imm:$cond),
27 [(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond,
29 def CMOV_FR64 : I<0, Pseudo,
30 (outs FR64:$dst), (ins FR64:$t, FR64:$f, i8imm:$cond),
32 [(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond,
34 def CMOV_V4F32 : I<0, Pseudo,
35 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
36 "#CMOV_V4F32 PSEUDO!",
38 (v4f32 (X86cmov VR128:$t, VR128:$f, imm:$cond,
40 def CMOV_V2F64 : I<0, Pseudo,
41 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
42 "#CMOV_V2F64 PSEUDO!",
44 (v2f64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
46 def CMOV_V2I64 : I<0, Pseudo,
47 (outs VR128:$dst), (ins VR128:$t, VR128:$f, i8imm:$cond),
48 "#CMOV_V2I64 PSEUDO!",
50 (v2i64 (X86cmov VR128:$t, VR128:$f, imm:$cond,
54 //===----------------------------------------------------------------------===//
55 // SSE 1 & 2 Instructions Classes
56 //===----------------------------------------------------------------------===//
58 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
59 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
60 RegisterClass RC, X86MemOperand x86memop,
62 let isCommutable = 1 in {
63 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
65 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
66 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
67 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
69 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
71 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
72 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
73 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
76 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
77 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
78 string asm, string SSEVer, string FPSizeStr,
79 Operand memopr, ComplexPattern mem_cpat,
81 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
83 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
84 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
85 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
86 !strconcat(SSEVer, !strconcat("_",
87 !strconcat(OpcodeStr, FPSizeStr))))
88 RC:$src1, RC:$src2))]>;
89 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
91 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
92 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
93 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_sse",
94 !strconcat(SSEVer, !strconcat("_",
95 !strconcat(OpcodeStr, FPSizeStr))))
96 RC:$src1, mem_cpat:$src2))]>;
99 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
100 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
101 RegisterClass RC, ValueType vt,
102 X86MemOperand x86memop, PatFrag mem_frag,
103 Domain d, bit Is2Addr = 1> {
104 let isCommutable = 1 in
105 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
107 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
108 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
109 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
111 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
113 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
114 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
115 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
118 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
119 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
120 string OpcodeStr, X86MemOperand x86memop,
121 list<dag> pat_rr, list<dag> pat_rm,
123 let isCommutable = 1 in
124 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
126 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
127 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
129 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
131 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
132 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
136 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
137 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
138 string asm, string SSEVer, string FPSizeStr,
139 X86MemOperand x86memop, PatFrag mem_frag,
140 Domain d, bit Is2Addr = 1> {
141 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
143 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
144 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
145 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
146 !strconcat(SSEVer, !strconcat("_",
147 !strconcat(OpcodeStr, FPSizeStr))))
148 RC:$src1, RC:$src2))], d>;
149 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
151 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
152 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
153 [(set RC:$dst, (!nameconcat<Intrinsic>("int_x86_",
154 !strconcat(SSEVer, !strconcat("_",
155 !strconcat(OpcodeStr, FPSizeStr))))
156 RC:$src1, (mem_frag addr:$src2)))], d>;
159 //===----------------------------------------------------------------------===//
160 // SSE 1 & 2 - Move Instructions
161 //===----------------------------------------------------------------------===//
163 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
164 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
165 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
167 // Loading from memory automatically zeroing upper bits.
168 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
169 PatFrag mem_pat, string OpcodeStr> :
170 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
171 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
172 [(set RC:$dst, (mem_pat addr:$src))]>;
174 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
175 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
176 // is used instead. Register-to-register movss/movsd is not modeled as an
177 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
178 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
179 let isAsmParserOnly = 1 in {
180 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
181 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
182 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
183 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
185 let canFoldAsLoad = 1, isReMaterializable = 1 in {
186 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
188 let AddedComplexity = 20 in
189 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
193 let Constraints = "$src1 = $dst" in {
194 def MOVSSrr : sse12_move_rr<FR32, v4f32,
195 "movss\t{$src2, $dst|$dst, $src2}">, XS;
196 def MOVSDrr : sse12_move_rr<FR64, v2f64,
197 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
200 let canFoldAsLoad = 1, isReMaterializable = 1 in {
201 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
203 let AddedComplexity = 20 in
204 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
207 let AddedComplexity = 15 in {
208 // Extract the low 32-bit value from one vector and insert it into another.
209 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
210 (MOVSSrr (v4f32 VR128:$src1),
211 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
212 // Extract the low 64-bit value from one vector and insert it into another.
213 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
214 (MOVSDrr (v2f64 VR128:$src1),
215 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
218 // Implicitly promote a 32-bit scalar to a vector.
219 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
220 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
221 // Implicitly promote a 64-bit scalar to a vector.
222 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
223 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
225 let AddedComplexity = 20 in {
226 // MOVSSrm zeros the high parts of the register; represent this
227 // with SUBREG_TO_REG.
228 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
229 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
230 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
231 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
232 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
233 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
234 // MOVSDrm zeros the high parts of the register; represent this
235 // with SUBREG_TO_REG.
236 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
237 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
238 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
239 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
240 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
241 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
242 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
243 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
244 def : Pat<(v2f64 (X86vzload addr:$src)),
245 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
248 // Store scalar value to memory.
249 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
250 "movss\t{$src, $dst|$dst, $src}",
251 [(store FR32:$src, addr:$dst)]>;
252 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
253 "movsd\t{$src, $dst|$dst, $src}",
254 [(store FR64:$src, addr:$dst)]>;
256 let isAsmParserOnly = 1 in {
257 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
258 "movss\t{$src, $dst|$dst, $src}",
259 [(store FR32:$src, addr:$dst)]>, XS, VEX;
260 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
261 "movsd\t{$src, $dst|$dst, $src}",
262 [(store FR64:$src, addr:$dst)]>, XD, VEX;
265 // Extract and store.
266 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
269 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
270 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
273 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
275 // Move Aligned/Unaligned floating point values
276 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
277 X86MemOperand x86memop, PatFrag ld_frag,
278 string asm, Domain d,
279 bit IsReMaterializable = 1> {
280 let neverHasSideEffects = 1 in
281 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
282 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
283 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
284 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
285 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
286 [(set RC:$dst, (ld_frag addr:$src))], d>;
289 let isAsmParserOnly = 1 in {
290 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
291 "movaps", SSEPackedSingle>, VEX;
292 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
293 "movapd", SSEPackedDouble>, OpSize, VEX;
294 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
295 "movups", SSEPackedSingle>, VEX;
296 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
297 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
299 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
300 "movaps", SSEPackedSingle>, VEX;
301 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
302 "movapd", SSEPackedDouble>, OpSize, VEX;
303 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
304 "movups", SSEPackedSingle>, VEX;
305 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
306 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
308 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
309 "movaps", SSEPackedSingle>, TB;
310 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
311 "movapd", SSEPackedDouble>, TB, OpSize;
312 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
313 "movups", SSEPackedSingle>, TB;
314 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
315 "movupd", SSEPackedDouble, 0>, TB, OpSize;
317 let isAsmParserOnly = 1 in {
318 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movaps\t{$src, $dst|$dst, $src}",
320 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
321 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
322 "movapd\t{$src, $dst|$dst, $src}",
323 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
324 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
327 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
328 "movupd\t{$src, $dst|$dst, $src}",
329 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
330 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
331 "movaps\t{$src, $dst|$dst, $src}",
332 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
333 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
334 "movapd\t{$src, $dst|$dst, $src}",
335 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
336 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
337 "movups\t{$src, $dst|$dst, $src}",
338 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
339 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
340 "movupd\t{$src, $dst|$dst, $src}",
341 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
344 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
345 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
346 (VMOVUPSYmr addr:$dst, VR256:$src)>;
348 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
349 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
350 (VMOVUPDYmr addr:$dst, VR256:$src)>;
352 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
353 "movaps\t{$src, $dst|$dst, $src}",
354 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
355 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
356 "movapd\t{$src, $dst|$dst, $src}",
357 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
358 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
359 "movups\t{$src, $dst|$dst, $src}",
360 [(store (v4f32 VR128:$src), addr:$dst)]>;
361 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
362 "movupd\t{$src, $dst|$dst, $src}",
363 [(store (v2f64 VR128:$src), addr:$dst)]>;
365 // Intrinsic forms of MOVUPS/D load and store
366 let isAsmParserOnly = 1 in {
367 let canFoldAsLoad = 1, isReMaterializable = 1 in
368 def VMOVUPSrm_Int : VPSI<0x10, MRMSrcMem, (outs VR128:$dst),
370 "movups\t{$src, $dst|$dst, $src}",
371 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>, VEX;
372 def VMOVUPDrm_Int : VPDI<0x10, MRMSrcMem, (outs VR128:$dst),
374 "movupd\t{$src, $dst|$dst, $src}",
375 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>, VEX;
376 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
377 (ins f128mem:$dst, VR128:$src),
378 "movups\t{$src, $dst|$dst, $src}",
379 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
380 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
381 (ins f128mem:$dst, VR128:$src),
382 "movupd\t{$src, $dst|$dst, $src}",
383 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
385 let canFoldAsLoad = 1, isReMaterializable = 1 in
386 def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
387 "movups\t{$src, $dst|$dst, $src}",
388 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
389 def MOVUPDrm_Int : PDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
390 "movupd\t{$src, $dst|$dst, $src}",
391 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
393 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
394 "movups\t{$src, $dst|$dst, $src}",
395 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
396 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
397 "movupd\t{$src, $dst|$dst, $src}",
398 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
400 // Move Low/High packed floating point values
401 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
402 PatFrag mov_frag, string base_opc,
404 def PSrm : PI<opc, MRMSrcMem,
405 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
406 !strconcat(!strconcat(base_opc,"s"), asm_opr),
409 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
410 SSEPackedSingle>, TB;
412 def PDrm : PI<opc, MRMSrcMem,
413 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
414 !strconcat(!strconcat(base_opc,"d"), asm_opr),
415 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
416 (scalar_to_vector (loadf64 addr:$src2)))))],
417 SSEPackedDouble>, TB, OpSize;
420 let isAsmParserOnly = 1, AddedComplexity = 20 in {
421 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
422 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
423 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
424 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
426 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
427 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
428 "\t{$src2, $dst|$dst, $src2}">;
429 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
430 "\t{$src2, $dst|$dst, $src2}">;
433 let isAsmParserOnly = 1 in {
434 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
435 "movlps\t{$src, $dst|$dst, $src}",
436 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
437 (iPTR 0))), addr:$dst)]>, VEX;
438 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
439 "movlpd\t{$src, $dst|$dst, $src}",
440 [(store (f64 (vector_extract (v2f64 VR128:$src),
441 (iPTR 0))), addr:$dst)]>, VEX;
443 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
444 "movlps\t{$src, $dst|$dst, $src}",
445 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
446 (iPTR 0))), addr:$dst)]>;
447 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
448 "movlpd\t{$src, $dst|$dst, $src}",
449 [(store (f64 (vector_extract (v2f64 VR128:$src),
450 (iPTR 0))), addr:$dst)]>;
452 // v2f64 extract element 1 is always custom lowered to unpack high to low
453 // and extract element 0 so the non-store version isn't too horrible.
454 let isAsmParserOnly = 1 in {
455 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
456 "movhps\t{$src, $dst|$dst, $src}",
457 [(store (f64 (vector_extract
458 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
459 (undef)), (iPTR 0))), addr:$dst)]>,
461 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
462 "movhpd\t{$src, $dst|$dst, $src}",
463 [(store (f64 (vector_extract
464 (v2f64 (unpckh VR128:$src, (undef))),
465 (iPTR 0))), addr:$dst)]>,
468 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
469 "movhps\t{$src, $dst|$dst, $src}",
470 [(store (f64 (vector_extract
471 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
472 (undef)), (iPTR 0))), addr:$dst)]>;
473 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
474 "movhpd\t{$src, $dst|$dst, $src}",
475 [(store (f64 (vector_extract
476 (v2f64 (unpckh VR128:$src, (undef))),
477 (iPTR 0))), addr:$dst)]>;
479 let isAsmParserOnly = 1, AddedComplexity = 20 in {
480 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
481 (ins VR128:$src1, VR128:$src2),
482 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
484 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
486 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
487 (ins VR128:$src1, VR128:$src2),
488 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
490 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
493 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
494 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
495 (ins VR128:$src1, VR128:$src2),
496 "movlhps\t{$src2, $dst|$dst, $src2}",
498 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
499 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
500 (ins VR128:$src1, VR128:$src2),
501 "movhlps\t{$src2, $dst|$dst, $src2}",
503 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
506 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
507 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
508 let AddedComplexity = 20 in {
509 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
510 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
511 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
512 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
515 //===----------------------------------------------------------------------===//
516 // SSE 1 & 2 - Conversion Instructions
517 //===----------------------------------------------------------------------===//
519 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
520 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
522 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
523 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
524 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
525 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
528 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
529 X86MemOperand x86memop, string asm> {
530 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
532 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
536 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
537 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
538 string asm, Domain d> {
539 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
540 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
541 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
542 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
545 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
546 X86MemOperand x86memop, string asm> {
547 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
548 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
549 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
550 (ins DstRC:$src1, x86memop:$src),
551 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
554 let isAsmParserOnly = 1 in {
555 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
556 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
557 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
558 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
560 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
561 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
562 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
563 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
566 // The assembler can recognize rr 64-bit instructions by seeing a rxx
567 // register, but the same isn't true when only using memory operands,
568 // provide other assembly "l" and "q" forms to address this explicitly
569 // where appropriate to do so.
570 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
572 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
574 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
576 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
578 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
582 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
583 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
584 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
585 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
586 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
587 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
588 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
589 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
590 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
591 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
592 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
593 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
594 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
595 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
596 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
597 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
599 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
600 // and/or XMM operand(s).
602 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
603 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
605 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
606 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
607 [(set DstRC:$dst, (Int SrcRC:$src))]>;
608 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
609 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
610 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
613 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
614 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
615 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
616 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
618 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
619 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
620 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
621 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
622 (ins DstRC:$src1, x86memop:$src2),
624 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
625 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
626 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
629 let isAsmParserOnly = 1 in {
630 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
631 f32mem, load, "cvtss2si">, XS, VEX;
632 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
633 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
635 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
636 f128mem, load, "cvtsd2si">, XD, VEX;
637 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
638 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
641 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
642 // Get rid of this hack or rename the intrinsics, there are several
643 // intructions that only match with the intrinsic form, why create duplicates
644 // to let them be recognized by the assembler?
645 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
646 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
647 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
648 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
650 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
651 f32mem, load, "cvtss2si">, XS;
652 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
653 f32mem, load, "cvtss2si{q}">, XS, REX_W;
654 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
655 f128mem, load, "cvtsd2si">, XD;
656 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
657 f128mem, load, "cvtsd2si">, XD, REX_W;
660 let isAsmParserOnly = 1 in {
661 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
662 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
663 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
664 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
666 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
667 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
668 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
669 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
673 let Constraints = "$src1 = $dst" in {
674 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
675 int_x86_sse_cvtsi2ss, i32mem, loadi32,
677 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
678 int_x86_sse_cvtsi642ss, i64mem, loadi64,
679 "cvtsi2ss{q}">, XS, REX_W;
680 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
681 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
683 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
684 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
685 "cvtsi2sd">, XD, REX_W;
690 // Aliases for intrinsics
691 let isAsmParserOnly = 1 in {
692 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
693 f32mem, load, "cvttss2si">, XS, VEX;
694 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
695 int_x86_sse_cvttss2si64, f32mem, load,
696 "cvttss2si">, XS, VEX, VEX_W;
697 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
698 f128mem, load, "cvttss2si">, XD, VEX;
699 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
700 int_x86_sse2_cvttsd2si64, f128mem, load,
701 "cvttss2si">, XD, VEX, VEX_W;
703 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
704 f32mem, load, "cvttss2si">, XS;
705 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
706 int_x86_sse_cvttss2si64, f32mem, load,
707 "cvttss2si{q}">, XS, REX_W;
708 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
709 f128mem, load, "cvttss2si">, XD;
710 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
711 int_x86_sse2_cvttsd2si64, f128mem, load,
712 "cvttss2si{q}">, XD, REX_W;
714 let isAsmParserOnly = 1, Pattern = []<dag> in {
715 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
716 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
717 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
718 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
720 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
721 "cvtdq2ps\t{$src, $dst|$dst, $src}",
722 SSEPackedSingle>, TB, VEX;
723 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
724 "cvtdq2ps\t{$src, $dst|$dst, $src}",
725 SSEPackedSingle>, TB, VEX;
727 let Pattern = []<dag> in {
728 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
729 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
730 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
731 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
732 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
733 "cvtdq2ps\t{$src, $dst|$dst, $src}",
734 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
739 // Convert scalar double to scalar single
740 let isAsmParserOnly = 1 in {
741 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
742 (ins FR64:$src1, FR64:$src2),
743 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
745 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
746 (ins FR64:$src1, f64mem:$src2),
747 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
748 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
750 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
751 "cvtsd2ss\t{$src, $dst|$dst, $src}",
752 [(set FR32:$dst, (fround FR64:$src))]>;
753 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
754 "cvtsd2ss\t{$src, $dst|$dst, $src}",
755 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
756 Requires<[HasSSE2, OptForSize]>;
758 let isAsmParserOnly = 1 in
759 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
760 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
762 let Constraints = "$src1 = $dst" in
763 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
764 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
766 // Convert scalar single to scalar double
767 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
768 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
769 (ins FR32:$src1, FR32:$src2),
770 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
771 []>, XS, Requires<[HasAVX]>, VEX_4V;
772 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
773 (ins FR32:$src1, f32mem:$src2),
774 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
775 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
777 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
778 "cvtss2sd\t{$src, $dst|$dst, $src}",
779 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
781 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
782 "cvtss2sd\t{$src, $dst|$dst, $src}",
783 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
784 Requires<[HasSSE2, OptForSize]>;
786 let isAsmParserOnly = 1 in {
787 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
788 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
789 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
790 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
791 VR128:$src2))]>, XS, VEX_4V,
793 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
794 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
795 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
796 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
797 (load addr:$src2)))]>, XS, VEX_4V,
800 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
801 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
802 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
803 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
804 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
807 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
808 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
809 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
810 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
811 (load addr:$src2)))]>, XS,
815 def : Pat<(extloadf32 addr:$src),
816 (CVTSS2SDrr (MOVSSrm addr:$src))>,
817 Requires<[HasSSE2, OptForSpeed]>;
819 // Convert doubleword to packed single/double fp
820 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
821 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
822 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
823 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
824 TB, VEX, Requires<[HasAVX]>;
825 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
826 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
827 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
828 (bitconvert (memopv2i64 addr:$src))))]>,
829 TB, VEX, Requires<[HasAVX]>;
831 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
832 "cvtdq2ps\t{$src, $dst|$dst, $src}",
833 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
834 TB, Requires<[HasSSE2]>;
835 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
836 "cvtdq2ps\t{$src, $dst|$dst, $src}",
837 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
838 (bitconvert (memopv2i64 addr:$src))))]>,
839 TB, Requires<[HasSSE2]>;
841 // FIXME: why the non-intrinsic version is described as SSE3?
842 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
843 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
844 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
845 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
846 XS, VEX, Requires<[HasAVX]>;
847 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
848 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
849 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
850 (bitconvert (memopv2i64 addr:$src))))]>,
851 XS, VEX, Requires<[HasAVX]>;
853 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
854 "cvtdq2pd\t{$src, $dst|$dst, $src}",
855 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
856 XS, Requires<[HasSSE2]>;
857 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
858 "cvtdq2pd\t{$src, $dst|$dst, $src}",
859 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
860 (bitconvert (memopv2i64 addr:$src))))]>,
861 XS, Requires<[HasSSE2]>;
864 // Convert packed single/double fp to doubleword
865 let isAsmParserOnly = 1 in {
866 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
867 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
868 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
869 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
870 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
871 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
872 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
873 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
875 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
876 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
877 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
878 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
880 let isAsmParserOnly = 1 in {
881 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
882 "cvtps2dq\t{$src, $dst|$dst, $src}",
883 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
885 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
887 "cvtps2dq\t{$src, $dst|$dst, $src}",
888 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
889 (memop addr:$src)))]>, VEX;
891 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
892 "cvtps2dq\t{$src, $dst|$dst, $src}",
893 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
894 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
895 "cvtps2dq\t{$src, $dst|$dst, $src}",
896 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
897 (memop addr:$src)))]>;
899 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
900 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
901 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
902 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
903 XD, VEX, Requires<[HasAVX]>;
904 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
905 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
906 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
907 (memop addr:$src)))]>,
908 XD, VEX, Requires<[HasAVX]>;
910 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
911 "cvtpd2dq\t{$src, $dst|$dst, $src}",
912 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
913 XD, Requires<[HasSSE2]>;
914 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
915 "cvtpd2dq\t{$src, $dst|$dst, $src}",
916 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
917 (memop addr:$src)))]>,
918 XD, Requires<[HasSSE2]>;
921 // Convert with truncation packed single/double fp to doubleword
922 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
923 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
924 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
925 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
926 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
927 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
928 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
929 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
930 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
932 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
933 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
934 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
935 "cvttps2dq\t{$src, $dst|$dst, $src}", []>;
938 let isAsmParserOnly = 1 in {
939 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
940 "vcvttps2dq\t{$src, $dst|$dst, $src}",
942 (int_x86_sse2_cvttps2dq VR128:$src))]>,
943 XS, VEX, Requires<[HasAVX]>;
944 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
945 "vcvttps2dq\t{$src, $dst|$dst, $src}",
946 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
947 (memop addr:$src)))]>,
948 XS, VEX, Requires<[HasAVX]>;
950 def Int_CVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvttps2dq\t{$src, $dst|$dst, $src}",
953 (int_x86_sse2_cvttps2dq VR128:$src))]>,
954 XS, Requires<[HasSSE2]>;
955 def Int_CVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
956 "cvttps2dq\t{$src, $dst|$dst, $src}",
957 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
958 (memop addr:$src)))]>,
959 XS, Requires<[HasSSE2]>;
961 let isAsmParserOnly = 1 in {
962 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
964 "cvttpd2dq\t{$src, $dst|$dst, $src}",
965 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
967 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
969 "cvttpd2dq\t{$src, $dst|$dst, $src}",
970 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
971 (memop addr:$src)))]>, VEX;
973 def Int_CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
974 "cvttpd2dq\t{$src, $dst|$dst, $src}",
975 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
976 def Int_CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
977 "cvttpd2dq\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
979 (memop addr:$src)))]>;
981 let isAsmParserOnly = 1 in {
982 // The assembler can recognize rr 256-bit instructions by seeing a ymm
983 // register, but the same isn't true when using memory operands instead.
984 // Provide other assembly rr and rm forms to address this explicitly.
985 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
986 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
987 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
988 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
991 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
992 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
993 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
994 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
997 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
998 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
999 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1000 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1003 // Convert packed single to packed double
1004 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1005 // SSE2 instructions without OpSize prefix
1006 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1007 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1008 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1009 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1010 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1011 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1012 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1013 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1015 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1016 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1017 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1018 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1020 let isAsmParserOnly = 1 in {
1021 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1022 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1023 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1024 VEX, Requires<[HasAVX]>;
1025 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1026 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1027 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1028 (load addr:$src)))]>,
1029 VEX, Requires<[HasAVX]>;
1031 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1032 "cvtps2pd\t{$src, $dst|$dst, $src}",
1033 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1034 TB, Requires<[HasSSE2]>;
1035 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1036 "cvtps2pd\t{$src, $dst|$dst, $src}",
1037 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1038 (load addr:$src)))]>,
1039 TB, Requires<[HasSSE2]>;
1041 // Convert packed double to packed single
1042 let isAsmParserOnly = 1 in {
1043 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1044 // register, but the same isn't true when using memory operands instead.
1045 // Provide other assembly rr and rm forms to address this explicitly.
1046 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1047 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1048 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1049 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1052 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1053 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1054 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1055 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1058 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1059 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1060 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1061 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1063 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1064 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1065 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1066 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1069 let isAsmParserOnly = 1 in {
1070 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1071 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1072 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1073 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1075 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1076 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1077 (memop addr:$src)))]>;
1079 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1080 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1081 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1082 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1083 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1084 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1085 (memop addr:$src)))]>;
1087 // AVX 256-bit register conversion intrinsics
1088 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1089 // whenever possible to avoid declaring two versions of each one.
1090 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1091 (VCVTDQ2PSYrr VR256:$src)>;
1092 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1093 (VCVTDQ2PSYrm addr:$src)>;
1095 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1096 (VCVTPD2PSYrr VR256:$src)>;
1097 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1098 (VCVTPD2PSYrm addr:$src)>;
1100 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1101 (VCVTPS2DQYrr VR256:$src)>;
1102 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1103 (VCVTPS2DQYrm addr:$src)>;
1105 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1106 (VCVTPS2PDYrr VR128:$src)>;
1107 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1108 (VCVTPS2PDYrm addr:$src)>;
1110 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1111 (VCVTTPD2DQYrr VR256:$src)>;
1112 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1113 (VCVTTPD2DQYrm addr:$src)>;
1115 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1116 (VCVTTPS2DQYrr VR256:$src)>;
1117 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1118 (VCVTTPS2DQYrm addr:$src)>;
1120 //===----------------------------------------------------------------------===//
1121 // SSE 1 & 2 - Compare Instructions
1122 //===----------------------------------------------------------------------===//
1124 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1125 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1126 string asm, string asm_alt> {
1127 def rr : SIi8<0xC2, MRMSrcReg,
1128 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1131 def rm : SIi8<0xC2, MRMSrcMem,
1132 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1134 // Accept explicit immediate argument form instead of comparison code.
1135 let isAsmParserOnly = 1 in {
1136 def rr_alt : SIi8<0xC2, MRMSrcReg,
1137 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1140 def rm_alt : SIi8<0xC2, MRMSrcMem,
1141 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1146 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1147 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1148 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1149 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1151 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1152 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1153 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1157 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1158 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1159 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1160 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1161 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1162 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1163 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1166 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1167 Intrinsic Int, string asm> {
1168 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1169 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1170 [(set VR128:$dst, (Int VR128:$src1,
1171 VR128:$src, imm:$cc))]>;
1172 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1173 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1174 [(set VR128:$dst, (Int VR128:$src1,
1175 (load addr:$src), imm:$cc))]>;
1178 // Aliases to match intrinsics which expect XMM operand(s).
1179 let isAsmParserOnly = 1 in {
1180 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1181 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1183 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1184 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1187 let Constraints = "$src1 = $dst" in {
1188 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1189 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1190 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1191 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1195 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1196 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1197 ValueType vt, X86MemOperand x86memop,
1198 PatFrag ld_frag, string OpcodeStr, Domain d> {
1199 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1200 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1201 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1202 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1203 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1204 [(set EFLAGS, (OpNode (vt RC:$src1),
1205 (ld_frag addr:$src2)))], d>;
1208 let Defs = [EFLAGS] in {
1209 let isAsmParserOnly = 1 in {
1210 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1211 "ucomiss", SSEPackedSingle>, VEX;
1212 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1213 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1214 let Pattern = []<dag> in {
1215 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1216 "comiss", SSEPackedSingle>, VEX;
1217 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1218 "comisd", SSEPackedDouble>, OpSize, VEX;
1221 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1222 load, "ucomiss", SSEPackedSingle>, VEX;
1223 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1224 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1226 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1227 load, "comiss", SSEPackedSingle>, VEX;
1228 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1229 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1231 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1232 "ucomiss", SSEPackedSingle>, TB;
1233 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1234 "ucomisd", SSEPackedDouble>, TB, OpSize;
1236 let Pattern = []<dag> in {
1237 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1238 "comiss", SSEPackedSingle>, TB;
1239 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1240 "comisd", SSEPackedDouble>, TB, OpSize;
1243 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1244 load, "ucomiss", SSEPackedSingle>, TB;
1245 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1246 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1248 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1249 "comiss", SSEPackedSingle>, TB;
1250 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1251 "comisd", SSEPackedDouble>, TB, OpSize;
1252 } // Defs = [EFLAGS]
1254 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1255 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1256 Intrinsic Int, string asm, string asm_alt,
1258 def rri : PIi8<0xC2, MRMSrcReg,
1259 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1260 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1261 def rmi : PIi8<0xC2, MRMSrcMem,
1262 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1263 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1264 // Accept explicit immediate argument form instead of comparison code.
1265 let isAsmParserOnly = 1 in {
1266 def rri_alt : PIi8<0xC2, MRMSrcReg,
1267 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1269 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1270 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1275 let isAsmParserOnly = 1 in {
1276 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1277 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1278 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1279 SSEPackedSingle>, VEX_4V;
1280 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1281 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1282 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1283 SSEPackedDouble>, OpSize, VEX_4V;
1284 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1285 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1286 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1287 SSEPackedSingle>, VEX_4V;
1288 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1289 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1290 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1291 SSEPackedDouble>, OpSize, VEX_4V;
1293 let Constraints = "$src1 = $dst" in {
1294 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1295 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1296 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1297 SSEPackedSingle>, TB;
1298 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1299 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1300 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1301 SSEPackedDouble>, TB, OpSize;
1304 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1305 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1306 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1307 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1308 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1309 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1310 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1311 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1313 //===----------------------------------------------------------------------===//
1314 // SSE 1 & 2 - Shuffle Instructions
1315 //===----------------------------------------------------------------------===//
1317 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1318 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1319 ValueType vt, string asm, PatFrag mem_frag,
1320 Domain d, bit IsConvertibleToThreeAddress = 0> {
1321 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1322 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1323 [(set RC:$dst, (vt (shufp:$src3
1324 RC:$src1, (mem_frag addr:$src2))))], d>;
1325 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1326 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1327 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1329 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1332 let isAsmParserOnly = 1 in {
1333 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1334 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1335 memopv4f32, SSEPackedSingle>, VEX_4V;
1336 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1337 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1338 memopv8f32, SSEPackedSingle>, VEX_4V;
1339 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1340 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1341 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1342 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1343 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1344 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1347 let Constraints = "$src1 = $dst" in {
1348 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1349 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1350 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1352 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1353 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1354 memopv2f64, SSEPackedDouble>, TB, OpSize;
1357 //===----------------------------------------------------------------------===//
1358 // SSE 1 & 2 - Unpack Instructions
1359 //===----------------------------------------------------------------------===//
1361 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1362 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1363 PatFrag mem_frag, RegisterClass RC,
1364 X86MemOperand x86memop, string asm,
1366 def rr : PI<opc, MRMSrcReg,
1367 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1369 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1370 def rm : PI<opc, MRMSrcMem,
1371 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1373 (vt (OpNode RC:$src1,
1374 (mem_frag addr:$src2))))], d>;
1377 let AddedComplexity = 10 in {
1378 let isAsmParserOnly = 1 in {
1379 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1380 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1381 SSEPackedSingle>, VEX_4V;
1382 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1383 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1384 SSEPackedDouble>, OpSize, VEX_4V;
1385 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1386 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1387 SSEPackedSingle>, VEX_4V;
1388 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1389 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1390 SSEPackedDouble>, OpSize, VEX_4V;
1392 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1393 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1394 SSEPackedSingle>, VEX_4V;
1395 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1396 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1397 SSEPackedDouble>, OpSize, VEX_4V;
1398 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1399 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1400 SSEPackedSingle>, VEX_4V;
1401 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1402 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1403 SSEPackedDouble>, OpSize, VEX_4V;
1406 let Constraints = "$src1 = $dst" in {
1407 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1408 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1409 SSEPackedSingle>, TB;
1410 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1411 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1412 SSEPackedDouble>, TB, OpSize;
1413 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1414 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1415 SSEPackedSingle>, TB;
1416 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1417 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1418 SSEPackedDouble>, TB, OpSize;
1419 } // Constraints = "$src1 = $dst"
1420 } // AddedComplexity
1422 //===----------------------------------------------------------------------===//
1423 // SSE 1 & 2 - Extract Floating-Point Sign mask
1424 //===----------------------------------------------------------------------===//
1426 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1427 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1429 def rr : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1430 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1431 [(set GR32:$dst, (Int RC:$src))], d>;
1435 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1436 SSEPackedSingle>, TB;
1437 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1438 SSEPackedDouble>, TB, OpSize;
1440 let isAsmParserOnly = 1 in {
1441 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1442 "movmskps", SSEPackedSingle>, VEX;
1443 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1444 "movmskpd", SSEPackedDouble>, OpSize,
1446 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1447 "movmskps", SSEPackedSingle>, VEX;
1448 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1449 "movmskpd", SSEPackedDouble>, OpSize,
1453 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1454 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1455 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1456 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1458 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1459 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1460 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1461 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1465 //===----------------------------------------------------------------------===//
1466 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1467 //===----------------------------------------------------------------------===//
1469 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1470 // names that start with 'Fs'.
1472 // Alias instructions that map fld0 to pxor for sse.
1473 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1474 canFoldAsLoad = 1 in {
1475 // FIXME: Set encoding to pseudo!
1476 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1477 [(set FR32:$dst, fp32imm0)]>,
1478 Requires<[HasSSE1]>, TB, OpSize;
1479 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1480 [(set FR64:$dst, fpimm0)]>,
1481 Requires<[HasSSE2]>, TB, OpSize;
1484 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1485 // bits are disregarded.
1486 let neverHasSideEffects = 1 in {
1487 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1488 "movaps\t{$src, $dst|$dst, $src}", []>;
1489 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1490 "movapd\t{$src, $dst|$dst, $src}", []>;
1493 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1494 // bits are disregarded.
1495 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1496 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1497 "movaps\t{$src, $dst|$dst, $src}",
1498 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1499 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1500 "movapd\t{$src, $dst|$dst, $src}",
1501 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1504 //===----------------------------------------------------------------------===//
1505 // SSE 1 & 2 - Logical Instructions
1506 //===----------------------------------------------------------------------===//
1508 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1510 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1512 let isAsmParserOnly = 1 in {
1513 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1514 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1516 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1517 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1520 let Constraints = "$src1 = $dst" in {
1521 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1522 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1524 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1525 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1529 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1530 let mayLoad = 0 in {
1531 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1532 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1533 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1536 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1537 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1539 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1541 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1542 SDNode OpNode, int HasPat = 0,
1543 list<list<dag>> Pattern = []> {
1544 let isAsmParserOnly = 1, Pattern = []<dag> in {
1545 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1546 !strconcat(OpcodeStr, "ps"), f128mem,
1547 !if(HasPat, Pattern[0], // rr
1548 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1550 !if(HasPat, Pattern[2], // rm
1551 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1552 (memopv2i64 addr:$src2)))]), 0>,
1555 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1556 !strconcat(OpcodeStr, "pd"), f128mem,
1557 !if(HasPat, Pattern[1], // rr
1558 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1561 !if(HasPat, Pattern[3], // rm
1562 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1563 (memopv2i64 addr:$src2)))]), 0>,
1566 let Constraints = "$src1 = $dst" in {
1567 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1568 !strconcat(OpcodeStr, "ps"), f128mem,
1569 !if(HasPat, Pattern[0], // rr
1570 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1572 !if(HasPat, Pattern[2], // rm
1573 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1574 (memopv2i64 addr:$src2)))])>, TB;
1576 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1577 !strconcat(OpcodeStr, "pd"), f128mem,
1578 !if(HasPat, Pattern[1], // rr
1579 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1582 !if(HasPat, Pattern[3], // rm
1583 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1584 (memopv2i64 addr:$src2)))])>,
1589 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1591 let isAsmParserOnly = 1 in {
1592 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1593 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1594 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1596 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1597 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1601 // AVX 256-bit packed logical ops forms
1602 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1603 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1604 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1605 let isCommutable = 0 in
1606 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1608 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1609 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1610 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1611 let isCommutable = 0 in
1612 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1614 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1615 (bc_v2i64 (v4i32 immAllOnesV))),
1618 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1619 (bc_v2i64 (v2f64 VR128:$src2))))],
1621 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1622 (bc_v2i64 (v4i32 immAllOnesV))),
1623 (memopv2i64 addr:$src2))))],
1625 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1626 (memopv2i64 addr:$src2)))]]>;
1628 //===----------------------------------------------------------------------===//
1629 // SSE 1 & 2 - Arithmetic Instructions
1630 //===----------------------------------------------------------------------===//
1632 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1635 /// In addition, we also have a special variant of the scalar form here to
1636 /// represent the associated intrinsic operation. This form is unlike the
1637 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1638 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1640 /// These three forms can each be reg+reg or reg+mem.
1643 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1645 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1647 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1648 OpNode, FR32, f32mem, Is2Addr>, XS;
1649 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1650 OpNode, FR64, f64mem, Is2Addr>, XD;
1653 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1655 let mayLoad = 0 in {
1656 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1657 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1658 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1659 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1663 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1665 let mayLoad = 0 in {
1666 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1667 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1668 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1669 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1673 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1675 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1676 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1677 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1678 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1681 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1683 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1684 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1685 SSEPackedSingle, Is2Addr>, TB;
1687 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1688 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1689 SSEPackedDouble, Is2Addr>, TB, OpSize;
1692 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1693 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1694 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1695 SSEPackedSingle, 0>, TB;
1697 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1698 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1699 SSEPackedDouble, 0>, TB, OpSize;
1702 // Binary Arithmetic instructions
1703 let isAsmParserOnly = 1 in {
1704 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1705 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1706 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1707 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1708 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1709 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1710 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1711 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1713 let isCommutable = 0 in {
1714 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1715 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1716 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1717 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1718 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1719 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1720 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1721 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1722 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1723 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1724 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1725 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1726 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1727 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1728 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1729 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1730 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1731 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1732 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1733 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1737 let Constraints = "$src1 = $dst" in {
1738 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1739 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1740 basic_sse12_fp_binop_s_int<0x58, "add">;
1741 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1742 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1743 basic_sse12_fp_binop_s_int<0x59, "mul">;
1745 let isCommutable = 0 in {
1746 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1747 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1748 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1749 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1750 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1751 basic_sse12_fp_binop_s_int<0x5E, "div">;
1752 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1753 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1754 basic_sse12_fp_binop_s_int<0x5F, "max">,
1755 basic_sse12_fp_binop_p_int<0x5F, "max">;
1756 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1757 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1758 basic_sse12_fp_binop_s_int<0x5D, "min">,
1759 basic_sse12_fp_binop_p_int<0x5D, "min">;
1764 /// In addition, we also have a special variant of the scalar form here to
1765 /// represent the associated intrinsic operation. This form is unlike the
1766 /// plain scalar form, in that it takes an entire vector (instead of a
1767 /// scalar) and leaves the top elements undefined.
1769 /// And, we have a special variant form for a full-vector intrinsic form.
1771 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1772 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1773 SDNode OpNode, Intrinsic F32Int> {
1774 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1775 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1776 [(set FR32:$dst, (OpNode FR32:$src))]>;
1777 // For scalar unary operations, fold a load into the operation
1778 // only in OptForSize mode. It eliminates an instruction, but it also
1779 // eliminates a whole-register clobber (the load), so it introduces a
1780 // partial register update condition.
1781 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1782 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1783 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1784 Requires<[HasSSE1, OptForSize]>;
1785 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1786 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1787 [(set VR128:$dst, (F32Int VR128:$src))]>;
1788 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1789 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1790 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1793 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1794 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1795 SDNode OpNode, Intrinsic F32Int> {
1796 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1797 !strconcat(OpcodeStr,
1798 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1799 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1800 !strconcat(OpcodeStr,
1801 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1802 []>, XS, Requires<[HasAVX, OptForSize]>;
1803 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1804 !strconcat(OpcodeStr,
1805 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1806 [(set VR128:$dst, (F32Int VR128:$src))]>;
1807 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1808 !strconcat(OpcodeStr,
1809 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1810 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1813 /// sse1_fp_unop_p - SSE1 unops in packed form.
1814 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1815 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1816 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1817 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1818 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1819 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1820 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1823 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1824 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1825 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1826 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1827 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1828 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1829 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1830 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1833 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1834 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1835 Intrinsic V4F32Int> {
1836 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1837 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1838 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1839 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1840 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1841 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1844 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1845 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1846 Intrinsic V4F32Int> {
1847 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1848 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1849 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1850 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1851 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1852 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1855 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1856 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1857 SDNode OpNode, Intrinsic F64Int> {
1858 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1859 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1860 [(set FR64:$dst, (OpNode FR64:$src))]>;
1861 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1862 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1863 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1864 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1865 Requires<[HasSSE2, OptForSize]>;
1866 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1867 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1868 [(set VR128:$dst, (F64Int VR128:$src))]>;
1869 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1870 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1871 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1874 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1875 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1876 SDNode OpNode, Intrinsic F64Int> {
1877 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1878 !strconcat(OpcodeStr,
1879 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1880 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1881 (ins FR64:$src1, f64mem:$src2),
1882 !strconcat(OpcodeStr,
1883 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1884 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1885 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1886 [(set VR128:$dst, (F64Int VR128:$src))]>;
1887 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1888 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1889 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1892 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1893 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1895 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1896 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1897 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1898 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1899 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1900 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1903 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1904 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1905 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1906 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1907 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1908 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1909 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1910 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1913 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1914 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1915 Intrinsic V2F64Int> {
1916 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1917 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1918 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1919 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1920 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1921 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1924 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1925 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1926 Intrinsic V2F64Int> {
1927 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1928 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1929 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1930 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1931 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1932 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1935 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1937 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1938 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1941 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1942 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1943 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1944 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1945 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1946 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1947 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1948 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1951 // Reciprocal approximations. Note that these typically require refinement
1952 // in order to obtain suitable precision.
1953 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1954 int_x86_sse_rsqrt_ss>, VEX_4V;
1955 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1956 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1957 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1958 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1960 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1962 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1963 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1964 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1965 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1969 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1970 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1971 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1972 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1973 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1974 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1976 // Reciprocal approximations. Note that these typically require refinement
1977 // in order to obtain suitable precision.
1978 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1979 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1980 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1981 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1982 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1983 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1985 // There is no f64 version of the reciprocal approximation instructions.
1987 //===----------------------------------------------------------------------===//
1988 // SSE 1 & 2 - Non-temporal stores
1989 //===----------------------------------------------------------------------===//
1991 let isAsmParserOnly = 1 in {
1992 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1993 (ins i128mem:$dst, VR128:$src),
1994 "movntps\t{$src, $dst|$dst, $src}",
1995 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1996 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1997 (ins i128mem:$dst, VR128:$src),
1998 "movntpd\t{$src, $dst|$dst, $src}",
1999 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
2001 let ExeDomain = SSEPackedInt in
2002 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
2003 (ins f128mem:$dst, VR128:$src),
2004 "movntdq\t{$src, $dst|$dst, $src}",
2005 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
2007 let AddedComplexity = 400 in { // Prefer non-temporal versions
2008 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2009 (ins f128mem:$dst, VR128:$src),
2010 "movntps\t{$src, $dst|$dst, $src}",
2011 [(alignednontemporalstore (v4f32 VR128:$src),
2013 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2014 (ins f128mem:$dst, VR128:$src),
2015 "movntpd\t{$src, $dst|$dst, $src}",
2016 [(alignednontemporalstore (v2f64 VR128:$src),
2018 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2019 (ins f128mem:$dst, VR128:$src),
2020 "movntdq\t{$src, $dst|$dst, $src}",
2021 [(alignednontemporalstore (v2f64 VR128:$src),
2023 let ExeDomain = SSEPackedInt in
2024 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2025 (ins f128mem:$dst, VR128:$src),
2026 "movntdq\t{$src, $dst|$dst, $src}",
2027 [(alignednontemporalstore (v4f32 VR128:$src),
2030 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2031 (ins f256mem:$dst, VR256:$src),
2032 "movntps\t{$src, $dst|$dst, $src}",
2033 [(alignednontemporalstore (v8f32 VR256:$src),
2035 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2036 (ins f256mem:$dst, VR256:$src),
2037 "movntpd\t{$src, $dst|$dst, $src}",
2038 [(alignednontemporalstore (v4f64 VR256:$src),
2040 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2041 (ins f256mem:$dst, VR256:$src),
2042 "movntdq\t{$src, $dst|$dst, $src}",
2043 [(alignednontemporalstore (v4f64 VR256:$src),
2045 let ExeDomain = SSEPackedInt in
2046 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2047 (ins f256mem:$dst, VR256:$src),
2048 "movntdq\t{$src, $dst|$dst, $src}",
2049 [(alignednontemporalstore (v8f32 VR256:$src),
2054 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2055 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2056 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2057 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2058 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2059 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2061 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2062 "movntps\t{$src, $dst|$dst, $src}",
2063 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2064 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2065 "movntpd\t{$src, $dst|$dst, $src}",
2066 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2068 let ExeDomain = SSEPackedInt in
2069 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2070 "movntdq\t{$src, $dst|$dst, $src}",
2071 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2073 let AddedComplexity = 400 in { // Prefer non-temporal versions
2074 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2075 "movntps\t{$src, $dst|$dst, $src}",
2076 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2077 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2078 "movntpd\t{$src, $dst|$dst, $src}",
2079 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2081 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2082 "movntdq\t{$src, $dst|$dst, $src}",
2083 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2085 let ExeDomain = SSEPackedInt in
2086 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2087 "movntdq\t{$src, $dst|$dst, $src}",
2088 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2090 // There is no AVX form for instructions below this point
2091 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2092 "movnti\t{$src, $dst|$dst, $src}",
2093 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2094 TB, Requires<[HasSSE2]>;
2096 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2097 "movnti\t{$src, $dst|$dst, $src}",
2098 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2099 TB, Requires<[HasSSE2]>;
2102 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2103 "movnti\t{$src, $dst|$dst, $src}",
2104 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2105 TB, Requires<[HasSSE2]>;
2107 //===----------------------------------------------------------------------===//
2108 // SSE 1 & 2 - Misc Instructions (No AVX form)
2109 //===----------------------------------------------------------------------===//
2111 // Prefetch intrinsic.
2112 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2113 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2114 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2115 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2116 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2117 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2118 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2119 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2121 // Load, store, and memory fence
2122 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2123 TB, Requires<[HasSSE1]>;
2124 def : Pat<(X86SFence), (SFENCE)>;
2126 // Alias instructions that map zero vector to pxor / xorp* for sse.
2127 // We set canFoldAsLoad because this can be converted to a constant-pool
2128 // load of an all-zeros value if folding it would be beneficial.
2129 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2130 // JIT implementatioan, it does not expand the instructions below like
2131 // X86MCInstLower does.
2132 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2133 isCodeGenOnly = 1 in {
2134 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2135 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2136 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2137 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2138 let ExeDomain = SSEPackedInt in
2139 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2140 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2143 // The same as done above but for AVX. The 128-bit versions are the
2144 // same, but re-encoded. The 256-bit does not support PI version.
2145 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2146 // JIT implementatioan, it does not expand the instructions below like
2147 // X86MCInstLower does.
2148 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2149 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2150 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2151 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2152 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2153 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2154 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2155 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2156 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2157 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2158 let ExeDomain = SSEPackedInt in
2159 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2160 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2163 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2164 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2165 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2167 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2168 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2170 //===----------------------------------------------------------------------===//
2171 // SSE 1 & 2 - Load/Store XCSR register
2172 //===----------------------------------------------------------------------===//
2174 let isAsmParserOnly = 1 in {
2175 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2176 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2177 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2178 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2181 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2182 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2183 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2184 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2186 //===---------------------------------------------------------------------===//
2187 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2188 //===---------------------------------------------------------------------===//
2190 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2192 let isAsmParserOnly = 1 in {
2193 let neverHasSideEffects = 1 in {
2194 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2195 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2196 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2197 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2199 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2200 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2201 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2202 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2204 let canFoldAsLoad = 1, mayLoad = 1 in {
2205 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2206 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2207 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2208 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2209 let Predicates = [HasAVX] in {
2210 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2211 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2212 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2213 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2217 let mayStore = 1 in {
2218 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2219 (ins i128mem:$dst, VR128:$src),
2220 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2221 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2222 (ins i256mem:$dst, VR256:$src),
2223 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2224 let Predicates = [HasAVX] in {
2225 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2226 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2227 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2228 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2233 let neverHasSideEffects = 1 in
2234 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2235 "movdqa\t{$src, $dst|$dst, $src}", []>;
2237 let canFoldAsLoad = 1, mayLoad = 1 in {
2238 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2239 "movdqa\t{$src, $dst|$dst, $src}",
2240 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2241 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2242 "movdqu\t{$src, $dst|$dst, $src}",
2243 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2244 XS, Requires<[HasSSE2]>;
2247 let mayStore = 1 in {
2248 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2249 "movdqa\t{$src, $dst|$dst, $src}",
2250 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2251 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2252 "movdqu\t{$src, $dst|$dst, $src}",
2253 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2254 XS, Requires<[HasSSE2]>;
2257 // Intrinsic forms of MOVDQU load and store
2258 let isAsmParserOnly = 1 in {
2259 let canFoldAsLoad = 1 in
2260 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2261 "vmovdqu\t{$src, $dst|$dst, $src}",
2262 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2263 XS, VEX, Requires<[HasAVX]>;
2264 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2265 "vmovdqu\t{$src, $dst|$dst, $src}",
2266 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2267 XS, VEX, Requires<[HasAVX]>;
2270 let canFoldAsLoad = 1 in
2271 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2272 "movdqu\t{$src, $dst|$dst, $src}",
2273 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2274 XS, Requires<[HasSSE2]>;
2275 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2276 "movdqu\t{$src, $dst|$dst, $src}",
2277 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2278 XS, Requires<[HasSSE2]>;
2280 } // ExeDomain = SSEPackedInt
2282 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2283 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2284 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2286 //===---------------------------------------------------------------------===//
2287 // SSE2 - Packed Integer Arithmetic Instructions
2288 //===---------------------------------------------------------------------===//
2290 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2292 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2293 bit IsCommutable = 0, bit Is2Addr = 1> {
2294 let isCommutable = IsCommutable in
2295 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2296 (ins VR128:$src1, VR128:$src2),
2298 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2299 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2300 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2301 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2302 (ins VR128:$src1, i128mem:$src2),
2304 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2305 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2306 [(set VR128:$dst, (IntId VR128:$src1,
2307 (bitconvert (memopv2i64 addr:$src2))))]>;
2310 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2311 string OpcodeStr, Intrinsic IntId,
2312 Intrinsic IntId2, bit Is2Addr = 1> {
2313 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2314 (ins VR128:$src1, VR128:$src2),
2316 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2317 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2318 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2319 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2320 (ins VR128:$src1, i128mem:$src2),
2322 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2323 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2324 [(set VR128:$dst, (IntId VR128:$src1,
2325 (bitconvert (memopv2i64 addr:$src2))))]>;
2326 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2327 (ins VR128:$src1, i32i8imm:$src2),
2329 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2330 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2331 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2334 /// PDI_binop_rm - Simple SSE2 binary operator.
2335 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2336 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2337 let isCommutable = IsCommutable in
2338 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2339 (ins VR128:$src1, VR128:$src2),
2341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2343 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2344 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2345 (ins VR128:$src1, i128mem:$src2),
2347 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2349 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2350 (bitconvert (memopv2i64 addr:$src2)))))]>;
2353 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2355 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2356 /// to collapse (bitconvert VT to VT) into its operand.
2358 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2359 bit IsCommutable = 0, bit Is2Addr = 1> {
2360 let isCommutable = IsCommutable in
2361 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2362 (ins VR128:$src1, VR128:$src2),
2364 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2366 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2367 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2368 (ins VR128:$src1, i128mem:$src2),
2370 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2372 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2375 } // ExeDomain = SSEPackedInt
2377 // 128-bit Integer Arithmetic
2379 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2380 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2381 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2382 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2383 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2384 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2385 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2386 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2387 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2388 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2391 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2393 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2395 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2397 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2399 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2401 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2403 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2405 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2407 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2409 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2411 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2413 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2415 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2417 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2419 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2421 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2423 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2425 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2427 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2431 let Constraints = "$src1 = $dst" in {
2432 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2433 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2434 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2435 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2436 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2437 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2438 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2439 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2440 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2443 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2444 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2445 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2446 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2447 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2448 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2449 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2450 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2451 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2452 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2453 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2454 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2455 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2456 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2457 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2458 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2459 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2460 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2461 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2463 } // Constraints = "$src1 = $dst"
2465 //===---------------------------------------------------------------------===//
2466 // SSE2 - Packed Integer Logical Instructions
2467 //===---------------------------------------------------------------------===//
2469 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2470 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2471 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2473 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2474 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2476 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2477 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2480 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2481 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2483 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2484 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2486 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2487 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2490 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2491 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2493 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2494 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2497 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2498 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2499 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2501 let ExeDomain = SSEPackedInt in {
2502 let neverHasSideEffects = 1 in {
2503 // 128-bit logical shifts.
2504 def VPSLLDQri : PDIi8<0x73, MRM7r,
2505 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2506 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2508 def VPSRLDQri : PDIi8<0x73, MRM3r,
2509 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2510 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2512 // PSRADQri doesn't exist in SSE[1-3].
2514 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2515 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2516 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2517 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2518 VR128:$src2)))]>, VEX_4V;
2520 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2521 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2522 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2523 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2524 (memopv2i64 addr:$src2))))]>,
2529 let Constraints = "$src1 = $dst" in {
2530 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2531 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2532 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2533 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2534 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2535 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2537 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2538 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2539 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2540 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2541 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2542 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2544 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2545 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2546 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2547 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2549 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2550 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2551 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2553 let ExeDomain = SSEPackedInt in {
2554 let neverHasSideEffects = 1 in {
2555 // 128-bit logical shifts.
2556 def PSLLDQri : PDIi8<0x73, MRM7r,
2557 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2558 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2559 def PSRLDQri : PDIi8<0x73, MRM3r,
2560 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2561 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2562 // PSRADQri doesn't exist in SSE[1-3].
2564 def PANDNrr : PDI<0xDF, MRMSrcReg,
2565 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2566 "pandn\t{$src2, $dst|$dst, $src2}",
2567 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2570 def PANDNrm : PDI<0xDF, MRMSrcMem,
2571 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2572 "pandn\t{$src2, $dst|$dst, $src2}",
2573 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2574 (memopv2i64 addr:$src2))))]>;
2576 } // Constraints = "$src1 = $dst"
2578 let Predicates = [HasAVX] in {
2579 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2580 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2581 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2582 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2583 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2584 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2585 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2586 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2587 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2588 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2590 // Shift up / down and insert zero's.
2591 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2592 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2593 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2594 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2597 let Predicates = [HasSSE2] in {
2598 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2599 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2600 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2601 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2602 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2603 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2604 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2605 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2606 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2607 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2609 // Shift up / down and insert zero's.
2610 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2611 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2612 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2613 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2616 //===---------------------------------------------------------------------===//
2617 // SSE2 - Packed Integer Comparison Instructions
2618 //===---------------------------------------------------------------------===//
2620 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2621 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2623 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2625 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2627 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2629 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2631 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2635 let Constraints = "$src1 = $dst" in {
2636 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2637 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2638 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2639 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2640 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2641 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2642 } // Constraints = "$src1 = $dst"
2644 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2645 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2646 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2647 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2648 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2649 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2650 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2651 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2652 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2653 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2654 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2655 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2657 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2658 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2659 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2660 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2661 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2662 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2663 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2664 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2665 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2666 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2667 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2668 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2670 //===---------------------------------------------------------------------===//
2671 // SSE2 - Packed Integer Pack Instructions
2672 //===---------------------------------------------------------------------===//
2674 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2675 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2677 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2679 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2683 let Constraints = "$src1 = $dst" in {
2684 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2685 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2686 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2687 } // Constraints = "$src1 = $dst"
2689 //===---------------------------------------------------------------------===//
2690 // SSE2 - Packed Integer Shuffle Instructions
2691 //===---------------------------------------------------------------------===//
2693 let ExeDomain = SSEPackedInt in {
2694 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2696 def ri : Ii8<0x70, MRMSrcReg,
2697 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2698 !strconcat(OpcodeStr,
2699 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2700 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2702 def mi : Ii8<0x70, MRMSrcMem,
2703 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2704 !strconcat(OpcodeStr,
2705 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2706 [(set VR128:$dst, (vt (pshuf_frag:$src2
2707 (bc_frag (memopv2i64 addr:$src1)),
2710 } // ExeDomain = SSEPackedInt
2712 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2713 let AddedComplexity = 5 in
2714 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2717 // SSE2 with ImmT == Imm8 and XS prefix.
2718 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2721 // SSE2 with ImmT == Imm8 and XD prefix.
2722 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2726 let Predicates = [HasSSE2] in {
2727 let AddedComplexity = 5 in
2728 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2730 // SSE2 with ImmT == Imm8 and XS prefix.
2731 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2733 // SSE2 with ImmT == Imm8 and XD prefix.
2734 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2737 //===---------------------------------------------------------------------===//
2738 // SSE2 - Packed Integer Unpack Instructions
2739 //===---------------------------------------------------------------------===//
2741 let ExeDomain = SSEPackedInt in {
2742 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2743 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2744 def rr : PDI<opc, MRMSrcReg,
2745 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2747 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2748 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2749 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2750 def rm : PDI<opc, MRMSrcMem,
2751 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2753 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2754 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2755 [(set VR128:$dst, (unp_frag VR128:$src1,
2756 (bc_frag (memopv2i64
2760 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2761 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2763 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2765 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2768 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2769 /// knew to collapse (bitconvert VT to VT) into its operand.
2770 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2771 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2772 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2774 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2775 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2776 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2777 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2779 (v2i64 (unpckl VR128:$src1,
2780 (memopv2i64 addr:$src2))))]>, VEX_4V;
2782 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2784 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2786 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2789 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2790 /// knew to collapse (bitconvert VT to VT) into its operand.
2791 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2792 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2793 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2795 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2796 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2797 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2798 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2800 (v2i64 (unpckh VR128:$src1,
2801 (memopv2i64 addr:$src2))))]>, VEX_4V;
2804 let Constraints = "$src1 = $dst" in {
2805 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2806 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2807 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2809 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2810 /// knew to collapse (bitconvert VT to VT) into its operand.
2811 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2812 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2813 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2815 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2816 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2817 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2818 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2820 (v2i64 (unpckl VR128:$src1,
2821 (memopv2i64 addr:$src2))))]>;
2823 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2824 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2825 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2827 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2828 /// knew to collapse (bitconvert VT to VT) into its operand.
2829 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2830 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2831 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2833 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2834 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2835 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2836 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2838 (v2i64 (unpckh VR128:$src1,
2839 (memopv2i64 addr:$src2))))]>;
2842 } // ExeDomain = SSEPackedInt
2844 //===---------------------------------------------------------------------===//
2845 // SSE2 - Packed Integer Extract and Insert
2846 //===---------------------------------------------------------------------===//
2848 let ExeDomain = SSEPackedInt in {
2849 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2850 def rri : Ii8<0xC4, MRMSrcReg,
2851 (outs VR128:$dst), (ins VR128:$src1,
2852 GR32:$src2, i32i8imm:$src3),
2854 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2855 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2857 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2858 def rmi : Ii8<0xC4, MRMSrcMem,
2859 (outs VR128:$dst), (ins VR128:$src1,
2860 i16mem:$src2, i32i8imm:$src3),
2862 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2863 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2865 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2870 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2871 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2872 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2873 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2874 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2875 imm:$src2))]>, OpSize, VEX;
2876 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2877 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2878 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2879 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2883 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2884 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2885 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2886 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2887 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2888 []>, OpSize, VEX_4V;
2891 let Constraints = "$src1 = $dst" in
2892 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2894 } // ExeDomain = SSEPackedInt
2896 //===---------------------------------------------------------------------===//
2897 // SSE2 - Packed Mask Creation
2898 //===---------------------------------------------------------------------===//
2900 let ExeDomain = SSEPackedInt in {
2902 let isAsmParserOnly = 1 in {
2903 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2904 "pmovmskb\t{$src, $dst|$dst, $src}",
2905 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2906 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2907 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2909 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2910 "pmovmskb\t{$src, $dst|$dst, $src}",
2911 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2913 } // ExeDomain = SSEPackedInt
2915 //===---------------------------------------------------------------------===//
2916 // SSE2 - Conditional Store
2917 //===---------------------------------------------------------------------===//
2919 let ExeDomain = SSEPackedInt in {
2921 let isAsmParserOnly = 1 in {
2923 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2924 (ins VR128:$src, VR128:$mask),
2925 "maskmovdqu\t{$mask, $src|$src, $mask}",
2926 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2928 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2929 (ins VR128:$src, VR128:$mask),
2930 "maskmovdqu\t{$mask, $src|$src, $mask}",
2931 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2935 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2936 "maskmovdqu\t{$mask, $src|$src, $mask}",
2937 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2939 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2940 "maskmovdqu\t{$mask, $src|$src, $mask}",
2941 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2943 } // ExeDomain = SSEPackedInt
2945 //===---------------------------------------------------------------------===//
2946 // SSE2 - Move Doubleword
2947 //===---------------------------------------------------------------------===//
2949 // Move Int Doubleword to Packed Double Int
2950 let isAsmParserOnly = 1 in {
2951 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2954 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2955 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2956 "movd\t{$src, $dst|$dst, $src}",
2958 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2961 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2962 "movd\t{$src, $dst|$dst, $src}",
2964 (v4i32 (scalar_to_vector GR32:$src)))]>;
2965 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2966 "movd\t{$src, $dst|$dst, $src}",
2968 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2971 // Move Int Doubleword to Single Scalar
2972 let isAsmParserOnly = 1 in {
2973 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2974 "movd\t{$src, $dst|$dst, $src}",
2975 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2977 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2978 "movd\t{$src, $dst|$dst, $src}",
2979 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2982 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2983 "movd\t{$src, $dst|$dst, $src}",
2984 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2986 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2987 "movd\t{$src, $dst|$dst, $src}",
2988 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2990 // Move Packed Doubleword Int to Packed Double Int
2991 let isAsmParserOnly = 1 in {
2992 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2993 "movd\t{$src, $dst|$dst, $src}",
2994 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2996 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2997 (ins i32mem:$dst, VR128:$src),
2998 "movd\t{$src, $dst|$dst, $src}",
2999 [(store (i32 (vector_extract (v4i32 VR128:$src),
3000 (iPTR 0))), addr:$dst)]>, VEX;
3002 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3003 "movd\t{$src, $dst|$dst, $src}",
3004 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3006 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3007 "movd\t{$src, $dst|$dst, $src}",
3008 [(store (i32 (vector_extract (v4i32 VR128:$src),
3009 (iPTR 0))), addr:$dst)]>;
3011 // Move Scalar Single to Double Int
3012 let isAsmParserOnly = 1 in {
3013 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3014 "movd\t{$src, $dst|$dst, $src}",
3015 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3016 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3017 "movd\t{$src, $dst|$dst, $src}",
3018 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3020 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3021 "movd\t{$src, $dst|$dst, $src}",
3022 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3023 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3024 "movd\t{$src, $dst|$dst, $src}",
3025 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3027 // movd / movq to XMM register zero-extends
3028 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3029 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3030 "movd\t{$src, $dst|$dst, $src}",
3031 [(set VR128:$dst, (v4i32 (X86vzmovl
3032 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3034 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3035 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3036 [(set VR128:$dst, (v2i64 (X86vzmovl
3037 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3040 let AddedComplexity = 15 in {
3041 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3042 "movd\t{$src, $dst|$dst, $src}",
3043 [(set VR128:$dst, (v4i32 (X86vzmovl
3044 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3045 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3046 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3047 [(set VR128:$dst, (v2i64 (X86vzmovl
3048 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3051 let AddedComplexity = 20 in {
3052 let isAsmParserOnly = 1 in
3053 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3054 "movd\t{$src, $dst|$dst, $src}",
3056 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3057 (loadi32 addr:$src))))))]>,
3059 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3060 "movd\t{$src, $dst|$dst, $src}",
3062 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3063 (loadi32 addr:$src))))))]>;
3065 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3066 (MOVZDI2PDIrm addr:$src)>;
3067 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3068 (MOVZDI2PDIrm addr:$src)>;
3069 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3070 (MOVZDI2PDIrm addr:$src)>;
3073 //===---------------------------------------------------------------------===//
3074 // SSE2 - Move Quadword
3075 //===---------------------------------------------------------------------===//
3077 // Move Quadword Int to Packed Quadword Int
3078 let isAsmParserOnly = 1 in
3079 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3080 "vmovq\t{$src, $dst|$dst, $src}",
3082 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3083 VEX, Requires<[HasAVX]>;
3084 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3085 "movq\t{$src, $dst|$dst, $src}",
3087 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3088 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3090 // Move Packed Quadword Int to Quadword Int
3091 let isAsmParserOnly = 1 in
3092 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3093 "movq\t{$src, $dst|$dst, $src}",
3094 [(store (i64 (vector_extract (v2i64 VR128:$src),
3095 (iPTR 0))), addr:$dst)]>, VEX;
3096 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3097 "movq\t{$src, $dst|$dst, $src}",
3098 [(store (i64 (vector_extract (v2i64 VR128:$src),
3099 (iPTR 0))), addr:$dst)]>;
3101 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3102 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3104 // Store / copy lower 64-bits of a XMM register.
3105 let isAsmParserOnly = 1 in
3106 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3107 "movq\t{$src, $dst|$dst, $src}",
3108 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3109 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3110 "movq\t{$src, $dst|$dst, $src}",
3111 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3113 let AddedComplexity = 20, isAsmParserOnly = 1 in
3114 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3115 "vmovq\t{$src, $dst|$dst, $src}",
3117 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3118 (loadi64 addr:$src))))))]>,
3119 XS, VEX, Requires<[HasAVX]>;
3121 let AddedComplexity = 20 in {
3122 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3123 "movq\t{$src, $dst|$dst, $src}",
3125 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3126 (loadi64 addr:$src))))))]>,
3127 XS, Requires<[HasSSE2]>;
3129 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3130 (MOVZQI2PQIrm addr:$src)>;
3131 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3132 (MOVZQI2PQIrm addr:$src)>;
3133 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3136 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3137 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3138 let isAsmParserOnly = 1, AddedComplexity = 15 in
3139 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3140 "vmovq\t{$src, $dst|$dst, $src}",
3141 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3142 XS, VEX, Requires<[HasAVX]>;
3143 let AddedComplexity = 15 in
3144 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3145 "movq\t{$src, $dst|$dst, $src}",
3146 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3147 XS, Requires<[HasSSE2]>;
3149 let AddedComplexity = 20, isAsmParserOnly = 1 in
3150 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3151 "vmovq\t{$src, $dst|$dst, $src}",
3152 [(set VR128:$dst, (v2i64 (X86vzmovl
3153 (loadv2i64 addr:$src))))]>,
3154 XS, VEX, Requires<[HasAVX]>;
3155 let AddedComplexity = 20 in {
3156 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3157 "movq\t{$src, $dst|$dst, $src}",
3158 [(set VR128:$dst, (v2i64 (X86vzmovl
3159 (loadv2i64 addr:$src))))]>,
3160 XS, Requires<[HasSSE2]>;
3162 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3163 (MOVZPQILo2PQIrm addr:$src)>;
3166 // Instructions to match in the assembler
3167 let isAsmParserOnly = 1 in {
3168 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3169 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3170 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3171 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3172 // Recognize "movd" with GR64 destination, but encode as a "movq"
3173 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3174 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3177 // Instructions for the disassembler
3178 // xr = XMM register
3181 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3182 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3183 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3184 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3185 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3187 //===---------------------------------------------------------------------===//
3188 // SSE2 - Misc Instructions
3189 //===---------------------------------------------------------------------===//
3192 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3193 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3194 TB, Requires<[HasSSE2]>;
3196 // Load, store, and memory fence
3197 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3198 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3199 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3200 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3201 def : Pat<(X86LFence), (LFENCE)>;
3202 def : Pat<(X86MFence), (MFENCE)>;
3205 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3206 // was introduced with SSE2, it's backward compatible.
3207 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3209 // Alias instructions that map zero vector to pxor / xorp* for sse.
3210 // We set canFoldAsLoad because this can be converted to a constant-pool
3211 // load of an all-ones value if folding it would be beneficial.
3212 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3213 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3214 // FIXME: Change encoding to pseudo.
3215 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3216 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3218 //===---------------------------------------------------------------------===//
3219 // SSE3 - Conversion Instructions
3220 //===---------------------------------------------------------------------===//
3222 // Convert Packed Double FP to Packed DW Integers
3223 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3224 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3225 // register, but the same isn't true when using memory operands instead.
3226 // Provide other assembly rr and rm forms to address this explicitly.
3227 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3228 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3229 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3230 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3233 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3234 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3235 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3236 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3239 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3240 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3241 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3242 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3245 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3246 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3247 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3248 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3250 // Convert Packed DW Integers to Packed Double FP
3251 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3252 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3253 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3254 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3255 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3256 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3257 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3258 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3259 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3262 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3263 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3264 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3265 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3267 // AVX 256-bit register conversion intrinsics
3268 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3269 (VCVTDQ2PDYrr VR128:$src)>;
3270 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3271 (VCVTDQ2PDYrm addr:$src)>;
3273 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3274 (VCVTPD2DQYrr VR256:$src)>;
3275 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3276 (VCVTPD2DQYrm addr:$src)>;
3278 //===---------------------------------------------------------------------===//
3279 // SSE3 - Move Instructions
3280 //===---------------------------------------------------------------------===//
3282 // Replicate Single FP
3283 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3284 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3285 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3286 [(set VR128:$dst, (v4f32 (rep_frag
3287 VR128:$src, (undef))))]>;
3288 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3289 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3290 [(set VR128:$dst, (rep_frag
3291 (memopv4f32 addr:$src), (undef)))]>;
3294 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3296 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3297 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3298 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3299 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3302 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3303 // FIXME: Merge above classes when we have patterns for the ymm version
3304 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3305 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3306 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3307 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3309 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3310 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3312 // Replicate Double FP
3313 multiclass sse3_replicate_dfp<string OpcodeStr> {
3314 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3315 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3316 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3317 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3318 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3320 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3324 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3325 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3326 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3328 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3329 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3333 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3334 // FIXME: Merge above classes when we have patterns for the ymm version
3335 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3336 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3338 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3340 // Move Unaligned Integer
3341 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3342 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3343 "vlddqu\t{$src, $dst|$dst, $src}",
3344 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3345 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3346 "vlddqu\t{$src, $dst|$dst, $src}",
3347 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3349 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3350 "lddqu\t{$src, $dst|$dst, $src}",
3351 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3353 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3355 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3357 // Several Move patterns
3358 let AddedComplexity = 5 in {
3359 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3360 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3361 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3362 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3363 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3364 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3365 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3366 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3369 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3370 let AddedComplexity = 15 in
3371 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3372 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3373 let AddedComplexity = 20 in
3374 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3375 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3377 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3378 let AddedComplexity = 15 in
3379 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3380 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3381 let AddedComplexity = 20 in
3382 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3383 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3385 //===---------------------------------------------------------------------===//
3386 // SSE3 - Arithmetic
3387 //===---------------------------------------------------------------------===//
3389 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3390 X86MemOperand x86memop, bit Is2Addr = 1> {
3391 def rr : I<0xD0, MRMSrcReg,
3392 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3394 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3395 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3396 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3397 def rm : I<0xD0, MRMSrcMem,
3398 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3400 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3401 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3402 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3405 let isAsmParserOnly = 1, Predicates = [HasAVX],
3406 ExeDomain = SSEPackedDouble in {
3407 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3408 f128mem, 0>, XD, VEX_4V;
3409 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3410 f128mem, 0>, OpSize, VEX_4V;
3411 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3412 f256mem, 0>, XD, VEX_4V;
3413 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3414 f256mem, 0>, OpSize, VEX_4V;
3416 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3417 ExeDomain = SSEPackedDouble in {
3418 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3420 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3421 f128mem>, TB, OpSize;
3424 //===---------------------------------------------------------------------===//
3425 // SSE3 Instructions
3426 //===---------------------------------------------------------------------===//
3429 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3430 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3431 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3433 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3434 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3435 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3437 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3439 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3440 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3441 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3443 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3444 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3445 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3447 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3449 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3451 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3453 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3454 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3455 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3458 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3459 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3460 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3461 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3462 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3463 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3464 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3465 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3466 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3467 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3468 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3469 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3470 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3471 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3472 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3473 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3474 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3477 let Constraints = "$src1 = $dst" in {
3478 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3479 int_x86_sse3_hadd_ps>;
3480 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3481 int_x86_sse3_hadd_pd>;
3482 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3483 int_x86_sse3_hsub_ps>;
3484 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3485 int_x86_sse3_hsub_pd>;
3488 //===---------------------------------------------------------------------===//
3489 // SSSE3 - Packed Absolute Instructions
3490 //===---------------------------------------------------------------------===//
3493 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3494 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3495 PatFrag mem_frag128, Intrinsic IntId128> {
3496 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3498 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3499 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3502 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3504 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3507 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3510 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3511 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3512 int_x86_ssse3_pabs_b_128>, VEX;
3513 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3514 int_x86_ssse3_pabs_w_128>, VEX;
3515 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3516 int_x86_ssse3_pabs_d_128>, VEX;
3519 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3520 int_x86_ssse3_pabs_b_128>;
3521 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3522 int_x86_ssse3_pabs_w_128>;
3523 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3524 int_x86_ssse3_pabs_d_128>;
3526 //===---------------------------------------------------------------------===//
3527 // SSSE3 - Packed Binary Operator Instructions
3528 //===---------------------------------------------------------------------===//
3530 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3531 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3532 PatFrag mem_frag128, Intrinsic IntId128,
3534 let isCommutable = 1 in
3535 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3536 (ins VR128:$src1, VR128:$src2),
3538 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3539 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3540 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3542 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3543 (ins VR128:$src1, i128mem:$src2),
3545 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3546 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3548 (IntId128 VR128:$src1,
3549 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3552 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3553 let isCommutable = 0 in {
3554 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3555 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3556 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3557 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3558 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3559 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3560 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3561 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3562 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3563 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3564 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3565 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3566 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3567 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3568 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3569 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3570 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3571 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3572 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3573 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3574 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3575 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3577 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3578 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3581 // None of these have i8 immediate fields.
3582 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3583 let isCommutable = 0 in {
3584 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3585 int_x86_ssse3_phadd_w_128>;
3586 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3587 int_x86_ssse3_phadd_d_128>;
3588 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3589 int_x86_ssse3_phadd_sw_128>;
3590 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3591 int_x86_ssse3_phsub_w_128>;
3592 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3593 int_x86_ssse3_phsub_d_128>;
3594 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3595 int_x86_ssse3_phsub_sw_128>;
3596 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3597 int_x86_ssse3_pmadd_ub_sw_128>;
3598 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3599 int_x86_ssse3_pshuf_b_128>;
3600 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3601 int_x86_ssse3_psign_b_128>;
3602 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3603 int_x86_ssse3_psign_w_128>;
3604 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3605 int_x86_ssse3_psign_d_128>;
3607 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3608 int_x86_ssse3_pmul_hr_sw_128>;
3611 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3612 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3613 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3614 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3616 //===---------------------------------------------------------------------===//
3617 // SSSE3 - Packed Align Instruction Patterns
3618 //===---------------------------------------------------------------------===//
3620 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3621 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3622 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3624 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3626 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3628 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3629 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3631 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3633 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3637 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3638 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3639 let Constraints = "$src1 = $dst" in
3640 defm PALIGN : ssse3_palign<"palignr">;
3642 let AddedComplexity = 5 in {
3643 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3644 (PALIGNR128rr VR128:$src2, VR128:$src1,
3645 (SHUFFLE_get_palign_imm VR128:$src3))>,
3646 Requires<[HasSSSE3]>;
3647 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3648 (PALIGNR128rr VR128:$src2, VR128:$src1,
3649 (SHUFFLE_get_palign_imm VR128:$src3))>,
3650 Requires<[HasSSSE3]>;
3651 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3652 (PALIGNR128rr VR128:$src2, VR128:$src1,
3653 (SHUFFLE_get_palign_imm VR128:$src3))>,
3654 Requires<[HasSSSE3]>;
3655 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3656 (PALIGNR128rr VR128:$src2, VR128:$src1,
3657 (SHUFFLE_get_palign_imm VR128:$src3))>,
3658 Requires<[HasSSSE3]>;
3661 //===---------------------------------------------------------------------===//
3662 // SSSE3 Misc Instructions
3663 //===---------------------------------------------------------------------===//
3665 // Thread synchronization
3666 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3667 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3668 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3669 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3671 //===---------------------------------------------------------------------===//
3672 // Non-Instruction Patterns
3673 //===---------------------------------------------------------------------===//
3675 // extload f32 -> f64. This matches load+fextend because we have a hack in
3676 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3678 // Since these loads aren't folded into the fextend, we have to match it
3680 let Predicates = [HasSSE2] in
3681 def : Pat<(fextend (loadf32 addr:$src)),
3682 (CVTSS2SDrm addr:$src)>;
3685 let Predicates = [HasSSE2] in {
3686 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3687 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3688 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3689 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3690 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3691 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3692 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3693 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3694 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3695 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3696 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3697 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3698 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3699 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3700 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3701 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3702 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3703 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3704 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3705 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3706 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3707 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3708 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3709 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3710 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3711 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3712 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3713 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3714 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3715 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3718 // Move scalar to XMM zero-extended
3719 // movd to XMM register zero-extends
3720 let AddedComplexity = 15 in {
3721 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3722 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3723 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3724 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3725 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3726 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3727 (MOVSSrr (v4f32 (V_SET0PS)),
3728 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3729 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3730 (MOVSSrr (v4i32 (V_SET0PI)),
3731 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3734 // Splat v2f64 / v2i64
3735 let AddedComplexity = 10 in {
3736 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3737 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3738 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3739 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3740 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3741 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3742 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3743 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3746 // Special unary SHUFPSrri case.
3747 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3748 (SHUFPSrri VR128:$src1, VR128:$src1,
3749 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3750 let AddedComplexity = 5 in
3751 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3752 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3753 Requires<[HasSSE2]>;
3754 // Special unary SHUFPDrri case.
3755 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3756 (SHUFPDrri VR128:$src1, VR128:$src1,
3757 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3758 Requires<[HasSSE2]>;
3759 // Special unary SHUFPDrri case.
3760 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3761 (SHUFPDrri VR128:$src1, VR128:$src1,
3762 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3763 Requires<[HasSSE2]>;
3764 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3765 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3766 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3767 Requires<[HasSSE2]>;
3769 // Special binary v4i32 shuffle cases with SHUFPS.
3770 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3771 (SHUFPSrri VR128:$src1, VR128:$src2,
3772 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3773 Requires<[HasSSE2]>;
3774 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3775 (SHUFPSrmi VR128:$src1, addr:$src2,
3776 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3777 Requires<[HasSSE2]>;
3778 // Special binary v2i64 shuffle cases using SHUFPDrri.
3779 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3780 (SHUFPDrri VR128:$src1, VR128:$src2,
3781 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3782 Requires<[HasSSE2]>;
3784 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3785 let AddedComplexity = 15 in {
3786 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3787 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3788 Requires<[OptForSpeed, HasSSE2]>;
3789 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3790 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3791 Requires<[OptForSpeed, HasSSE2]>;
3793 let AddedComplexity = 10 in {
3794 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3795 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3796 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3797 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3798 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3799 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3800 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3801 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3804 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3805 let AddedComplexity = 15 in {
3806 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3807 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3808 Requires<[OptForSpeed, HasSSE2]>;
3809 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3810 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3811 Requires<[OptForSpeed, HasSSE2]>;
3813 let AddedComplexity = 10 in {
3814 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3815 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3816 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3817 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3818 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3819 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3820 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3821 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3824 let AddedComplexity = 20 in {
3825 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3826 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3827 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3829 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3830 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3831 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3833 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3834 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3835 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3836 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3837 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3840 let AddedComplexity = 20 in {
3841 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3842 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3843 (MOVLPSrm VR128:$src1, addr:$src2)>;
3844 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3845 (MOVLPDrm VR128:$src1, addr:$src2)>;
3846 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3847 (MOVLPSrm VR128:$src1, addr:$src2)>;
3848 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3849 (MOVLPDrm VR128:$src1, addr:$src2)>;
3852 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3853 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3854 (MOVLPSmr addr:$src1, VR128:$src2)>;
3855 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3856 (MOVLPDmr addr:$src1, VR128:$src2)>;
3857 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3859 (MOVLPSmr addr:$src1, VR128:$src2)>;
3860 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3861 (MOVLPDmr addr:$src1, VR128:$src2)>;
3863 let AddedComplexity = 15 in {
3864 // Setting the lowest element in the vector.
3865 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3866 (MOVSSrr (v4i32 VR128:$src1),
3867 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3868 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3869 (MOVSDrr (v2i64 VR128:$src1),
3870 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3872 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3873 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3874 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3875 Requires<[HasSSE2]>;
3876 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3877 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3878 Requires<[HasSSE2]>;
3881 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3882 // fall back to this for SSE1)
3883 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3884 (SHUFPSrri VR128:$src2, VR128:$src1,
3885 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3887 // Set lowest element and zero upper elements.
3888 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3889 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3891 // Some special case pandn patterns.
3892 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3894 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3895 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3897 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3898 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3900 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3902 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3903 (memop addr:$src2))),
3904 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3905 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3906 (memop addr:$src2))),
3907 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3908 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3909 (memop addr:$src2))),
3910 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3912 // vector -> vector casts
3913 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3914 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3915 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3916 (Int_CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3918 // Use movaps / movups for SSE integer load / store (one byte shorter).
3919 let Predicates = [HasSSE1] in {
3920 def : Pat<(alignedloadv4i32 addr:$src),
3921 (MOVAPSrm addr:$src)>;
3922 def : Pat<(loadv4i32 addr:$src),
3923 (MOVUPSrm addr:$src)>;
3924 def : Pat<(alignedloadv2i64 addr:$src),
3925 (MOVAPSrm addr:$src)>;
3926 def : Pat<(loadv2i64 addr:$src),
3927 (MOVUPSrm addr:$src)>;
3929 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3930 (MOVAPSmr addr:$dst, VR128:$src)>;
3931 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3932 (MOVAPSmr addr:$dst, VR128:$src)>;
3933 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3934 (MOVAPSmr addr:$dst, VR128:$src)>;
3935 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3936 (MOVAPSmr addr:$dst, VR128:$src)>;
3937 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3938 (MOVUPSmr addr:$dst, VR128:$src)>;
3939 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3940 (MOVUPSmr addr:$dst, VR128:$src)>;
3941 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3942 (MOVUPSmr addr:$dst, VR128:$src)>;
3943 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3944 (MOVUPSmr addr:$dst, VR128:$src)>;
3947 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3948 let Predicates = [HasAVX] in {
3949 def : Pat<(alignedloadv4i32 addr:$src),
3950 (VMOVAPSrm addr:$src)>;
3951 def : Pat<(loadv4i32 addr:$src),
3952 (VMOVUPSrm addr:$src)>;
3953 def : Pat<(alignedloadv2i64 addr:$src),
3954 (VMOVAPSrm addr:$src)>;
3955 def : Pat<(loadv2i64 addr:$src),
3956 (VMOVUPSrm addr:$src)>;
3958 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3959 (VMOVAPSmr addr:$dst, VR128:$src)>;
3960 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3961 (VMOVAPSmr addr:$dst, VR128:$src)>;
3962 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3963 (VMOVAPSmr addr:$dst, VR128:$src)>;
3964 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3965 (VMOVAPSmr addr:$dst, VR128:$src)>;
3966 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3967 (VMOVUPSmr addr:$dst, VR128:$src)>;
3968 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3969 (VMOVUPSmr addr:$dst, VR128:$src)>;
3970 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3971 (VMOVUPSmr addr:$dst, VR128:$src)>;
3972 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3973 (VMOVUPSmr addr:$dst, VR128:$src)>;
3976 //===----------------------------------------------------------------------===//
3977 // SSE4.1 - Packed Move with Sign/Zero Extend
3978 //===----------------------------------------------------------------------===//
3980 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3981 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3982 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3983 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3985 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3986 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3988 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3992 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3993 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3995 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3997 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3999 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4001 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4003 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4007 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4008 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4009 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4010 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4011 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4012 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4014 // Common patterns involving scalar load.
4015 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4016 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4017 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4018 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4020 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4021 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4022 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4023 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4025 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4026 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4027 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4028 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4030 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4031 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4032 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4033 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4035 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4036 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4037 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4038 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4040 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4041 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4042 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4043 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4046 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4047 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4048 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4049 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4051 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4052 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4054 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4058 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4059 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4061 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4063 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4065 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4069 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4070 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4071 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4072 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4074 // Common patterns involving scalar load
4075 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4076 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4077 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4078 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4080 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4081 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4082 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4083 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4086 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4087 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4088 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4089 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4091 // Expecting a i16 load any extended to i32 value.
4092 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4093 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4094 [(set VR128:$dst, (IntId (bitconvert
4095 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4099 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4100 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4102 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4105 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4106 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4108 // Common patterns involving scalar load
4109 def : Pat<(int_x86_sse41_pmovsxbq
4110 (bitconvert (v4i32 (X86vzmovl
4111 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4112 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4114 def : Pat<(int_x86_sse41_pmovzxbq
4115 (bitconvert (v4i32 (X86vzmovl
4116 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4117 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4119 //===----------------------------------------------------------------------===//
4120 // SSE4.1 - Extract Instructions
4121 //===----------------------------------------------------------------------===//
4123 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4124 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4125 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4126 (ins VR128:$src1, i32i8imm:$src2),
4127 !strconcat(OpcodeStr,
4128 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4129 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4131 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4132 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4133 !strconcat(OpcodeStr,
4134 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4137 // There's an AssertZext in the way of writing the store pattern
4138 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4141 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4142 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4143 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4144 (ins VR128:$src1, i32i8imm:$src2),
4145 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4148 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4151 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4152 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4153 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4154 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4155 !strconcat(OpcodeStr,
4156 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4159 // There's an AssertZext in the way of writing the store pattern
4160 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4163 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4164 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4166 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4169 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4170 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4171 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4172 (ins VR128:$src1, i32i8imm:$src2),
4173 !strconcat(OpcodeStr,
4174 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4176 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4177 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4178 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4179 !strconcat(OpcodeStr,
4180 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4181 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4182 addr:$dst)]>, OpSize;
4185 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4186 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4188 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4190 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4191 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4192 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4193 (ins VR128:$src1, i32i8imm:$src2),
4194 !strconcat(OpcodeStr,
4195 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4197 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4198 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4199 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4200 !strconcat(OpcodeStr,
4201 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4202 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4203 addr:$dst)]>, OpSize, REX_W;
4206 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4207 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4209 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4211 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4213 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4214 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4215 (ins VR128:$src1, i32i8imm:$src2),
4216 !strconcat(OpcodeStr,
4217 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4219 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4221 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4222 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4223 !strconcat(OpcodeStr,
4224 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4225 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4226 addr:$dst)]>, OpSize;
4229 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4230 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4231 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4232 (ins VR128:$src1, i32i8imm:$src2),
4233 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4236 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4238 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4239 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4242 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4243 Requires<[HasSSE41]>;
4245 //===----------------------------------------------------------------------===//
4246 // SSE4.1 - Insert Instructions
4247 //===----------------------------------------------------------------------===//
4249 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4250 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4251 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4253 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4255 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4257 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4258 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4259 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4261 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4263 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4265 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4266 imm:$src3))]>, OpSize;
4269 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4270 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4271 let Constraints = "$src1 = $dst" in
4272 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4274 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4275 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4276 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4278 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4280 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4282 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4284 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4285 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4287 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4289 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4291 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4292 imm:$src3)))]>, OpSize;
4295 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4296 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4297 let Constraints = "$src1 = $dst" in
4298 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4300 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4301 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4302 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4304 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4306 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4308 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4310 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4311 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4313 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4315 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4317 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4318 imm:$src3)))]>, OpSize;
4321 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4322 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4323 let Constraints = "$src1 = $dst" in
4324 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4326 // insertps has a few different modes, there's the first two here below which
4327 // are optimized inserts that won't zero arbitrary elements in the destination
4328 // vector. The next one matches the intrinsic and could zero arbitrary elements
4329 // in the target vector.
4330 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4331 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4332 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4334 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4336 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4338 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4340 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4341 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4343 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4345 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4347 (X86insrtps VR128:$src1,
4348 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4349 imm:$src3))]>, OpSize;
4352 let Constraints = "$src1 = $dst" in
4353 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4354 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4355 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4357 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4358 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4360 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4361 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4362 Requires<[HasSSE41]>;
4364 //===----------------------------------------------------------------------===//
4365 // SSE4.1 - Round Instructions
4366 //===----------------------------------------------------------------------===//
4368 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4369 X86MemOperand x86memop, RegisterClass RC,
4370 PatFrag mem_frag32, PatFrag mem_frag64,
4371 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4372 // Intrinsic operation, reg.
4373 // Vector intrinsic operation, reg
4374 def PSr_Int : SS4AIi8<opcps, MRMSrcReg,
4375 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4376 !strconcat(OpcodeStr,
4377 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4378 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4381 // Vector intrinsic operation, mem
4382 def PSm_Int : Ii8<opcps, MRMSrcMem,
4383 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4384 !strconcat(OpcodeStr,
4385 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4387 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4389 Requires<[HasSSE41]>;
4391 // Vector intrinsic operation, reg
4392 def PDr_Int : SS4AIi8<opcpd, MRMSrcReg,
4393 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4394 !strconcat(OpcodeStr,
4395 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4396 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4399 // Vector intrinsic operation, mem
4400 def PDm_Int : SS4AIi8<opcpd, MRMSrcMem,
4401 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4402 !strconcat(OpcodeStr,
4403 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4405 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4409 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4410 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4411 // Intrinsic operation, reg.
4412 // Vector intrinsic operation, reg
4413 def PSr : SS4AIi8<opcps, MRMSrcReg,
4414 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4415 !strconcat(OpcodeStr,
4416 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4419 // Vector intrinsic operation, mem
4420 def PSm : Ii8<opcps, MRMSrcMem,
4421 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4422 !strconcat(OpcodeStr,
4423 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4424 []>, TA, OpSize, Requires<[HasSSE41]>;
4426 // Vector intrinsic operation, reg
4427 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4428 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4429 !strconcat(OpcodeStr,
4430 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4433 // Vector intrinsic operation, mem
4434 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4435 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4436 !strconcat(OpcodeStr,
4437 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4441 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4444 Intrinsic F64Int, bit Is2Addr = 1> {
4445 // Intrinsic operation, reg.
4446 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
4447 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4449 !strconcat(OpcodeStr,
4450 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4451 !strconcat(OpcodeStr,
4452 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4453 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4456 // Intrinsic operation, mem.
4457 def SSm_Int : SS4AIi8<opcss, MRMSrcMem,
4458 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4460 !strconcat(OpcodeStr,
4461 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4462 !strconcat(OpcodeStr,
4463 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4465 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4468 // Intrinsic operation, reg.
4469 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
4470 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4472 !strconcat(OpcodeStr,
4473 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4474 !strconcat(OpcodeStr,
4475 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4476 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4479 // Intrinsic operation, mem.
4480 def SDm_Int : SS4AIi8<opcsd, MRMSrcMem,
4481 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4483 !strconcat(OpcodeStr,
4484 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4485 !strconcat(OpcodeStr,
4486 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4488 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4492 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4494 // Intrinsic operation, reg.
4495 def SSr : SS4AIi8<opcss, MRMSrcReg,
4496 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4497 !strconcat(OpcodeStr,
4498 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4501 // Intrinsic operation, mem.
4502 def SSm : SS4AIi8<opcss, MRMSrcMem,
4503 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4504 !strconcat(OpcodeStr,
4505 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4508 // Intrinsic operation, reg.
4509 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4510 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4511 !strconcat(OpcodeStr,
4512 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4515 // Intrinsic operation, mem.
4516 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4517 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4518 !strconcat(OpcodeStr,
4519 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4523 // FP round - roundss, roundps, roundsd, roundpd
4524 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4526 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4527 memopv4f32, memopv2f64,
4528 int_x86_sse41_round_ps,
4529 int_x86_sse41_round_pd>, VEX;
4530 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4531 memopv8f32, memopv4f64,
4532 int_x86_avx_round_ps_256,
4533 int_x86_avx_round_pd_256>, VEX;
4534 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4535 int_x86_sse41_round_ss,
4536 int_x86_sse41_round_sd, 0>, VEX_4V;
4538 // Instructions for the assembler
4539 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4541 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4543 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4546 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4547 memopv4f32, memopv2f64,
4548 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4549 let Constraints = "$src1 = $dst" in
4550 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4551 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4553 //===----------------------------------------------------------------------===//
4554 // SSE4.1 - Packed Bit Test
4555 //===----------------------------------------------------------------------===//
4557 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4558 // the intel intrinsic that corresponds to this.
4559 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4560 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4561 "vptest\t{$src2, $src1|$src1, $src2}",
4562 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4564 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4565 "vptest\t{$src2, $src1|$src1, $src2}",
4566 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4569 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4570 "vptest\t{$src2, $src1|$src1, $src2}",
4571 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4573 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4574 "vptest\t{$src2, $src1|$src1, $src2}",
4575 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4579 let Defs = [EFLAGS] in {
4580 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4581 "ptest \t{$src2, $src1|$src1, $src2}",
4582 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4584 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4585 "ptest \t{$src2, $src1|$src1, $src2}",
4586 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4590 // The bit test instructions below are AVX only
4591 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4592 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4593 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4594 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4595 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4596 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4597 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4598 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4602 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4603 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4604 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4605 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4606 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4609 //===----------------------------------------------------------------------===//
4610 // SSE4.1 - Misc Instructions
4611 //===----------------------------------------------------------------------===//
4613 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4614 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4615 Intrinsic IntId128> {
4616 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4618 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4619 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4620 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4622 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4625 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4628 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4629 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4630 int_x86_sse41_phminposuw>, VEX;
4631 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4632 int_x86_sse41_phminposuw>;
4634 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4635 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4636 Intrinsic IntId128, bit Is2Addr = 1> {
4637 let isCommutable = 1 in
4638 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4639 (ins VR128:$src1, VR128:$src2),
4641 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4642 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4643 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4644 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4645 (ins VR128:$src1, i128mem:$src2),
4647 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4648 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4650 (IntId128 VR128:$src1,
4651 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4654 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4655 let isCommutable = 0 in
4656 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4658 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4660 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4662 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4664 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4666 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4668 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4670 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4672 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4674 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4676 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4680 let Constraints = "$src1 = $dst" in {
4681 let isCommutable = 0 in
4682 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4683 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4684 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4685 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4686 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4687 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4688 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4689 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4690 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4691 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4692 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4695 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4696 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4697 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4698 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4700 /// SS48I_binop_rm - Simple SSE41 binary operator.
4701 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4702 ValueType OpVT, bit Is2Addr = 1> {
4703 let isCommutable = 1 in
4704 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4705 (ins VR128:$src1, VR128:$src2),
4707 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4708 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4709 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4711 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4712 (ins VR128:$src1, i128mem:$src2),
4714 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4715 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4716 [(set VR128:$dst, (OpNode VR128:$src1,
4717 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4721 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4722 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4723 let Constraints = "$src1 = $dst" in
4724 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4726 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4727 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4728 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4729 X86MemOperand x86memop, bit Is2Addr = 1> {
4730 let isCommutable = 1 in
4731 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4732 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4734 !strconcat(OpcodeStr,
4735 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4736 !strconcat(OpcodeStr,
4737 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4738 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4740 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4741 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4743 !strconcat(OpcodeStr,
4744 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4745 !strconcat(OpcodeStr,
4746 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4749 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4753 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4754 let isCommutable = 0 in {
4755 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4756 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4757 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4758 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4759 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4760 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4761 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4762 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4763 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4764 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4765 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4766 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4768 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4769 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4770 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4771 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4772 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4773 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4776 let Constraints = "$src1 = $dst" in {
4777 let isCommutable = 0 in {
4778 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4779 VR128, memopv16i8, i128mem>;
4780 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4781 VR128, memopv16i8, i128mem>;
4782 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4783 VR128, memopv16i8, i128mem>;
4784 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4785 VR128, memopv16i8, i128mem>;
4787 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4788 VR128, memopv16i8, i128mem>;
4789 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4790 VR128, memopv16i8, i128mem>;
4793 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4794 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4795 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4796 RegisterClass RC, X86MemOperand x86memop,
4797 PatFrag mem_frag, Intrinsic IntId> {
4798 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4799 (ins RC:$src1, RC:$src2, RC:$src3),
4800 !strconcat(OpcodeStr,
4801 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4802 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4803 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4805 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4806 (ins RC:$src1, x86memop:$src2, RC:$src3),
4807 !strconcat(OpcodeStr,
4808 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4810 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4812 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4816 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4817 memopv16i8, int_x86_sse41_blendvpd>;
4818 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4819 memopv16i8, int_x86_sse41_blendvps>;
4820 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4821 memopv16i8, int_x86_sse41_pblendvb>;
4822 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4823 memopv32i8, int_x86_avx_blendv_pd_256>;
4824 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4825 memopv32i8, int_x86_avx_blendv_ps_256>;
4827 /// SS41I_ternary_int - SSE 4.1 ternary operator
4828 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4829 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4830 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4831 (ins VR128:$src1, VR128:$src2),
4832 !strconcat(OpcodeStr,
4833 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4834 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4837 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4838 (ins VR128:$src1, i128mem:$src2),
4839 !strconcat(OpcodeStr,
4840 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4843 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4847 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4848 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4849 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4851 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4852 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4853 "vmovntdqa\t{$src, $dst|$dst, $src}",
4854 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4856 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4857 "movntdqa\t{$src, $dst|$dst, $src}",
4858 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4861 //===----------------------------------------------------------------------===//
4862 // SSE4.2 - Compare Instructions
4863 //===----------------------------------------------------------------------===//
4865 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4866 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4867 Intrinsic IntId128, bit Is2Addr = 1> {
4868 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4869 (ins VR128:$src1, VR128:$src2),
4871 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4872 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4873 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4875 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4876 (ins VR128:$src1, i128mem:$src2),
4878 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4879 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4881 (IntId128 VR128:$src1,
4882 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4885 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4886 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4888 let Constraints = "$src1 = $dst" in
4889 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4891 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4892 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4893 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4894 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4896 //===----------------------------------------------------------------------===//
4897 // SSE4.2 - String/text Processing Instructions
4898 //===----------------------------------------------------------------------===//
4900 // Packed Compare Implicit Length Strings, Return Mask
4901 multiclass pseudo_pcmpistrm<string asm> {
4902 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4903 (ins VR128:$src1, VR128:$src2, i8imm:$src3), !strconcat(asm, "rr PSEUDO"),
4904 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4906 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4907 (ins VR128:$src1, i128mem:$src2, i8imm:$src3), !strconcat(asm, "rm PSEUDO"),
4908 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4909 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4912 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4913 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4914 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4917 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4918 Predicates = [HasAVX] in {
4919 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4920 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4921 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4922 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4923 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4924 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4927 let Defs = [XMM0, EFLAGS] in {
4928 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4929 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4930 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4931 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4932 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4933 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4936 // Packed Compare Explicit Length Strings, Return Mask
4937 multiclass pseudo_pcmpestrm<string asm> {
4938 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4939 (ins VR128:$src1, VR128:$src3, i8imm:$src5), !strconcat(asm, "rr PSEUDO"),
4940 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4941 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4942 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4943 (ins VR128:$src1, i128mem:$src3, i8imm:$src5), !strconcat(asm, "rm PSEUDO"),
4944 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4945 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4948 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4949 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4950 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4953 let isAsmParserOnly = 1, Predicates = [HasAVX],
4954 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4955 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4956 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4957 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4958 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4959 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4960 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4963 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4964 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4965 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4966 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4967 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4968 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4969 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4972 // Packed Compare Implicit Length Strings, Return Index
4973 let Defs = [ECX, EFLAGS] in {
4974 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4975 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4976 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4977 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4978 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4979 (implicit EFLAGS)]>, OpSize;
4980 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4981 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4982 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4983 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4984 (implicit EFLAGS)]>, OpSize;
4988 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4989 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4991 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4993 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4995 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4997 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4999 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5003 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5004 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5005 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5006 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5007 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5008 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5010 // Packed Compare Explicit Length Strings, Return Index
5011 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5012 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5013 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5014 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5015 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5016 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5017 (implicit EFLAGS)]>, OpSize;
5018 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5019 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5020 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5022 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5023 (implicit EFLAGS)]>, OpSize;
5027 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5028 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5030 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5032 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5034 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5036 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5038 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5042 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5043 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5044 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5045 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5046 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5047 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5049 //===----------------------------------------------------------------------===//
5050 // SSE4.2 - CRC Instructions
5051 //===----------------------------------------------------------------------===//
5053 // No CRC instructions have AVX equivalents
5055 // crc intrinsic instruction
5056 // This set of instructions are only rm, the only difference is the size
5058 let Constraints = "$src1 = $dst" in {
5059 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5060 (ins GR32:$src1, i8mem:$src2),
5061 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5063 (int_x86_sse42_crc32_8 GR32:$src1,
5064 (load addr:$src2)))]>;
5065 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5066 (ins GR32:$src1, GR8:$src2),
5067 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5069 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5070 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5071 (ins GR32:$src1, i16mem:$src2),
5072 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5074 (int_x86_sse42_crc32_16 GR32:$src1,
5075 (load addr:$src2)))]>,
5077 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5078 (ins GR32:$src1, GR16:$src2),
5079 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5081 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5083 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5084 (ins GR32:$src1, i32mem:$src2),
5085 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5087 (int_x86_sse42_crc32_32 GR32:$src1,
5088 (load addr:$src2)))]>;
5089 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5090 (ins GR32:$src1, GR32:$src2),
5091 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5093 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5094 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5095 (ins GR64:$src1, i8mem:$src2),
5096 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5098 (int_x86_sse42_crc64_8 GR64:$src1,
5099 (load addr:$src2)))]>,
5101 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5102 (ins GR64:$src1, GR8:$src2),
5103 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5105 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5107 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5108 (ins GR64:$src1, i64mem:$src2),
5109 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5111 (int_x86_sse42_crc64_64 GR64:$src1,
5112 (load addr:$src2)))]>,
5114 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5115 (ins GR64:$src1, GR64:$src2),
5116 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5118 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5122 //===----------------------------------------------------------------------===//
5123 // AES-NI Instructions
5124 //===----------------------------------------------------------------------===//
5126 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5127 Intrinsic IntId128, bit Is2Addr = 1> {
5128 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5129 (ins VR128:$src1, VR128:$src2),
5131 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5132 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5133 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5135 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5136 (ins VR128:$src1, i128mem:$src2),
5138 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5139 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5141 (IntId128 VR128:$src1,
5142 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5145 // Perform One Round of an AES Encryption/Decryption Flow
5146 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5147 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5148 int_x86_aesni_aesenc, 0>, VEX_4V;
5149 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5150 int_x86_aesni_aesenclast, 0>, VEX_4V;
5151 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5152 int_x86_aesni_aesdec, 0>, VEX_4V;
5153 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5154 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5157 let Constraints = "$src1 = $dst" in {
5158 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5159 int_x86_aesni_aesenc>;
5160 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5161 int_x86_aesni_aesenclast>;
5162 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5163 int_x86_aesni_aesdec>;
5164 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5165 int_x86_aesni_aesdeclast>;
5168 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5169 (AESENCrr VR128:$src1, VR128:$src2)>;
5170 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5171 (AESENCrm VR128:$src1, addr:$src2)>;
5172 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5173 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5174 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5175 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5176 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5177 (AESDECrr VR128:$src1, VR128:$src2)>;
5178 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5179 (AESDECrm VR128:$src1, addr:$src2)>;
5180 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5181 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5182 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5183 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5185 // Perform the AES InvMixColumn Transformation
5186 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5187 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5189 "vaesimc\t{$src1, $dst|$dst, $src1}",
5191 (int_x86_aesni_aesimc VR128:$src1))]>,
5193 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5194 (ins i128mem:$src1),
5195 "vaesimc\t{$src1, $dst|$dst, $src1}",
5197 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5200 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5202 "aesimc\t{$src1, $dst|$dst, $src1}",
5204 (int_x86_aesni_aesimc VR128:$src1))]>,
5206 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5207 (ins i128mem:$src1),
5208 "aesimc\t{$src1, $dst|$dst, $src1}",
5210 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5213 // AES Round Key Generation Assist
5214 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5215 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5216 (ins VR128:$src1, i8imm:$src2),
5217 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5219 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5221 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5222 (ins i128mem:$src1, i8imm:$src2),
5223 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5225 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5229 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5230 (ins VR128:$src1, i8imm:$src2),
5231 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5233 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5235 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5236 (ins i128mem:$src1, i8imm:$src2),
5237 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5239 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5243 //===----------------------------------------------------------------------===//
5244 // CLMUL Instructions
5245 //===----------------------------------------------------------------------===//
5247 // Only the AVX version of CLMUL instructions are described here.
5249 // Carry-less Multiplication instructions
5250 let isAsmParserOnly = 1 in {
5251 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5252 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5253 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5256 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5257 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5258 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5262 multiclass avx_vpclmul<string asm> {
5263 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5264 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5267 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5268 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5271 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5272 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5273 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5274 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5276 } // isAsmParserOnly
5278 //===----------------------------------------------------------------------===//
5280 //===----------------------------------------------------------------------===//
5282 let isAsmParserOnly = 1 in {
5284 // Load from memory and broadcast to all elements of the destination operand
5285 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5286 X86MemOperand x86memop, Intrinsic Int> :
5287 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5288 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5289 [(set RC:$dst, (Int addr:$src))]>, VEX;
5291 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5292 int_x86_avx_vbroadcastss>;
5293 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5294 int_x86_avx_vbroadcastss_256>;
5295 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5296 int_x86_avx_vbroadcast_sd_256>;
5297 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5298 int_x86_avx_vbroadcastf128_pd_256>;
5300 // Insert packed floating-point values
5301 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5302 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5303 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5305 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5306 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5307 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5310 // Extract packed floating-point values
5311 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5312 (ins VR256:$src1, i8imm:$src2),
5313 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5315 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5316 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5317 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5320 // Conditional SIMD Packed Loads and Stores
5321 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5322 Intrinsic IntLd, Intrinsic IntLd256,
5323 Intrinsic IntSt, Intrinsic IntSt256,
5324 PatFrag pf128, PatFrag pf256> {
5325 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5326 (ins VR128:$src1, f128mem:$src2),
5327 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5328 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5330 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5331 (ins VR256:$src1, f256mem:$src2),
5332 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5333 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5335 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5336 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5337 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5338 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5339 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5340 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5341 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5342 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5345 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5346 int_x86_avx_maskload_ps,
5347 int_x86_avx_maskload_ps_256,
5348 int_x86_avx_maskstore_ps,
5349 int_x86_avx_maskstore_ps_256,
5350 memopv4f32, memopv8f32>;
5351 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5352 int_x86_avx_maskload_pd,
5353 int_x86_avx_maskload_pd_256,
5354 int_x86_avx_maskstore_pd,
5355 int_x86_avx_maskstore_pd_256,
5356 memopv2f64, memopv4f64>;
5358 // Permute Floating-Point Values
5359 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5360 RegisterClass RC, X86MemOperand x86memop_f,
5361 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5362 Intrinsic IntVar, Intrinsic IntImm> {
5363 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5364 (ins RC:$src1, RC:$src2),
5365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5366 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5367 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5368 (ins RC:$src1, x86memop_i:$src2),
5369 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5370 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5372 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5373 (ins RC:$src1, i8imm:$src2),
5374 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5375 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5376 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5377 (ins x86memop_f:$src1, i8imm:$src2),
5378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5379 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5382 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5383 memopv4f32, memopv4i32,
5384 int_x86_avx_vpermilvar_ps,
5385 int_x86_avx_vpermil_ps>;
5386 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5387 memopv8f32, memopv8i32,
5388 int_x86_avx_vpermilvar_ps_256,
5389 int_x86_avx_vpermil_ps_256>;
5390 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5391 memopv2f64, memopv2i64,
5392 int_x86_avx_vpermilvar_pd,
5393 int_x86_avx_vpermil_pd>;
5394 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5395 memopv4f64, memopv4i64,
5396 int_x86_avx_vpermilvar_pd_256,
5397 int_x86_avx_vpermil_pd_256>;
5399 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5400 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5401 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5403 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5404 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5405 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5408 // Zero All YMM registers
5409 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5410 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5412 // Zero Upper bits of YMM registers
5413 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5414 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5416 } // isAsmParserOnly
5418 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5419 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5420 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5421 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5422 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5423 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5425 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5426 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5427 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5428 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5429 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5430 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5432 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5433 (VBROADCASTF128 addr:$src)>;
5435 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5436 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5437 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5438 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5439 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5440 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5442 def : Pat<(int_x86_avx_vperm2f128_ps_256
5443 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5444 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5445 def : Pat<(int_x86_avx_vperm2f128_pd_256
5446 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5447 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5448 def : Pat<(int_x86_avx_vperm2f128_si_256
5449 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5450 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5452 //===----------------------------------------------------------------------===//
5453 // SSE Shuffle pattern fragments
5454 //===----------------------------------------------------------------------===//
5456 // This is part of a "work in progress" refactoring. The idea is that all
5457 // vector shuffles are going to be translated into target specific nodes and
5458 // directly matched by the patterns below (which can be changed along the way)
5459 // The AVX version of some but not all of them are described here, and more
5460 // should come in a near future.
5462 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5463 // SSE2 loads, which are always promoted to v2i64. The last one should match
5464 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5465 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5466 // we investigate further.
5467 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5469 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5470 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5472 (PSHUFDmi addr:$src1, imm:$imm)>;
5473 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5475 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5477 // Shuffle with PSHUFD instruction.
5478 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5479 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5480 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5481 (PSHUFDri VR128:$src1, imm:$imm)>;
5483 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5484 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5485 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5486 (PSHUFDri VR128:$src1, imm:$imm)>;
5488 // Shuffle with SHUFPD instruction.
5489 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5490 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5491 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5492 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5493 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5494 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5496 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5497 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5498 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5499 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5501 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5502 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5503 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5504 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5506 // Shuffle with SHUFPS instruction.
5507 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5508 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5509 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5510 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5511 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5512 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5514 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5515 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5516 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5517 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5519 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5520 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5521 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5522 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5523 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5524 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5526 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5527 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5528 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5529 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5531 // Shuffle with MOVHLPS instruction
5532 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5533 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5534 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5535 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5537 // Shuffle with MOVDDUP instruction
5538 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5539 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5540 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5541 (MOVDDUPrm addr:$src)>;
5543 def : Pat<(X86Movddup (bc_v4f32 (memopv2f64 addr:$src))),
5544 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5545 def : Pat<(X86Movddup (bc_v4f32 (memopv2f64 addr:$src))),
5546 (MOVDDUPrm addr:$src)>;
5548 def : Pat<(X86Movddup (memopv2i64 addr:$src)),
5549 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5550 def : Pat<(X86Movddup (memopv2i64 addr:$src)),
5551 (MOVDDUPrm addr:$src)>;
5553 def : Pat<(X86Movddup (bc_v4i32 (memopv2i64 addr:$src))),
5554 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5555 def : Pat<(X86Movddup (bc_v4i32 (memopv2i64 addr:$src))),
5556 (MOVDDUPrm addr:$src)>;
5558 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5559 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5560 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5561 (MOVDDUPrm addr:$src)>;
5563 def : Pat<(X86Movddup (bc_v2f64
5564 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5565 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5566 def : Pat<(X86Movddup (bc_v2f64
5567 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5568 (MOVDDUPrm addr:$src)>;
5570 // Shuffle with UNPCKLPS
5571 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5572 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5573 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5574 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5576 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5577 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5578 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5579 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5581 // Shuffle with UNPCKHPS
5582 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5583 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5584 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5585 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5587 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5588 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5589 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5590 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5592 // Shuffle with UNPCKLPD
5593 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5594 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5595 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5596 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5598 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5599 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5600 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5601 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5603 // Shuffle with UNPCKHPD
5604 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5605 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5606 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5607 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5609 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5610 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5611 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5612 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5614 // Shuffle with PUNPCKLBW
5615 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5616 (bc_v16i8 (memopv2i64 addr:$src2)))),
5617 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5618 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5619 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5621 // Shuffle with PUNPCKLWD
5622 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5623 (bc_v8i16 (memopv2i64 addr:$src2)))),
5624 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5625 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5626 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5628 // Shuffle with PUNPCKLDQ
5629 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5630 (bc_v4i32 (memopv2i64 addr:$src2)))),
5631 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5632 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5633 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5635 // Shuffle with PUNPCKLQDQ
5636 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5637 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5638 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5639 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5641 // Shuffle with PUNPCKHBW
5642 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5643 (bc_v16i8 (memopv2i64 addr:$src2)))),
5644 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5645 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5646 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5648 // Shuffle with PUNPCKHWD
5649 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5650 (bc_v8i16 (memopv2i64 addr:$src2)))),
5651 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5652 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5653 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5655 // Shuffle with PUNPCKHDQ
5656 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5657 (bc_v4i32 (memopv2i64 addr:$src2)))),
5658 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5659 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5660 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5662 // Shuffle with PUNPCKHQDQ
5663 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5664 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5665 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5666 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5668 // Shuffle with MOVLHPS
5669 def : Pat<(X86Movlhps VR128:$src1,
5670 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5671 (MOVHPSrm VR128:$src1, addr:$src2)>;
5672 def : Pat<(X86Movlhps VR128:$src1,
5673 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5674 (MOVHPSrm VR128:$src1, addr:$src2)>;
5675 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5676 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5677 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5678 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5679 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5680 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5681 // FIXME: Instead of X86Movddup, there should be a X86Movlhps here, the problem
5682 // is during lowering, where it's not possible to recognize the load fold cause
5683 // it has two uses through a bitcast. One use disappears at isel time and the
5684 // fold opportunity reappears.
5685 def : Pat<(v2i64 (X86Movddup VR128:$src)),
5686 (MOVLHPSrr VR128:$src, VR128:$src)>;
5687 def : Pat<(v4f32 (X86Movddup VR128:$src)),
5688 (MOVLHPSrr VR128:$src, VR128:$src)>;
5689 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5690 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5692 // Shuffle with MOVLHPD
5693 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5694 (scalar_to_vector (loadf64 addr:$src2)))),
5695 (MOVHPDrm VR128:$src1, addr:$src2)>;
5696 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5697 // is during lowering, where it's not possible to recognize the load fold cause
5698 // it has two uses through a bitcast. One use disappears at isel time and the
5699 // fold opportunity reappears.
5700 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5701 (scalar_to_vector (loadf64 addr:$src2)))),
5702 (MOVHPDrm VR128:$src1, addr:$src2)>;
5704 // Shuffle with MOVSS
5705 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5706 (MOVSSrr VR128:$src1, FR32:$src2)>;
5707 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5708 (MOVSSrr (v4i32 VR128:$src1),
5709 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5710 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5711 (MOVSSrr (v4f32 VR128:$src1),
5712 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5713 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5714 // is during lowering, where it's not possible to recognize the load fold cause
5715 // it has two uses through a bitcast. One use disappears at isel time and the
5716 // fold opportunity reappears.
5717 def : Pat<(X86Movss VR128:$src1,
5718 (bc_v4i32 (v2i64 (load addr:$src2)))),
5719 (MOVLPSrm VR128:$src1, addr:$src2)>;
5721 // Shuffle with MOVSD
5722 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5723 (MOVSDrr VR128:$src1, FR64:$src2)>;
5724 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5725 (MOVSDrr (v2i64 VR128:$src1),
5726 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5727 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5728 (MOVSDrr (v2f64 VR128:$src1),
5729 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5730 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5731 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5732 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5733 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5735 // Shuffle with MOVSHDUP
5736 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5737 (MOVSHDUPrr VR128:$src)>;
5738 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5739 (MOVSHDUPrm addr:$src)>;
5741 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5742 (MOVSHDUPrr VR128:$src)>;
5743 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5744 (MOVSHDUPrm addr:$src)>;
5746 // Shuffle with MOVSLDUP
5747 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5748 (MOVSLDUPrr VR128:$src)>;
5749 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5750 (MOVSLDUPrm addr:$src)>;
5752 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5753 (MOVSLDUPrr VR128:$src)>;
5754 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5755 (MOVSLDUPrm addr:$src)>;
5757 // Shuffle with PSHUFHW
5758 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5759 (PSHUFHWri VR128:$src, imm:$imm)>;
5760 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5761 (PSHUFHWmi addr:$src, imm:$imm)>;
5763 // Shuffle with PSHUFLW
5764 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5765 (PSHUFLWri VR128:$src, imm:$imm)>;
5766 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5767 (PSHUFLWmi addr:$src, imm:$imm)>;
5769 // Shuffle with PALIGN
5770 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5771 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5772 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5773 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5774 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5775 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5776 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5777 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5779 // Shuffle with MOVLPS
5780 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5781 (MOVLPSrm VR128:$src1, addr:$src2)>;
5782 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5783 (MOVLPSrm VR128:$src1, addr:$src2)>;
5784 def : Pat<(X86Movlps VR128:$src1,
5785 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5786 (MOVLPSrm VR128:$src1, addr:$src2)>;
5788 // Shuffle with MOVLPD
5789 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5790 (MOVLPDrm VR128:$src1, addr:$src2)>;
5791 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5792 (MOVLPDrm VR128:$src1, addr:$src2)>;
5793 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5794 (scalar_to_vector (loadf64 addr:$src2)))),
5795 (MOVLPDrm VR128:$src1, addr:$src2)>;
5797 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5798 def : Pat<(store (f64 (vector_extract
5799 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5800 (MOVHPSmr addr:$dst, VR128:$src)>;
5801 def : Pat<(store (f64 (vector_extract
5802 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5803 (MOVHPDmr addr:$dst, VR128:$src)>;
5805 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5806 (MOVLPSmr addr:$src1, VR128:$src2)>;
5807 def : Pat<(store (v4i32 (X86Movlps
5808 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5809 (MOVLPSmr addr:$src1, VR128:$src2)>;
5811 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5812 (MOVLPDmr addr:$src1, VR128:$src2)>;
5813 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5814 (MOVLPDmr addr:$src1, VR128:$src2)>;