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, $src}">, XD, VEX, VEX_W;
650 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
651 f32mem, load, "cvtss2si">, XS;
652 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
653 f32mem, load, "cvtss2si{q}">, XS, REX_W;
654 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
655 f128mem, load, "cvtsd2si{l}">, XD;
656 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
657 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
660 let isAsmParserOnly = 1 in {
661 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
662 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
663 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
664 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
666 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
667 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
668 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
669 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
673 let Constraints = "$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, $src}",
935 (int_x86_sse2_cvttps2dq VR128:$src))]>;
936 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
937 "cvttps2dq\t{$src, $dst|$dst, $src}",
939 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
942 let isAsmParserOnly = 1 in {
943 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
944 "vcvttps2dq\t{$src, $dst|$dst, $src}",
946 (int_x86_sse2_cvttps2dq VR128:$src))]>,
947 XS, VEX, Requires<[HasAVX]>;
948 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
949 "vcvttps2dq\t{$src, $dst|$dst, $src}",
950 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
951 (memop addr:$src)))]>,
952 XS, VEX, Requires<[HasAVX]>;
955 let isAsmParserOnly = 1 in {
956 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
958 "cvttpd2dq\t{$src, $dst|$dst, $src}",
959 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
961 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
963 "cvttpd2dq\t{$src, $dst|$dst, $src}",
964 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
965 (memop addr:$src)))]>, VEX;
967 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
968 "cvttpd2dq\t{$src, $dst|$dst, $src}",
969 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
970 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
971 "cvttpd2dq\t{$src, $dst|$dst, $src}",
972 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
973 (memop addr:$src)))]>;
975 let isAsmParserOnly = 1 in {
976 // The assembler can recognize rr 256-bit instructions by seeing a ymm
977 // register, but the same isn't true when using memory operands instead.
978 // Provide other assembly rr and rm forms to address this explicitly.
979 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
980 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
981 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
982 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
985 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
986 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
987 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
988 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
991 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
992 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
993 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
994 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
997 // Convert packed single to packed double
998 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
999 // SSE2 instructions without OpSize prefix
1000 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1001 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1002 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1003 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1004 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1005 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1006 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1007 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1009 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1010 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1011 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1012 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1014 let isAsmParserOnly = 1 in {
1015 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1016 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1017 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1018 VEX, Requires<[HasAVX]>;
1019 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1020 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1021 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1022 (load addr:$src)))]>,
1023 VEX, Requires<[HasAVX]>;
1025 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1026 "cvtps2pd\t{$src, $dst|$dst, $src}",
1027 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1028 TB, Requires<[HasSSE2]>;
1029 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1030 "cvtps2pd\t{$src, $dst|$dst, $src}",
1031 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1032 (load addr:$src)))]>,
1033 TB, Requires<[HasSSE2]>;
1035 // Convert packed double to packed single
1036 let isAsmParserOnly = 1 in {
1037 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1038 // register, but the same isn't true when using memory operands instead.
1039 // Provide other assembly rr and rm forms to address this explicitly.
1040 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1041 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1042 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1043 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1046 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1047 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1048 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1049 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1052 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1053 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1054 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1055 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1057 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1058 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1059 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1060 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1063 let isAsmParserOnly = 1 in {
1064 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1065 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1066 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1067 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1069 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1070 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1071 (memop addr:$src)))]>;
1073 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1074 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1075 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1076 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1077 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1078 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1079 (memop addr:$src)))]>;
1081 // AVX 256-bit register conversion intrinsics
1082 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1083 // whenever possible to avoid declaring two versions of each one.
1084 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1085 (VCVTDQ2PSYrr VR256:$src)>;
1086 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1087 (VCVTDQ2PSYrm addr:$src)>;
1089 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1090 (VCVTPD2PSYrr VR256:$src)>;
1091 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1092 (VCVTPD2PSYrm addr:$src)>;
1094 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1095 (VCVTPS2DQYrr VR256:$src)>;
1096 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1097 (VCVTPS2DQYrm addr:$src)>;
1099 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1100 (VCVTPS2PDYrr VR128:$src)>;
1101 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1102 (VCVTPS2PDYrm addr:$src)>;
1104 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1105 (VCVTTPD2DQYrr VR256:$src)>;
1106 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1107 (VCVTTPD2DQYrm addr:$src)>;
1109 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1110 (VCVTTPS2DQYrr VR256:$src)>;
1111 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1112 (VCVTTPS2DQYrm addr:$src)>;
1114 //===----------------------------------------------------------------------===//
1115 // SSE 1 & 2 - Compare Instructions
1116 //===----------------------------------------------------------------------===//
1118 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1119 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1120 string asm, string asm_alt> {
1121 def rr : SIi8<0xC2, MRMSrcReg,
1122 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1125 def rm : SIi8<0xC2, MRMSrcMem,
1126 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1128 // Accept explicit immediate argument form instead of comparison code.
1129 let isAsmParserOnly = 1 in {
1130 def rr_alt : SIi8<0xC2, MRMSrcReg,
1131 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1134 def rm_alt : SIi8<0xC2, MRMSrcMem,
1135 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1140 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1141 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1142 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1143 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1145 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1146 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1147 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1151 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1152 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1153 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1154 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1155 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1156 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1157 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1160 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1161 Intrinsic Int, string asm> {
1162 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1163 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1164 [(set VR128:$dst, (Int VR128:$src1,
1165 VR128:$src, imm:$cc))]>;
1166 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1167 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1168 [(set VR128:$dst, (Int VR128:$src1,
1169 (load addr:$src), imm:$cc))]>;
1172 // Aliases to match intrinsics which expect XMM operand(s).
1173 let isAsmParserOnly = 1 in {
1174 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1175 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1177 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1178 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1181 let Constraints = "$src1 = $dst" in {
1182 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1183 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1184 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1185 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1189 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1190 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1191 ValueType vt, X86MemOperand x86memop,
1192 PatFrag ld_frag, string OpcodeStr, Domain d> {
1193 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1194 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1195 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1196 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1197 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1198 [(set EFLAGS, (OpNode (vt RC:$src1),
1199 (ld_frag addr:$src2)))], d>;
1202 let Defs = [EFLAGS] in {
1203 let isAsmParserOnly = 1 in {
1204 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1205 "ucomiss", SSEPackedSingle>, VEX;
1206 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1207 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1208 let Pattern = []<dag> in {
1209 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1210 "comiss", SSEPackedSingle>, VEX;
1211 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1212 "comisd", SSEPackedDouble>, OpSize, VEX;
1215 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1216 load, "ucomiss", SSEPackedSingle>, VEX;
1217 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1218 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1220 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1221 load, "comiss", SSEPackedSingle>, VEX;
1222 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1223 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1225 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1226 "ucomiss", SSEPackedSingle>, TB;
1227 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1228 "ucomisd", SSEPackedDouble>, TB, OpSize;
1230 let Pattern = []<dag> in {
1231 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1232 "comiss", SSEPackedSingle>, TB;
1233 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1234 "comisd", SSEPackedDouble>, TB, OpSize;
1237 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1238 load, "ucomiss", SSEPackedSingle>, TB;
1239 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1240 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1242 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1243 "comiss", SSEPackedSingle>, TB;
1244 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1245 "comisd", SSEPackedDouble>, TB, OpSize;
1246 } // Defs = [EFLAGS]
1248 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1249 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1250 Intrinsic Int, string asm, string asm_alt,
1252 def rri : PIi8<0xC2, MRMSrcReg,
1253 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1254 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1255 def rmi : PIi8<0xC2, MRMSrcMem,
1256 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1257 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1258 // Accept explicit immediate argument form instead of comparison code.
1259 let isAsmParserOnly = 1 in {
1260 def rri_alt : PIi8<0xC2, MRMSrcReg,
1261 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1263 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1264 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1269 let isAsmParserOnly = 1 in {
1270 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1271 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1272 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1273 SSEPackedSingle>, VEX_4V;
1274 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1275 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1276 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1277 SSEPackedDouble>, OpSize, VEX_4V;
1278 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1279 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1280 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1281 SSEPackedSingle>, VEX_4V;
1282 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1283 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1284 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1285 SSEPackedDouble>, OpSize, VEX_4V;
1287 let Constraints = "$src1 = $dst" in {
1288 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1289 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1290 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1291 SSEPackedSingle>, TB;
1292 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1293 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1294 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1295 SSEPackedDouble>, TB, OpSize;
1298 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1299 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1300 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1301 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1302 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1303 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1304 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1305 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1307 //===----------------------------------------------------------------------===//
1308 // SSE 1 & 2 - Shuffle Instructions
1309 //===----------------------------------------------------------------------===//
1311 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1312 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1313 ValueType vt, string asm, PatFrag mem_frag,
1314 Domain d, bit IsConvertibleToThreeAddress = 0> {
1315 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1316 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1317 [(set RC:$dst, (vt (shufp:$src3
1318 RC:$src1, (mem_frag addr:$src2))))], d>;
1319 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1320 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1321 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1323 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1326 let isAsmParserOnly = 1 in {
1327 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1328 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1329 memopv4f32, SSEPackedSingle>, VEX_4V;
1330 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1331 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1332 memopv8f32, SSEPackedSingle>, VEX_4V;
1333 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1334 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1335 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1336 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1337 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1338 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1341 let Constraints = "$src1 = $dst" in {
1342 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1343 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1344 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1346 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1347 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1348 memopv2f64, SSEPackedDouble>, TB, OpSize;
1351 //===----------------------------------------------------------------------===//
1352 // SSE 1 & 2 - Unpack Instructions
1353 //===----------------------------------------------------------------------===//
1355 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1356 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1357 PatFrag mem_frag, RegisterClass RC,
1358 X86MemOperand x86memop, string asm,
1360 def rr : PI<opc, MRMSrcReg,
1361 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1363 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1364 def rm : PI<opc, MRMSrcMem,
1365 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1367 (vt (OpNode RC:$src1,
1368 (mem_frag addr:$src2))))], d>;
1371 let AddedComplexity = 10 in {
1372 let isAsmParserOnly = 1 in {
1373 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1374 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1375 SSEPackedSingle>, VEX_4V;
1376 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1377 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1378 SSEPackedDouble>, OpSize, VEX_4V;
1379 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1380 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1381 SSEPackedSingle>, VEX_4V;
1382 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1383 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1384 SSEPackedDouble>, OpSize, VEX_4V;
1386 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1387 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1388 SSEPackedSingle>, VEX_4V;
1389 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1390 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1391 SSEPackedDouble>, OpSize, VEX_4V;
1392 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1393 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1394 SSEPackedSingle>, VEX_4V;
1395 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1396 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1397 SSEPackedDouble>, OpSize, VEX_4V;
1400 let Constraints = "$src1 = $dst" in {
1401 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1402 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1403 SSEPackedSingle>, TB;
1404 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1405 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1406 SSEPackedDouble>, TB, OpSize;
1407 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1408 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1409 SSEPackedSingle>, TB;
1410 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1411 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1412 SSEPackedDouble>, TB, OpSize;
1413 } // Constraints = "$src1 = $dst"
1414 } // AddedComplexity
1416 //===----------------------------------------------------------------------===//
1417 // SSE 1 & 2 - Extract Floating-Point Sign mask
1418 //===----------------------------------------------------------------------===//
1420 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1421 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1423 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1424 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1425 [(set GR32:$dst, (Int RC:$src))], d>;
1426 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1427 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1431 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1432 SSEPackedSingle>, TB;
1433 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1434 SSEPackedDouble>, TB, OpSize;
1436 let isAsmParserOnly = 1 in {
1437 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1438 "movmskps", SSEPackedSingle>, VEX;
1439 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1440 "movmskpd", SSEPackedDouble>, OpSize,
1442 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1443 "movmskps", SSEPackedSingle>, VEX;
1444 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1445 "movmskpd", SSEPackedDouble>, OpSize,
1449 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1450 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1451 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1452 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1454 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1455 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1456 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1457 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1461 //===----------------------------------------------------------------------===//
1462 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1463 //===----------------------------------------------------------------------===//
1465 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1466 // names that start with 'Fs'.
1468 // Alias instructions that map fld0 to pxor for sse.
1469 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1470 canFoldAsLoad = 1 in {
1471 // FIXME: Set encoding to pseudo!
1472 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1473 [(set FR32:$dst, fp32imm0)]>,
1474 Requires<[HasSSE1]>, TB, OpSize;
1475 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1476 [(set FR64:$dst, fpimm0)]>,
1477 Requires<[HasSSE2]>, TB, OpSize;
1480 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1481 // bits are disregarded.
1482 let neverHasSideEffects = 1 in {
1483 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1484 "movaps\t{$src, $dst|$dst, $src}", []>;
1485 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1486 "movapd\t{$src, $dst|$dst, $src}", []>;
1489 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1490 // bits are disregarded.
1491 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1492 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1493 "movaps\t{$src, $dst|$dst, $src}",
1494 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1495 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1496 "movapd\t{$src, $dst|$dst, $src}",
1497 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1500 //===----------------------------------------------------------------------===//
1501 // SSE 1 & 2 - Logical Instructions
1502 //===----------------------------------------------------------------------===//
1504 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1506 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1508 let isAsmParserOnly = 1 in {
1509 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1510 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1512 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1513 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1516 let Constraints = "$src1 = $dst" in {
1517 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1518 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1520 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1521 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1525 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1526 let mayLoad = 0 in {
1527 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1528 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1529 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1532 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1533 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1535 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1537 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1538 SDNode OpNode, int HasPat = 0,
1539 list<list<dag>> Pattern = []> {
1540 let isAsmParserOnly = 1, Pattern = []<dag> in {
1541 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1542 !strconcat(OpcodeStr, "ps"), f128mem,
1543 !if(HasPat, Pattern[0], // rr
1544 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1546 !if(HasPat, Pattern[2], // rm
1547 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1548 (memopv2i64 addr:$src2)))]), 0>,
1551 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1552 !strconcat(OpcodeStr, "pd"), f128mem,
1553 !if(HasPat, Pattern[1], // rr
1554 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1557 !if(HasPat, Pattern[3], // rm
1558 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1559 (memopv2i64 addr:$src2)))]), 0>,
1562 let Constraints = "$src1 = $dst" in {
1563 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1564 !strconcat(OpcodeStr, "ps"), f128mem,
1565 !if(HasPat, Pattern[0], // rr
1566 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1568 !if(HasPat, Pattern[2], // rm
1569 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1570 (memopv2i64 addr:$src2)))])>, TB;
1572 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1573 !strconcat(OpcodeStr, "pd"), f128mem,
1574 !if(HasPat, Pattern[1], // rr
1575 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1578 !if(HasPat, Pattern[3], // rm
1579 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1580 (memopv2i64 addr:$src2)))])>,
1585 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1587 let isAsmParserOnly = 1 in {
1588 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1589 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1590 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1592 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1593 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1597 // AVX 256-bit packed logical ops forms
1598 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1599 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1600 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1601 let isCommutable = 0 in
1602 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1604 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1605 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1606 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1607 let isCommutable = 0 in
1608 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1610 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1611 (bc_v2i64 (v4i32 immAllOnesV))),
1614 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1615 (bc_v2i64 (v2f64 VR128:$src2))))],
1617 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1618 (bc_v2i64 (v4i32 immAllOnesV))),
1619 (memopv2i64 addr:$src2))))],
1621 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1622 (memopv2i64 addr:$src2)))]]>;
1624 //===----------------------------------------------------------------------===//
1625 // SSE 1 & 2 - Arithmetic Instructions
1626 //===----------------------------------------------------------------------===//
1628 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1631 /// In addition, we also have a special variant of the scalar form here to
1632 /// represent the associated intrinsic operation. This form is unlike the
1633 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1634 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1636 /// These three forms can each be reg+reg or reg+mem.
1639 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1641 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1643 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1644 OpNode, FR32, f32mem, Is2Addr>, XS;
1645 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1646 OpNode, FR64, f64mem, Is2Addr>, XD;
1649 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1651 let mayLoad = 0 in {
1652 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1653 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1654 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1655 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1659 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1661 let mayLoad = 0 in {
1662 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1663 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1664 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1665 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1669 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1671 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1672 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1673 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1674 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1677 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1679 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1680 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1681 SSEPackedSingle, Is2Addr>, TB;
1683 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1684 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1685 SSEPackedDouble, Is2Addr>, TB, OpSize;
1688 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1689 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1690 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1691 SSEPackedSingle, 0>, TB;
1693 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1694 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1695 SSEPackedDouble, 0>, TB, OpSize;
1698 // Binary Arithmetic instructions
1699 let isAsmParserOnly = 1 in {
1700 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1701 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1702 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1703 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1704 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1705 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1706 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1707 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1709 let isCommutable = 0 in {
1710 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1711 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1712 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1713 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1714 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1715 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1716 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1717 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1718 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1719 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1720 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1721 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1722 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1723 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1724 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1725 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1726 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1727 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1728 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1729 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1733 let Constraints = "$src1 = $dst" in {
1734 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1735 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1736 basic_sse12_fp_binop_s_int<0x58, "add">;
1737 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1738 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1739 basic_sse12_fp_binop_s_int<0x59, "mul">;
1741 let isCommutable = 0 in {
1742 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1743 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1744 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1745 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1746 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1747 basic_sse12_fp_binop_s_int<0x5E, "div">;
1748 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1749 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1750 basic_sse12_fp_binop_s_int<0x5F, "max">,
1751 basic_sse12_fp_binop_p_int<0x5F, "max">;
1752 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1753 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1754 basic_sse12_fp_binop_s_int<0x5D, "min">,
1755 basic_sse12_fp_binop_p_int<0x5D, "min">;
1760 /// In addition, we also have a special variant of the scalar form here to
1761 /// represent the associated intrinsic operation. This form is unlike the
1762 /// plain scalar form, in that it takes an entire vector (instead of a
1763 /// scalar) and leaves the top elements undefined.
1765 /// And, we have a special variant form for a full-vector intrinsic form.
1767 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1768 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1769 SDNode OpNode, Intrinsic F32Int> {
1770 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1771 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1772 [(set FR32:$dst, (OpNode FR32:$src))]>;
1773 // For scalar unary operations, fold a load into the operation
1774 // only in OptForSize mode. It eliminates an instruction, but it also
1775 // eliminates a whole-register clobber (the load), so it introduces a
1776 // partial register update condition.
1777 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1778 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1779 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1780 Requires<[HasSSE1, OptForSize]>;
1781 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1782 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1783 [(set VR128:$dst, (F32Int VR128:$src))]>;
1784 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1785 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1786 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1789 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1790 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1791 SDNode OpNode, Intrinsic F32Int> {
1792 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1793 !strconcat(OpcodeStr,
1794 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1795 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1796 !strconcat(OpcodeStr,
1797 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1798 []>, XS, Requires<[HasAVX, OptForSize]>;
1799 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1800 !strconcat(OpcodeStr,
1801 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1802 [(set VR128:$dst, (F32Int VR128:$src))]>;
1803 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1804 !strconcat(OpcodeStr,
1805 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1806 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1809 /// sse1_fp_unop_p - SSE1 unops in packed form.
1810 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1811 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1812 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1813 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1814 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1815 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1816 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1819 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1820 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1821 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1822 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1823 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1824 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1825 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1826 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1829 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1830 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1831 Intrinsic V4F32Int> {
1832 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1833 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1834 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1835 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1836 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1837 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1840 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1841 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1842 Intrinsic V4F32Int> {
1843 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1844 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1845 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1846 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1847 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1848 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1851 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1852 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1853 SDNode OpNode, Intrinsic F64Int> {
1854 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1855 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1856 [(set FR64:$dst, (OpNode FR64:$src))]>;
1857 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1858 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1859 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1860 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1861 Requires<[HasSSE2, OptForSize]>;
1862 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1863 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1864 [(set VR128:$dst, (F64Int VR128:$src))]>;
1865 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1866 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1867 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1870 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1871 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1872 SDNode OpNode, Intrinsic F64Int> {
1873 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1874 !strconcat(OpcodeStr,
1875 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1876 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1877 (ins FR64:$src1, f64mem:$src2),
1878 !strconcat(OpcodeStr,
1879 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1880 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1881 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1882 [(set VR128:$dst, (F64Int VR128:$src))]>;
1883 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1884 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1885 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1888 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1889 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1891 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1892 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1893 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1894 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1895 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1896 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1899 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1900 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1901 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1902 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1903 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1904 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1905 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1906 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1909 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1910 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1911 Intrinsic V2F64Int> {
1912 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1913 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1914 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1915 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1916 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1917 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1920 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1921 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1922 Intrinsic V2F64Int> {
1923 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1924 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1925 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1926 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1927 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1928 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1931 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1933 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1934 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1937 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1938 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1939 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1940 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1941 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1942 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1943 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1944 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1947 // Reciprocal approximations. Note that these typically require refinement
1948 // in order to obtain suitable precision.
1949 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1950 int_x86_sse_rsqrt_ss>, VEX_4V;
1951 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1952 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1953 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1954 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1956 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1958 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1959 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1960 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1961 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1965 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1966 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1967 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1968 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1969 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1970 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1972 // Reciprocal approximations. Note that these typically require refinement
1973 // in order to obtain suitable precision.
1974 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1975 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1976 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1977 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1978 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1979 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1981 // There is no f64 version of the reciprocal approximation instructions.
1983 //===----------------------------------------------------------------------===//
1984 // SSE 1 & 2 - Non-temporal stores
1985 //===----------------------------------------------------------------------===//
1987 let isAsmParserOnly = 1 in {
1988 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1989 (ins i128mem:$dst, VR128:$src),
1990 "movntps\t{$src, $dst|$dst, $src}",
1991 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1992 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1993 (ins i128mem:$dst, VR128:$src),
1994 "movntpd\t{$src, $dst|$dst, $src}",
1995 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1997 let ExeDomain = SSEPackedInt in
1998 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1999 (ins f128mem:$dst, VR128:$src),
2000 "movntdq\t{$src, $dst|$dst, $src}",
2001 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
2003 let AddedComplexity = 400 in { // Prefer non-temporal versions
2004 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2005 (ins f128mem:$dst, VR128:$src),
2006 "movntps\t{$src, $dst|$dst, $src}",
2007 [(alignednontemporalstore (v4f32 VR128:$src),
2009 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2010 (ins f128mem:$dst, VR128:$src),
2011 "movntpd\t{$src, $dst|$dst, $src}",
2012 [(alignednontemporalstore (v2f64 VR128:$src),
2014 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2015 (ins f128mem:$dst, VR128:$src),
2016 "movntdq\t{$src, $dst|$dst, $src}",
2017 [(alignednontemporalstore (v2f64 VR128:$src),
2019 let ExeDomain = SSEPackedInt in
2020 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2021 (ins f128mem:$dst, VR128:$src),
2022 "movntdq\t{$src, $dst|$dst, $src}",
2023 [(alignednontemporalstore (v4f32 VR128:$src),
2026 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2027 (ins f256mem:$dst, VR256:$src),
2028 "movntps\t{$src, $dst|$dst, $src}",
2029 [(alignednontemporalstore (v8f32 VR256:$src),
2031 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2032 (ins f256mem:$dst, VR256:$src),
2033 "movntpd\t{$src, $dst|$dst, $src}",
2034 [(alignednontemporalstore (v4f64 VR256:$src),
2036 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2037 (ins f256mem:$dst, VR256:$src),
2038 "movntdq\t{$src, $dst|$dst, $src}",
2039 [(alignednontemporalstore (v4f64 VR256:$src),
2041 let ExeDomain = SSEPackedInt in
2042 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2043 (ins f256mem:$dst, VR256:$src),
2044 "movntdq\t{$src, $dst|$dst, $src}",
2045 [(alignednontemporalstore (v8f32 VR256:$src),
2050 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2051 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2052 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2053 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2054 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2055 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2057 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2058 "movntps\t{$src, $dst|$dst, $src}",
2059 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2060 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2061 "movntpd\t{$src, $dst|$dst, $src}",
2062 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2064 let ExeDomain = SSEPackedInt in
2065 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2066 "movntdq\t{$src, $dst|$dst, $src}",
2067 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2069 let AddedComplexity = 400 in { // Prefer non-temporal versions
2070 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2071 "movntps\t{$src, $dst|$dst, $src}",
2072 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2073 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2074 "movntpd\t{$src, $dst|$dst, $src}",
2075 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2077 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2078 "movntdq\t{$src, $dst|$dst, $src}",
2079 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2081 let ExeDomain = SSEPackedInt in
2082 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2083 "movntdq\t{$src, $dst|$dst, $src}",
2084 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2086 // There is no AVX form for instructions below this point
2087 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2088 "movnti\t{$src, $dst|$dst, $src}",
2089 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2090 TB, Requires<[HasSSE2]>;
2092 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2093 "movnti\t{$src, $dst|$dst, $src}",
2094 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2095 TB, Requires<[HasSSE2]>;
2098 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2099 "movnti\t{$src, $dst|$dst, $src}",
2100 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2101 TB, Requires<[HasSSE2]>;
2103 //===----------------------------------------------------------------------===//
2104 // SSE 1 & 2 - Misc Instructions (No AVX form)
2105 //===----------------------------------------------------------------------===//
2107 // Prefetch intrinsic.
2108 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2109 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2110 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2111 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2112 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2113 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2114 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2115 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2117 // Load, store, and memory fence
2118 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2119 TB, Requires<[HasSSE1]>;
2120 def : Pat<(X86SFence), (SFENCE)>;
2122 // Alias instructions that map zero vector to pxor / xorp* for sse.
2123 // We set canFoldAsLoad because this can be converted to a constant-pool
2124 // load of an all-zeros value if folding it would be beneficial.
2125 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2126 // JIT implementatioan, it does not expand the instructions below like
2127 // X86MCInstLower does.
2128 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2129 isCodeGenOnly = 1 in {
2130 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2131 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2132 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2133 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2134 let ExeDomain = SSEPackedInt in
2135 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2136 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2139 // The same as done above but for AVX. The 128-bit versions are the
2140 // same, but re-encoded. The 256-bit does not support PI version.
2141 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2142 // JIT implementatioan, it does not expand the instructions below like
2143 // X86MCInstLower does.
2144 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2145 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2146 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2147 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2148 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2149 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2150 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2151 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2152 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2153 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2154 let ExeDomain = SSEPackedInt in
2155 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2156 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2159 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2160 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2161 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2163 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2164 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2166 //===----------------------------------------------------------------------===//
2167 // SSE 1 & 2 - Load/Store XCSR register
2168 //===----------------------------------------------------------------------===//
2170 let isAsmParserOnly = 1 in {
2171 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2172 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2173 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2174 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2177 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2178 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2179 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2180 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2182 //===---------------------------------------------------------------------===//
2183 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2184 //===---------------------------------------------------------------------===//
2186 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2188 let isAsmParserOnly = 1 in {
2189 let neverHasSideEffects = 1 in {
2190 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2191 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2192 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2193 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2195 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2196 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2197 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2198 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2200 let canFoldAsLoad = 1, mayLoad = 1 in {
2201 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2202 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2203 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2204 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2205 let Predicates = [HasAVX] in {
2206 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2207 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2208 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2209 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2213 let mayStore = 1 in {
2214 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2215 (ins i128mem:$dst, VR128:$src),
2216 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2217 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2218 (ins i256mem:$dst, VR256:$src),
2219 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2220 let Predicates = [HasAVX] in {
2221 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2222 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2223 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2224 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2229 let neverHasSideEffects = 1 in
2230 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2231 "movdqa\t{$src, $dst|$dst, $src}", []>;
2233 let canFoldAsLoad = 1, mayLoad = 1 in {
2234 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2235 "movdqa\t{$src, $dst|$dst, $src}",
2236 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2237 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2238 "movdqu\t{$src, $dst|$dst, $src}",
2239 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2240 XS, Requires<[HasSSE2]>;
2243 let mayStore = 1 in {
2244 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2245 "movdqa\t{$src, $dst|$dst, $src}",
2246 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2247 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2248 "movdqu\t{$src, $dst|$dst, $src}",
2249 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2250 XS, Requires<[HasSSE2]>;
2253 // Intrinsic forms of MOVDQU load and store
2254 let isAsmParserOnly = 1 in {
2255 let canFoldAsLoad = 1 in
2256 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2257 "vmovdqu\t{$src, $dst|$dst, $src}",
2258 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2259 XS, VEX, Requires<[HasAVX]>;
2260 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2261 "vmovdqu\t{$src, $dst|$dst, $src}",
2262 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2263 XS, VEX, Requires<[HasAVX]>;
2266 let canFoldAsLoad = 1 in
2267 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2268 "movdqu\t{$src, $dst|$dst, $src}",
2269 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2270 XS, Requires<[HasSSE2]>;
2271 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2272 "movdqu\t{$src, $dst|$dst, $src}",
2273 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2274 XS, Requires<[HasSSE2]>;
2276 } // ExeDomain = SSEPackedInt
2278 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2279 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2280 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2282 //===---------------------------------------------------------------------===//
2283 // SSE2 - Packed Integer Arithmetic Instructions
2284 //===---------------------------------------------------------------------===//
2286 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2288 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2289 bit IsCommutable = 0, bit Is2Addr = 1> {
2290 let isCommutable = IsCommutable in
2291 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2292 (ins VR128:$src1, VR128:$src2),
2294 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2295 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2296 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2297 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2298 (ins VR128:$src1, i128mem:$src2),
2300 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2301 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2302 [(set VR128:$dst, (IntId VR128:$src1,
2303 (bitconvert (memopv2i64 addr:$src2))))]>;
2306 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2307 string OpcodeStr, Intrinsic IntId,
2308 Intrinsic IntId2, bit Is2Addr = 1> {
2309 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2310 (ins VR128:$src1, VR128:$src2),
2312 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2314 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2315 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2316 (ins VR128:$src1, i128mem:$src2),
2318 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2320 [(set VR128:$dst, (IntId VR128:$src1,
2321 (bitconvert (memopv2i64 addr:$src2))))]>;
2322 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2323 (ins VR128:$src1, i32i8imm:$src2),
2325 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2327 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2330 /// PDI_binop_rm - Simple SSE2 binary operator.
2331 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2332 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2333 let isCommutable = IsCommutable in
2334 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2335 (ins VR128:$src1, VR128:$src2),
2337 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2339 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2340 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2341 (ins VR128:$src1, i128mem:$src2),
2343 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2345 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2346 (bitconvert (memopv2i64 addr:$src2)))))]>;
2349 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2351 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2352 /// to collapse (bitconvert VT to VT) into its operand.
2354 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2355 bit IsCommutable = 0, bit Is2Addr = 1> {
2356 let isCommutable = IsCommutable in
2357 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2358 (ins VR128:$src1, VR128:$src2),
2360 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2362 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2363 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2364 (ins VR128:$src1, i128mem:$src2),
2366 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2367 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2368 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2371 } // ExeDomain = SSEPackedInt
2373 // 128-bit Integer Arithmetic
2375 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2376 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2377 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2378 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2379 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2380 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2381 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2382 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2383 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2384 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2387 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2389 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2391 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2393 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2395 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2397 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2399 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2401 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2403 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2405 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2407 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2409 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2411 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2413 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2415 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2417 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2419 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2421 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2423 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2427 let Constraints = "$src1 = $dst" in {
2428 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2429 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2430 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2431 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2432 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2433 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2434 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2435 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2436 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2439 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2440 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2441 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2442 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2443 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2444 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2445 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2446 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2447 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2448 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2449 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2450 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2451 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2452 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2453 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2454 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2455 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2456 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2457 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2459 } // Constraints = "$src1 = $dst"
2461 //===---------------------------------------------------------------------===//
2462 // SSE2 - Packed Integer Logical Instructions
2463 //===---------------------------------------------------------------------===//
2465 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2466 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2467 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2469 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2470 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2472 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2473 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2476 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2477 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2479 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2480 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2482 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2483 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2486 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2487 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2489 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2490 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2493 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2494 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2495 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2497 let ExeDomain = SSEPackedInt in {
2498 let neverHasSideEffects = 1 in {
2499 // 128-bit logical shifts.
2500 def VPSLLDQri : PDIi8<0x73, MRM7r,
2501 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2502 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2504 def VPSRLDQri : PDIi8<0x73, MRM3r,
2505 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2506 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2508 // PSRADQri doesn't exist in SSE[1-3].
2510 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2511 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2512 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2513 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2514 VR128:$src2)))]>, VEX_4V;
2516 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2517 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2518 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2519 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2520 (memopv2i64 addr:$src2))))]>,
2525 let Constraints = "$src1 = $dst" in {
2526 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2527 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2528 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2529 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2530 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2531 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2533 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2534 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2535 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2536 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2537 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2538 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2540 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2541 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2542 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2543 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2545 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2546 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2547 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2549 let ExeDomain = SSEPackedInt in {
2550 let neverHasSideEffects = 1 in {
2551 // 128-bit logical shifts.
2552 def PSLLDQri : PDIi8<0x73, MRM7r,
2553 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2554 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2555 def PSRLDQri : PDIi8<0x73, MRM3r,
2556 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2557 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2558 // PSRADQri doesn't exist in SSE[1-3].
2560 def PANDNrr : PDI<0xDF, MRMSrcReg,
2561 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2562 "pandn\t{$src2, $dst|$dst, $src2}",
2563 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2566 def PANDNrm : PDI<0xDF, MRMSrcMem,
2567 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2568 "pandn\t{$src2, $dst|$dst, $src2}",
2569 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2570 (memopv2i64 addr:$src2))))]>;
2572 } // Constraints = "$src1 = $dst"
2574 let Predicates = [HasAVX] in {
2575 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2576 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2577 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2578 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2579 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2580 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2581 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2582 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2583 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2584 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2586 // Shift up / down and insert zero's.
2587 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2588 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2589 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2590 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2593 let Predicates = [HasSSE2] in {
2594 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2595 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2596 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2597 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2598 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2599 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2600 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2601 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2602 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2603 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2605 // Shift up / down and insert zero's.
2606 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2607 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2608 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2609 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2612 //===---------------------------------------------------------------------===//
2613 // SSE2 - Packed Integer Comparison Instructions
2614 //===---------------------------------------------------------------------===//
2616 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2617 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2619 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2621 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2623 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2625 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2627 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2631 let Constraints = "$src1 = $dst" in {
2632 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2633 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2634 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2635 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2636 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2637 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2638 } // Constraints = "$src1 = $dst"
2640 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2641 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2642 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2643 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2644 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2645 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2646 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2647 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2648 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2649 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2650 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2651 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2653 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2654 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2655 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2656 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2657 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2658 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2659 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2660 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2661 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2662 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2663 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2664 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2666 //===---------------------------------------------------------------------===//
2667 // SSE2 - Packed Integer Pack Instructions
2668 //===---------------------------------------------------------------------===//
2670 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2671 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2673 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2675 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2679 let Constraints = "$src1 = $dst" in {
2680 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2681 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2682 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2683 } // Constraints = "$src1 = $dst"
2685 //===---------------------------------------------------------------------===//
2686 // SSE2 - Packed Integer Shuffle Instructions
2687 //===---------------------------------------------------------------------===//
2689 let ExeDomain = SSEPackedInt in {
2690 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2692 def ri : Ii8<0x70, MRMSrcReg,
2693 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2694 !strconcat(OpcodeStr,
2695 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2696 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2698 def mi : Ii8<0x70, MRMSrcMem,
2699 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2700 !strconcat(OpcodeStr,
2701 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2702 [(set VR128:$dst, (vt (pshuf_frag:$src2
2703 (bc_frag (memopv2i64 addr:$src1)),
2706 } // ExeDomain = SSEPackedInt
2708 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2709 let AddedComplexity = 5 in
2710 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2713 // SSE2 with ImmT == Imm8 and XS prefix.
2714 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2717 // SSE2 with ImmT == Imm8 and XD prefix.
2718 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2722 let Predicates = [HasSSE2] in {
2723 let AddedComplexity = 5 in
2724 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2726 // SSE2 with ImmT == Imm8 and XS prefix.
2727 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2729 // SSE2 with ImmT == Imm8 and XD prefix.
2730 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2733 //===---------------------------------------------------------------------===//
2734 // SSE2 - Packed Integer Unpack Instructions
2735 //===---------------------------------------------------------------------===//
2737 let ExeDomain = SSEPackedInt in {
2738 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2739 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2740 def rr : PDI<opc, MRMSrcReg,
2741 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2743 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2744 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2745 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2746 def rm : PDI<opc, MRMSrcMem,
2747 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2749 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2750 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2751 [(set VR128:$dst, (unp_frag VR128:$src1,
2752 (bc_frag (memopv2i64
2756 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2757 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2759 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2761 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2764 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2765 /// knew to collapse (bitconvert VT to VT) into its operand.
2766 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2767 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2768 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2770 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2771 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2772 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2773 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2775 (v2i64 (unpckl VR128:$src1,
2776 (memopv2i64 addr:$src2))))]>, VEX_4V;
2778 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2780 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2782 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2785 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2786 /// knew to collapse (bitconvert VT to VT) into its operand.
2787 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2788 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2789 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2791 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2792 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2793 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2794 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2796 (v2i64 (unpckh VR128:$src1,
2797 (memopv2i64 addr:$src2))))]>, VEX_4V;
2800 let Constraints = "$src1 = $dst" in {
2801 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2802 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2803 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2805 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2806 /// knew to collapse (bitconvert VT to VT) into its operand.
2807 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2808 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2809 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2811 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2812 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2813 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2814 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2816 (v2i64 (unpckl VR128:$src1,
2817 (memopv2i64 addr:$src2))))]>;
2819 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2820 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2821 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2823 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2824 /// knew to collapse (bitconvert VT to VT) into its operand.
2825 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2826 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2827 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2829 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2830 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2831 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2832 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2834 (v2i64 (unpckh VR128:$src1,
2835 (memopv2i64 addr:$src2))))]>;
2838 } // ExeDomain = SSEPackedInt
2840 //===---------------------------------------------------------------------===//
2841 // SSE2 - Packed Integer Extract and Insert
2842 //===---------------------------------------------------------------------===//
2844 let ExeDomain = SSEPackedInt in {
2845 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2846 def rri : Ii8<0xC4, MRMSrcReg,
2847 (outs VR128:$dst), (ins VR128:$src1,
2848 GR32:$src2, i32i8imm:$src3),
2850 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2851 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2853 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2854 def rmi : Ii8<0xC4, MRMSrcMem,
2855 (outs VR128:$dst), (ins VR128:$src1,
2856 i16mem:$src2, i32i8imm:$src3),
2858 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2859 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2861 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2866 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2867 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2868 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2869 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2870 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2871 imm:$src2))]>, OpSize, VEX;
2872 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2873 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2874 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2875 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2879 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2880 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2881 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2882 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2883 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2884 []>, OpSize, VEX_4V;
2887 let Constraints = "$src1 = $dst" in
2888 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2890 } // ExeDomain = SSEPackedInt
2892 //===---------------------------------------------------------------------===//
2893 // SSE2 - Packed Mask Creation
2894 //===---------------------------------------------------------------------===//
2896 let ExeDomain = SSEPackedInt in {
2898 let isAsmParserOnly = 1 in {
2899 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2900 "pmovmskb\t{$src, $dst|$dst, $src}",
2901 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2902 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2903 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2905 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2906 "pmovmskb\t{$src, $dst|$dst, $src}",
2907 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2909 } // ExeDomain = SSEPackedInt
2911 //===---------------------------------------------------------------------===//
2912 // SSE2 - Conditional Store
2913 //===---------------------------------------------------------------------===//
2915 let ExeDomain = SSEPackedInt in {
2917 let isAsmParserOnly = 1 in {
2919 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2920 (ins VR128:$src, VR128:$mask),
2921 "maskmovdqu\t{$mask, $src|$src, $mask}",
2922 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2924 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2925 (ins VR128:$src, VR128:$mask),
2926 "maskmovdqu\t{$mask, $src|$src, $mask}",
2927 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2931 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2932 "maskmovdqu\t{$mask, $src|$src, $mask}",
2933 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2935 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2936 "maskmovdqu\t{$mask, $src|$src, $mask}",
2937 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2939 } // ExeDomain = SSEPackedInt
2941 //===---------------------------------------------------------------------===//
2942 // SSE2 - Move Doubleword
2943 //===---------------------------------------------------------------------===//
2945 // Move Int Doubleword to Packed Double Int
2946 let isAsmParserOnly = 1 in {
2947 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2948 "movd\t{$src, $dst|$dst, $src}",
2950 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2951 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2954 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2957 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2958 "movd\t{$src, $dst|$dst, $src}",
2960 (v4i32 (scalar_to_vector GR32:$src)))]>;
2961 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2962 "movd\t{$src, $dst|$dst, $src}",
2964 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2967 // Move Int Doubleword to Single Scalar
2968 let isAsmParserOnly = 1 in {
2969 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2970 "movd\t{$src, $dst|$dst, $src}",
2971 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2973 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2974 "movd\t{$src, $dst|$dst, $src}",
2975 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2978 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2979 "movd\t{$src, $dst|$dst, $src}",
2980 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2982 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2983 "movd\t{$src, $dst|$dst, $src}",
2984 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2986 // Move Packed Doubleword Int to Packed Double Int
2987 let isAsmParserOnly = 1 in {
2988 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2989 "movd\t{$src, $dst|$dst, $src}",
2990 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2992 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2993 (ins i32mem:$dst, VR128:$src),
2994 "movd\t{$src, $dst|$dst, $src}",
2995 [(store (i32 (vector_extract (v4i32 VR128:$src),
2996 (iPTR 0))), addr:$dst)]>, VEX;
2998 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2999 "movd\t{$src, $dst|$dst, $src}",
3000 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3002 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3003 "movd\t{$src, $dst|$dst, $src}",
3004 [(store (i32 (vector_extract (v4i32 VR128:$src),
3005 (iPTR 0))), addr:$dst)]>;
3007 // Move Scalar Single to Double Int
3008 let isAsmParserOnly = 1 in {
3009 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3010 "movd\t{$src, $dst|$dst, $src}",
3011 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3012 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3013 "movd\t{$src, $dst|$dst, $src}",
3014 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3016 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3017 "movd\t{$src, $dst|$dst, $src}",
3018 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3019 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3020 "movd\t{$src, $dst|$dst, $src}",
3021 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3023 // movd / movq to XMM register zero-extends
3024 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3025 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3026 "movd\t{$src, $dst|$dst, $src}",
3027 [(set VR128:$dst, (v4i32 (X86vzmovl
3028 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3030 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3031 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3032 [(set VR128:$dst, (v2i64 (X86vzmovl
3033 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3036 let AddedComplexity = 15 in {
3037 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3038 "movd\t{$src, $dst|$dst, $src}",
3039 [(set VR128:$dst, (v4i32 (X86vzmovl
3040 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3041 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3042 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3043 [(set VR128:$dst, (v2i64 (X86vzmovl
3044 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3047 let AddedComplexity = 20 in {
3048 let isAsmParserOnly = 1 in
3049 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3050 "movd\t{$src, $dst|$dst, $src}",
3052 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3053 (loadi32 addr:$src))))))]>,
3055 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3056 "movd\t{$src, $dst|$dst, $src}",
3058 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3059 (loadi32 addr:$src))))))]>;
3061 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3062 (MOVZDI2PDIrm addr:$src)>;
3063 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3064 (MOVZDI2PDIrm addr:$src)>;
3065 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3066 (MOVZDI2PDIrm addr:$src)>;
3069 //===---------------------------------------------------------------------===//
3070 // SSE2 - Move Quadword
3071 //===---------------------------------------------------------------------===//
3073 // Move Quadword Int to Packed Quadword Int
3074 let isAsmParserOnly = 1 in
3075 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3076 "vmovq\t{$src, $dst|$dst, $src}",
3078 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3079 VEX, Requires<[HasAVX]>;
3080 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3081 "movq\t{$src, $dst|$dst, $src}",
3083 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3084 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3086 // Move Packed Quadword Int to Quadword Int
3087 let isAsmParserOnly = 1 in
3088 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3089 "movq\t{$src, $dst|$dst, $src}",
3090 [(store (i64 (vector_extract (v2i64 VR128:$src),
3091 (iPTR 0))), addr:$dst)]>, VEX;
3092 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3093 "movq\t{$src, $dst|$dst, $src}",
3094 [(store (i64 (vector_extract (v2i64 VR128:$src),
3095 (iPTR 0))), addr:$dst)]>;
3097 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3098 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3100 // Store / copy lower 64-bits of a XMM register.
3101 let isAsmParserOnly = 1 in
3102 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3103 "movq\t{$src, $dst|$dst, $src}",
3104 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3105 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3106 "movq\t{$src, $dst|$dst, $src}",
3107 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3109 let AddedComplexity = 20, isAsmParserOnly = 1 in
3110 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3111 "vmovq\t{$src, $dst|$dst, $src}",
3113 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3114 (loadi64 addr:$src))))))]>,
3115 XS, VEX, Requires<[HasAVX]>;
3117 let AddedComplexity = 20 in {
3118 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3119 "movq\t{$src, $dst|$dst, $src}",
3121 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3122 (loadi64 addr:$src))))))]>,
3123 XS, Requires<[HasSSE2]>;
3125 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3126 (MOVZQI2PQIrm addr:$src)>;
3127 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3128 (MOVZQI2PQIrm addr:$src)>;
3129 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3132 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3133 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3134 let isAsmParserOnly = 1, AddedComplexity = 15 in
3135 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3136 "vmovq\t{$src, $dst|$dst, $src}",
3137 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3138 XS, VEX, Requires<[HasAVX]>;
3139 let AddedComplexity = 15 in
3140 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3141 "movq\t{$src, $dst|$dst, $src}",
3142 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3143 XS, Requires<[HasSSE2]>;
3145 let AddedComplexity = 20, isAsmParserOnly = 1 in
3146 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3147 "vmovq\t{$src, $dst|$dst, $src}",
3148 [(set VR128:$dst, (v2i64 (X86vzmovl
3149 (loadv2i64 addr:$src))))]>,
3150 XS, VEX, Requires<[HasAVX]>;
3151 let AddedComplexity = 20 in {
3152 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3153 "movq\t{$src, $dst|$dst, $src}",
3154 [(set VR128:$dst, (v2i64 (X86vzmovl
3155 (loadv2i64 addr:$src))))]>,
3156 XS, Requires<[HasSSE2]>;
3158 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3159 (MOVZPQILo2PQIrm addr:$src)>;
3162 // Instructions to match in the assembler
3163 let isAsmParserOnly = 1 in {
3164 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3165 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3166 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3167 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3168 // Recognize "movd" with GR64 destination, but encode as a "movq"
3169 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3170 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3173 // Instructions for the disassembler
3174 // xr = XMM register
3177 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3178 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3179 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3180 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3181 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3183 //===---------------------------------------------------------------------===//
3184 // SSE2 - Misc Instructions
3185 //===---------------------------------------------------------------------===//
3188 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3189 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3190 TB, Requires<[HasSSE2]>;
3192 // Load, store, and memory fence
3193 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3194 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3195 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3196 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3197 def : Pat<(X86LFence), (LFENCE)>;
3198 def : Pat<(X86MFence), (MFENCE)>;
3201 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3202 // was introduced with SSE2, it's backward compatible.
3203 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3205 // Alias instructions that map zero vector to pxor / xorp* for sse.
3206 // We set canFoldAsLoad because this can be converted to a constant-pool
3207 // load of an all-ones value if folding it would be beneficial.
3208 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3209 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3210 // FIXME: Change encoding to pseudo.
3211 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3212 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3214 //===---------------------------------------------------------------------===//
3215 // SSE3 - Conversion Instructions
3216 //===---------------------------------------------------------------------===//
3218 // Convert Packed Double FP to Packed DW Integers
3219 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3220 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3221 // register, but the same isn't true when using memory operands instead.
3222 // Provide other assembly rr and rm forms to address this explicitly.
3223 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3224 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3225 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3226 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3229 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3230 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3231 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3232 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3235 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3236 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3237 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3238 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3241 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3242 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3243 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3244 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3246 // Convert Packed DW Integers to Packed Double FP
3247 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3248 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3249 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3250 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3251 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3252 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3253 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3254 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3255 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3258 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3259 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3260 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3261 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3263 // AVX 256-bit register conversion intrinsics
3264 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3265 (VCVTDQ2PDYrr VR128:$src)>;
3266 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3267 (VCVTDQ2PDYrm addr:$src)>;
3269 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3270 (VCVTPD2DQYrr VR256:$src)>;
3271 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3272 (VCVTPD2DQYrm addr:$src)>;
3274 //===---------------------------------------------------------------------===//
3275 // SSE3 - Move Instructions
3276 //===---------------------------------------------------------------------===//
3278 // Replicate Single FP
3279 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3280 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3281 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3282 [(set VR128:$dst, (v4f32 (rep_frag
3283 VR128:$src, (undef))))]>;
3284 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3285 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3286 [(set VR128:$dst, (rep_frag
3287 (memopv4f32 addr:$src), (undef)))]>;
3290 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3292 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3293 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3294 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3295 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3298 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3299 // FIXME: Merge above classes when we have patterns for the ymm version
3300 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3301 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3302 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3303 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3305 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3306 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3308 // Replicate Double FP
3309 multiclass sse3_replicate_dfp<string OpcodeStr> {
3310 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3311 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3312 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3313 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3314 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3316 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3320 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3321 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3322 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3324 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3325 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3329 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3330 // FIXME: Merge above classes when we have patterns for the ymm version
3331 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3332 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3334 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3336 // Move Unaligned Integer
3337 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3338 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3339 "vlddqu\t{$src, $dst|$dst, $src}",
3340 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3341 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3342 "vlddqu\t{$src, $dst|$dst, $src}",
3343 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3345 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3346 "lddqu\t{$src, $dst|$dst, $src}",
3347 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3349 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3351 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3353 // Several Move patterns
3354 let AddedComplexity = 5 in {
3355 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3356 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3357 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3358 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3359 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3360 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3361 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3362 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3365 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3366 let AddedComplexity = 15 in
3367 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3368 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3369 let AddedComplexity = 20 in
3370 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3371 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3373 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3374 let AddedComplexity = 15 in
3375 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3376 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3377 let AddedComplexity = 20 in
3378 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3379 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3381 //===---------------------------------------------------------------------===//
3382 // SSE3 - Arithmetic
3383 //===---------------------------------------------------------------------===//
3385 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3386 X86MemOperand x86memop, bit Is2Addr = 1> {
3387 def rr : I<0xD0, MRMSrcReg,
3388 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3390 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3391 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3392 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3393 def rm : I<0xD0, MRMSrcMem,
3394 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3396 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3397 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3398 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3401 let isAsmParserOnly = 1, Predicates = [HasAVX],
3402 ExeDomain = SSEPackedDouble in {
3403 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3404 f128mem, 0>, XD, VEX_4V;
3405 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3406 f128mem, 0>, OpSize, VEX_4V;
3407 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3408 f256mem, 0>, XD, VEX_4V;
3409 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3410 f256mem, 0>, OpSize, VEX_4V;
3412 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3413 ExeDomain = SSEPackedDouble in {
3414 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3416 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3417 f128mem>, TB, OpSize;
3420 //===---------------------------------------------------------------------===//
3421 // SSE3 Instructions
3422 //===---------------------------------------------------------------------===//
3425 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3426 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3427 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3429 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3430 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3431 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3433 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3435 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3436 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3437 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3439 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3440 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3441 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3443 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3444 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3445 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3447 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3449 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3450 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3451 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3454 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3455 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3456 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3457 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3458 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3459 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3460 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3461 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3462 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3463 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3464 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3465 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3466 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3467 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3468 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3469 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3470 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3473 let Constraints = "$src1 = $dst" in {
3474 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3475 int_x86_sse3_hadd_ps>;
3476 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3477 int_x86_sse3_hadd_pd>;
3478 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3479 int_x86_sse3_hsub_ps>;
3480 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3481 int_x86_sse3_hsub_pd>;
3484 //===---------------------------------------------------------------------===//
3485 // SSSE3 - Packed Absolute Instructions
3486 //===---------------------------------------------------------------------===//
3489 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3490 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3491 PatFrag mem_frag128, Intrinsic IntId128> {
3492 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3495 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3498 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3503 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3506 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3507 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3508 int_x86_ssse3_pabs_b_128>, VEX;
3509 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3510 int_x86_ssse3_pabs_w_128>, VEX;
3511 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3512 int_x86_ssse3_pabs_d_128>, VEX;
3515 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3516 int_x86_ssse3_pabs_b_128>;
3517 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3518 int_x86_ssse3_pabs_w_128>;
3519 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3520 int_x86_ssse3_pabs_d_128>;
3522 //===---------------------------------------------------------------------===//
3523 // SSSE3 - Packed Binary Operator Instructions
3524 //===---------------------------------------------------------------------===//
3526 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3527 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3528 PatFrag mem_frag128, Intrinsic IntId128,
3530 let isCommutable = 1 in
3531 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3532 (ins VR128:$src1, VR128:$src2),
3534 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3535 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3536 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3538 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3539 (ins VR128:$src1, i128mem:$src2),
3541 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3542 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3544 (IntId128 VR128:$src1,
3545 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3548 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3549 let isCommutable = 0 in {
3550 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3551 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3552 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3553 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3554 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3555 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3556 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3557 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3558 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3559 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3560 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3561 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3562 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3563 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3564 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3565 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3566 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3567 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3568 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3569 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3570 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3571 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3573 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3574 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3577 // None of these have i8 immediate fields.
3578 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3579 let isCommutable = 0 in {
3580 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3581 int_x86_ssse3_phadd_w_128>;
3582 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3583 int_x86_ssse3_phadd_d_128>;
3584 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3585 int_x86_ssse3_phadd_sw_128>;
3586 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3587 int_x86_ssse3_phsub_w_128>;
3588 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3589 int_x86_ssse3_phsub_d_128>;
3590 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3591 int_x86_ssse3_phsub_sw_128>;
3592 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3593 int_x86_ssse3_pmadd_ub_sw_128>;
3594 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3595 int_x86_ssse3_pshuf_b_128>;
3596 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3597 int_x86_ssse3_psign_b_128>;
3598 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3599 int_x86_ssse3_psign_w_128>;
3600 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3601 int_x86_ssse3_psign_d_128>;
3603 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3604 int_x86_ssse3_pmul_hr_sw_128>;
3607 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3608 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3609 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3610 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3612 //===---------------------------------------------------------------------===//
3613 // SSSE3 - Packed Align Instruction Patterns
3614 //===---------------------------------------------------------------------===//
3616 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3617 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3618 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3620 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3622 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3624 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3625 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3627 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3629 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3633 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3634 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3635 let Constraints = "$src1 = $dst" in
3636 defm PALIGN : ssse3_palign<"palignr">;
3638 let AddedComplexity = 5 in {
3639 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3640 (PALIGNR128rr VR128:$src2, VR128:$src1,
3641 (SHUFFLE_get_palign_imm VR128:$src3))>,
3642 Requires<[HasSSSE3]>;
3643 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3644 (PALIGNR128rr VR128:$src2, VR128:$src1,
3645 (SHUFFLE_get_palign_imm VR128:$src3))>,
3646 Requires<[HasSSSE3]>;
3647 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3648 (PALIGNR128rr VR128:$src2, VR128:$src1,
3649 (SHUFFLE_get_palign_imm VR128:$src3))>,
3650 Requires<[HasSSSE3]>;
3651 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3652 (PALIGNR128rr VR128:$src2, VR128:$src1,
3653 (SHUFFLE_get_palign_imm VR128:$src3))>,
3654 Requires<[HasSSSE3]>;
3657 //===---------------------------------------------------------------------===//
3658 // SSSE3 Misc Instructions
3659 //===---------------------------------------------------------------------===//
3661 // Thread synchronization
3662 def MONITOR : I<0x01, MRM_C8, (outs), (ins), "monitor",
3663 [(int_x86_sse3_monitor EAX, ECX, EDX)]>,TB, Requires<[HasSSE3]>;
3664 def MWAIT : I<0x01, MRM_C9, (outs), (ins), "mwait",
3665 [(int_x86_sse3_mwait ECX, EAX)]>, TB, Requires<[HasSSE3]>;
3667 //===---------------------------------------------------------------------===//
3668 // Non-Instruction Patterns
3669 //===---------------------------------------------------------------------===//
3671 // extload f32 -> f64. This matches load+fextend because we have a hack in
3672 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3674 // Since these loads aren't folded into the fextend, we have to match it
3676 let Predicates = [HasSSE2] in
3677 def : Pat<(fextend (loadf32 addr:$src)),
3678 (CVTSS2SDrm addr:$src)>;
3681 let Predicates = [HasSSE2] in {
3682 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3683 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3684 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3685 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3686 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3687 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3688 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3689 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3690 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3691 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3692 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3693 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3694 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3695 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3696 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3697 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3698 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3699 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3700 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3701 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3702 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3703 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3704 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3705 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3706 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3707 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3708 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3709 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3710 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3711 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3714 // Move scalar to XMM zero-extended
3715 // movd to XMM register zero-extends
3716 let AddedComplexity = 15 in {
3717 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3718 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3719 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3720 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3721 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3722 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3723 (MOVSSrr (v4f32 (V_SET0PS)),
3724 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3725 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3726 (MOVSSrr (v4i32 (V_SET0PI)),
3727 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3730 // Splat v2f64 / v2i64
3731 let AddedComplexity = 10 in {
3732 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3733 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3734 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3735 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3736 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3737 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3738 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3739 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3742 // Special unary SHUFPSrri case.
3743 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3744 (SHUFPSrri VR128:$src1, VR128:$src1,
3745 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3746 let AddedComplexity = 5 in
3747 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3748 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3749 Requires<[HasSSE2]>;
3750 // Special unary SHUFPDrri case.
3751 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3752 (SHUFPDrri VR128:$src1, VR128:$src1,
3753 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3754 Requires<[HasSSE2]>;
3755 // Special unary SHUFPDrri case.
3756 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3757 (SHUFPDrri VR128:$src1, VR128:$src1,
3758 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3759 Requires<[HasSSE2]>;
3760 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3761 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3762 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3763 Requires<[HasSSE2]>;
3765 // Special binary v4i32 shuffle cases with SHUFPS.
3766 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3767 (SHUFPSrri VR128:$src1, VR128:$src2,
3768 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3769 Requires<[HasSSE2]>;
3770 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3771 (SHUFPSrmi VR128:$src1, addr:$src2,
3772 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3773 Requires<[HasSSE2]>;
3774 // Special binary v2i64 shuffle cases using SHUFPDrri.
3775 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3776 (SHUFPDrri VR128:$src1, VR128:$src2,
3777 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3778 Requires<[HasSSE2]>;
3780 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3781 let AddedComplexity = 15 in {
3782 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3783 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3784 Requires<[OptForSpeed, HasSSE2]>;
3785 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3786 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3787 Requires<[OptForSpeed, HasSSE2]>;
3789 let AddedComplexity = 10 in {
3790 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3791 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3792 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3793 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3794 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3795 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3796 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3797 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3800 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3801 let AddedComplexity = 15 in {
3802 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3803 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3804 Requires<[OptForSpeed, HasSSE2]>;
3805 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3806 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3807 Requires<[OptForSpeed, HasSSE2]>;
3809 let AddedComplexity = 10 in {
3810 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3811 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3812 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3813 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3814 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3815 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3816 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3817 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3820 let AddedComplexity = 20 in {
3821 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3822 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3823 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3825 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3826 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3827 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3829 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3830 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3831 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3832 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3833 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3836 let AddedComplexity = 20 in {
3837 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3838 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3839 (MOVLPSrm VR128:$src1, addr:$src2)>;
3840 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3841 (MOVLPDrm VR128:$src1, addr:$src2)>;
3842 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3843 (MOVLPSrm VR128:$src1, addr:$src2)>;
3844 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3845 (MOVLPDrm VR128:$src1, addr:$src2)>;
3848 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3849 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3850 (MOVLPSmr addr:$src1, VR128:$src2)>;
3851 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3852 (MOVLPDmr addr:$src1, VR128:$src2)>;
3853 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3855 (MOVLPSmr addr:$src1, VR128:$src2)>;
3856 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3857 (MOVLPDmr addr:$src1, VR128:$src2)>;
3859 let AddedComplexity = 15 in {
3860 // Setting the lowest element in the vector.
3861 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3862 (MOVSSrr (v4i32 VR128:$src1),
3863 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3864 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3865 (MOVSDrr (v2i64 VR128:$src1),
3866 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3868 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3869 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3870 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3871 Requires<[HasSSE2]>;
3872 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3873 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3874 Requires<[HasSSE2]>;
3877 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3878 // fall back to this for SSE1)
3879 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3880 (SHUFPSrri VR128:$src2, VR128:$src1,
3881 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3883 // Set lowest element and zero upper elements.
3884 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3885 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3887 // Some special case pandn patterns.
3888 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3890 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3891 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3893 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3894 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3896 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3898 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3899 (memop addr:$src2))),
3900 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3901 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3902 (memop addr:$src2))),
3903 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3904 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3905 (memop addr:$src2))),
3906 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3908 // vector -> vector casts
3909 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3910 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3911 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3912 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3914 // Use movaps / movups for SSE integer load / store (one byte shorter).
3915 let Predicates = [HasSSE1] in {
3916 def : Pat<(alignedloadv4i32 addr:$src),
3917 (MOVAPSrm addr:$src)>;
3918 def : Pat<(loadv4i32 addr:$src),
3919 (MOVUPSrm addr:$src)>;
3920 def : Pat<(alignedloadv2i64 addr:$src),
3921 (MOVAPSrm addr:$src)>;
3922 def : Pat<(loadv2i64 addr:$src),
3923 (MOVUPSrm addr:$src)>;
3925 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3926 (MOVAPSmr addr:$dst, VR128:$src)>;
3927 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3928 (MOVAPSmr addr:$dst, VR128:$src)>;
3929 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3930 (MOVAPSmr addr:$dst, VR128:$src)>;
3931 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3932 (MOVAPSmr addr:$dst, VR128:$src)>;
3933 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3934 (MOVUPSmr addr:$dst, VR128:$src)>;
3935 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3936 (MOVUPSmr addr:$dst, VR128:$src)>;
3937 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3938 (MOVUPSmr addr:$dst, VR128:$src)>;
3939 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3940 (MOVUPSmr addr:$dst, VR128:$src)>;
3943 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3944 let Predicates = [HasAVX] in {
3945 def : Pat<(alignedloadv4i32 addr:$src),
3946 (VMOVAPSrm addr:$src)>;
3947 def : Pat<(loadv4i32 addr:$src),
3948 (VMOVUPSrm addr:$src)>;
3949 def : Pat<(alignedloadv2i64 addr:$src),
3950 (VMOVAPSrm addr:$src)>;
3951 def : Pat<(loadv2i64 addr:$src),
3952 (VMOVUPSrm addr:$src)>;
3954 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3955 (VMOVAPSmr addr:$dst, VR128:$src)>;
3956 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3957 (VMOVAPSmr addr:$dst, VR128:$src)>;
3958 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3959 (VMOVAPSmr addr:$dst, VR128:$src)>;
3960 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3961 (VMOVAPSmr addr:$dst, VR128:$src)>;
3962 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3963 (VMOVUPSmr addr:$dst, VR128:$src)>;
3964 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3965 (VMOVUPSmr addr:$dst, VR128:$src)>;
3966 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3967 (VMOVUPSmr addr:$dst, VR128:$src)>;
3968 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3969 (VMOVUPSmr addr:$dst, VR128:$src)>;
3972 //===----------------------------------------------------------------------===//
3973 // SSE4.1 - Packed Move with Sign/Zero Extend
3974 //===----------------------------------------------------------------------===//
3976 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3977 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3978 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3979 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3981 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3982 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3984 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3988 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3989 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3991 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3993 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3995 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3997 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3999 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4003 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4004 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4005 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4006 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4007 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4008 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4010 // Common patterns involving scalar load.
4011 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4012 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4013 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4014 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4016 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4017 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4018 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4019 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4021 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4022 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4023 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4024 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4026 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4027 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4028 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4029 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4031 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4032 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4033 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4034 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4036 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4037 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4038 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4039 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4042 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4043 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4044 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4045 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4047 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4048 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4050 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4054 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4055 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4057 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4059 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4061 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4065 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4066 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4067 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4068 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4070 // Common patterns involving scalar load
4071 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4072 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4073 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4074 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4076 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4077 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4078 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4079 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4082 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4083 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4084 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4085 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4087 // Expecting a i16 load any extended to i32 value.
4088 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4089 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4090 [(set VR128:$dst, (IntId (bitconvert
4091 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4095 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4096 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4098 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4101 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4102 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4104 // Common patterns involving scalar load
4105 def : Pat<(int_x86_sse41_pmovsxbq
4106 (bitconvert (v4i32 (X86vzmovl
4107 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4108 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4110 def : Pat<(int_x86_sse41_pmovzxbq
4111 (bitconvert (v4i32 (X86vzmovl
4112 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4113 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4115 //===----------------------------------------------------------------------===//
4116 // SSE4.1 - Extract Instructions
4117 //===----------------------------------------------------------------------===//
4119 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4120 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4121 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4122 (ins VR128:$src1, i32i8imm:$src2),
4123 !strconcat(OpcodeStr,
4124 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4125 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4127 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4128 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4129 !strconcat(OpcodeStr,
4130 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4133 // There's an AssertZext in the way of writing the store pattern
4134 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4137 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4138 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4139 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4140 (ins VR128:$src1, i32i8imm:$src2),
4141 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4144 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4147 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4148 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4149 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4150 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4151 !strconcat(OpcodeStr,
4152 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4155 // There's an AssertZext in the way of writing the store pattern
4156 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4159 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4160 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4162 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4165 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4166 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4167 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4168 (ins VR128:$src1, i32i8imm:$src2),
4169 !strconcat(OpcodeStr,
4170 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4172 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4173 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4174 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4175 !strconcat(OpcodeStr,
4176 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4177 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4178 addr:$dst)]>, OpSize;
4181 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4182 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4184 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4186 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4187 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4188 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4189 (ins VR128:$src1, i32i8imm:$src2),
4190 !strconcat(OpcodeStr,
4191 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4193 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4194 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4195 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4196 !strconcat(OpcodeStr,
4197 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4198 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4199 addr:$dst)]>, OpSize, REX_W;
4202 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4203 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4205 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4207 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4209 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4210 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4211 (ins VR128:$src1, i32i8imm:$src2),
4212 !strconcat(OpcodeStr,
4213 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4215 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4217 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4218 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4219 !strconcat(OpcodeStr,
4220 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4221 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4222 addr:$dst)]>, OpSize;
4225 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4226 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4227 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4228 (ins VR128:$src1, i32i8imm:$src2),
4229 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4232 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4234 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4235 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4238 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4239 Requires<[HasSSE41]>;
4241 //===----------------------------------------------------------------------===//
4242 // SSE4.1 - Insert Instructions
4243 //===----------------------------------------------------------------------===//
4245 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4246 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4247 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4249 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4251 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4253 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4254 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4255 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4257 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4259 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4261 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4262 imm:$src3))]>, OpSize;
4265 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4266 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4267 let Constraints = "$src1 = $dst" in
4268 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4270 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4271 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4272 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4274 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4276 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4278 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4280 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4281 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4283 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4285 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4287 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4288 imm:$src3)))]>, OpSize;
4291 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4292 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4293 let Constraints = "$src1 = $dst" in
4294 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4296 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4297 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4298 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4300 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4302 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4304 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4306 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4307 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4309 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4311 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4313 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4314 imm:$src3)))]>, OpSize;
4317 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4318 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4319 let Constraints = "$src1 = $dst" in
4320 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4322 // insertps has a few different modes, there's the first two here below which
4323 // are optimized inserts that won't zero arbitrary elements in the destination
4324 // vector. The next one matches the intrinsic and could zero arbitrary elements
4325 // in the target vector.
4326 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4327 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4328 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4330 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4332 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4334 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4336 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4337 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4339 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4341 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4343 (X86insrtps VR128:$src1,
4344 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4345 imm:$src3))]>, OpSize;
4348 let Constraints = "$src1 = $dst" in
4349 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4350 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4351 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4353 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4354 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4356 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4357 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4358 Requires<[HasSSE41]>;
4360 //===----------------------------------------------------------------------===//
4361 // SSE4.1 - Round Instructions
4362 //===----------------------------------------------------------------------===//
4364 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4365 X86MemOperand x86memop, RegisterClass RC,
4366 PatFrag mem_frag32, PatFrag mem_frag64,
4367 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4368 // Intrinsic operation, reg.
4369 // Vector intrinsic operation, reg
4370 def PSr : SS4AIi8<opcps, MRMSrcReg,
4371 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4372 !strconcat(OpcodeStr,
4373 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4374 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4377 // Vector intrinsic operation, mem
4378 def PSm : Ii8<opcps, MRMSrcMem,
4379 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4380 !strconcat(OpcodeStr,
4381 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4383 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4385 Requires<[HasSSE41]>;
4387 // Vector intrinsic operation, reg
4388 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4389 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4390 !strconcat(OpcodeStr,
4391 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4392 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4395 // Vector intrinsic operation, mem
4396 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4397 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4398 !strconcat(OpcodeStr,
4399 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4401 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4405 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4406 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4407 // Intrinsic operation, reg.
4408 // Vector intrinsic operation, reg
4409 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4410 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4411 !strconcat(OpcodeStr,
4412 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4415 // Vector intrinsic operation, mem
4416 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4417 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4418 !strconcat(OpcodeStr,
4419 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4420 []>, TA, OpSize, Requires<[HasSSE41]>;
4422 // Vector intrinsic operation, reg
4423 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4424 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4425 !strconcat(OpcodeStr,
4426 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4429 // Vector intrinsic operation, mem
4430 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4431 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4432 !strconcat(OpcodeStr,
4433 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4437 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4440 Intrinsic F64Int, bit Is2Addr = 1> {
4441 // Intrinsic operation, reg.
4442 def SSr : SS4AIi8<opcss, MRMSrcReg,
4443 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4445 !strconcat(OpcodeStr,
4446 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4447 !strconcat(OpcodeStr,
4448 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4449 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4452 // Intrinsic operation, mem.
4453 def SSm : SS4AIi8<opcss, MRMSrcMem,
4454 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4456 !strconcat(OpcodeStr,
4457 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4458 !strconcat(OpcodeStr,
4459 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4461 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4464 // Intrinsic operation, reg.
4465 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4466 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4468 !strconcat(OpcodeStr,
4469 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4470 !strconcat(OpcodeStr,
4471 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4472 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4475 // Intrinsic operation, mem.
4476 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4477 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4479 !strconcat(OpcodeStr,
4480 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4481 !strconcat(OpcodeStr,
4482 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4484 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4488 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4490 // Intrinsic operation, reg.
4491 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4492 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4493 !strconcat(OpcodeStr,
4494 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4497 // Intrinsic operation, mem.
4498 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4499 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4500 !strconcat(OpcodeStr,
4501 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4504 // Intrinsic operation, reg.
4505 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4506 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4507 !strconcat(OpcodeStr,
4508 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4511 // Intrinsic operation, mem.
4512 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4513 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4514 !strconcat(OpcodeStr,
4515 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4519 // FP round - roundss, roundps, roundsd, roundpd
4520 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4522 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4523 memopv4f32, memopv2f64,
4524 int_x86_sse41_round_ps,
4525 int_x86_sse41_round_pd>, VEX;
4526 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4527 memopv8f32, memopv4f64,
4528 int_x86_avx_round_ps_256,
4529 int_x86_avx_round_pd_256>, VEX;
4530 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4531 int_x86_sse41_round_ss,
4532 int_x86_sse41_round_sd, 0>, VEX_4V;
4534 // Instructions for the assembler
4535 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4537 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4539 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4542 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4543 memopv4f32, memopv2f64,
4544 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4545 let Constraints = "$src1 = $dst" in
4546 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4547 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4549 //===----------------------------------------------------------------------===//
4550 // SSE4.1 - Packed Bit Test
4551 //===----------------------------------------------------------------------===//
4553 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4554 // the intel intrinsic that corresponds to this.
4555 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4556 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4557 "vptest\t{$src2, $src1|$src1, $src2}",
4558 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4560 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4561 "vptest\t{$src2, $src1|$src1, $src2}",
4562 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4565 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4566 "vptest\t{$src2, $src1|$src1, $src2}",
4567 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4569 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4570 "vptest\t{$src2, $src1|$src1, $src2}",
4571 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4575 let Defs = [EFLAGS] in {
4576 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4577 "ptest \t{$src2, $src1|$src1, $src2}",
4578 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4580 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4581 "ptest \t{$src2, $src1|$src1, $src2}",
4582 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4586 // The bit test instructions below are AVX only
4587 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4588 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4589 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4590 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4591 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4592 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4593 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4594 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4598 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4599 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4600 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4601 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4602 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4605 //===----------------------------------------------------------------------===//
4606 // SSE4.1 - Misc Instructions
4607 //===----------------------------------------------------------------------===//
4609 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4610 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4611 Intrinsic IntId128> {
4612 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4614 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4615 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4616 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4618 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4621 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4624 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4625 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4626 int_x86_sse41_phminposuw>, VEX;
4627 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4628 int_x86_sse41_phminposuw>;
4630 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4631 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4632 Intrinsic IntId128, bit Is2Addr = 1> {
4633 let isCommutable = 1 in
4634 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4635 (ins VR128:$src1, VR128:$src2),
4637 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4638 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4639 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4640 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4641 (ins VR128:$src1, i128mem:$src2),
4643 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4644 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4646 (IntId128 VR128:$src1,
4647 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4650 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4651 let isCommutable = 0 in
4652 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4654 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4656 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4658 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4660 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4662 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4664 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4666 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4668 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4670 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4672 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4676 let Constraints = "$src1 = $dst" in {
4677 let isCommutable = 0 in
4678 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4679 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4680 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4681 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4682 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4683 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4684 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4685 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4686 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4687 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4688 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4691 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4692 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4693 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4694 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4696 /// SS48I_binop_rm - Simple SSE41 binary operator.
4697 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4698 ValueType OpVT, bit Is2Addr = 1> {
4699 let isCommutable = 1 in
4700 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4701 (ins VR128:$src1, VR128:$src2),
4703 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4704 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4705 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4707 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4708 (ins VR128:$src1, i128mem:$src2),
4710 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4711 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4712 [(set VR128:$dst, (OpNode VR128:$src1,
4713 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4717 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4718 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4719 let Constraints = "$src1 = $dst" in
4720 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4722 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4723 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4724 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4725 X86MemOperand x86memop, bit Is2Addr = 1> {
4726 let isCommutable = 1 in
4727 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4728 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4730 !strconcat(OpcodeStr,
4731 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4732 !strconcat(OpcodeStr,
4733 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4734 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4736 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4737 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4739 !strconcat(OpcodeStr,
4740 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4741 !strconcat(OpcodeStr,
4742 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4745 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4749 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4750 let isCommutable = 0 in {
4751 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4752 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4753 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4754 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4755 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4756 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4757 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4758 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4759 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4760 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4761 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4762 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4764 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4765 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4766 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4767 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4768 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4769 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4772 let Constraints = "$src1 = $dst" in {
4773 let isCommutable = 0 in {
4774 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4775 VR128, memopv16i8, i128mem>;
4776 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4777 VR128, memopv16i8, i128mem>;
4778 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4779 VR128, memopv16i8, i128mem>;
4780 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4781 VR128, memopv16i8, i128mem>;
4783 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4784 VR128, memopv16i8, i128mem>;
4785 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4786 VR128, memopv16i8, i128mem>;
4789 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4790 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4791 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4792 RegisterClass RC, X86MemOperand x86memop,
4793 PatFrag mem_frag, Intrinsic IntId> {
4794 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4795 (ins RC:$src1, RC:$src2, RC:$src3),
4796 !strconcat(OpcodeStr,
4797 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4798 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4799 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4801 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4802 (ins RC:$src1, x86memop:$src2, RC:$src3),
4803 !strconcat(OpcodeStr,
4804 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4806 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4808 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4812 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4813 memopv16i8, int_x86_sse41_blendvpd>;
4814 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4815 memopv16i8, int_x86_sse41_blendvps>;
4816 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4817 memopv16i8, int_x86_sse41_pblendvb>;
4818 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4819 memopv32i8, int_x86_avx_blendv_pd_256>;
4820 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4821 memopv32i8, int_x86_avx_blendv_ps_256>;
4823 /// SS41I_ternary_int - SSE 4.1 ternary operator
4824 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4825 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4826 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4827 (ins VR128:$src1, VR128:$src2),
4828 !strconcat(OpcodeStr,
4829 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4830 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4833 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4834 (ins VR128:$src1, i128mem:$src2),
4835 !strconcat(OpcodeStr,
4836 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4839 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4843 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4844 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4845 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4847 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4848 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4849 "vmovntdqa\t{$src, $dst|$dst, $src}",
4850 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4852 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4853 "movntdqa\t{$src, $dst|$dst, $src}",
4854 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4857 //===----------------------------------------------------------------------===//
4858 // SSE4.2 - Compare Instructions
4859 //===----------------------------------------------------------------------===//
4861 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4862 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4863 Intrinsic IntId128, bit Is2Addr = 1> {
4864 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4865 (ins VR128:$src1, VR128:$src2),
4867 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4868 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4869 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4871 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4872 (ins VR128:$src1, i128mem:$src2),
4874 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4875 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4877 (IntId128 VR128:$src1,
4878 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4881 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4882 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4884 let Constraints = "$src1 = $dst" in
4885 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4887 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4888 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4889 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4890 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4892 //===----------------------------------------------------------------------===//
4893 // SSE4.2 - String/text Processing Instructions
4894 //===----------------------------------------------------------------------===//
4896 // Packed Compare Implicit Length Strings, Return Mask
4897 multiclass pseudo_pcmpistrm<string asm> {
4898 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4899 (ins VR128:$src1, VR128:$src2, i8imm:$src3), !strconcat(asm, "rr PSEUDO"),
4900 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4902 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4903 (ins VR128:$src1, i128mem:$src2, i8imm:$src3), !strconcat(asm, "rm PSEUDO"),
4904 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4905 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4908 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4909 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4910 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4913 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4914 Predicates = [HasAVX] in {
4915 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4916 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4917 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4918 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4919 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4920 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4923 let Defs = [XMM0, EFLAGS] in {
4924 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4925 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4926 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4927 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4928 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4929 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4932 // Packed Compare Explicit Length Strings, Return Mask
4933 multiclass pseudo_pcmpestrm<string asm> {
4934 def REG : Ii8<0, Pseudo, (outs VR128:$dst),
4935 (ins VR128:$src1, VR128:$src3, i8imm:$src5), !strconcat(asm, "rr PSEUDO"),
4936 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4937 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4938 def MEM : Ii8<0, Pseudo, (outs VR128:$dst),
4939 (ins VR128:$src1, i128mem:$src3, i8imm:$src5), !strconcat(asm, "rm PSEUDO"),
4940 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4941 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4944 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4945 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4946 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4949 let isAsmParserOnly = 1, Predicates = [HasAVX],
4950 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4951 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4952 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4953 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4954 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4955 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4956 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4959 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4960 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4961 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4962 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4963 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4964 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4965 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4968 // Packed Compare Implicit Length Strings, Return Index
4969 let Defs = [ECX, EFLAGS] in {
4970 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4971 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4972 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4973 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4974 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4975 (implicit EFLAGS)]>, OpSize;
4976 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4977 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4978 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4979 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4980 (implicit EFLAGS)]>, OpSize;
4984 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4985 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4987 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4989 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4991 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4993 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4995 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4999 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5000 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5001 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5002 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5003 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5004 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5006 // Packed Compare Explicit Length Strings, Return Index
5007 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5008 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5009 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5010 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5011 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5012 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5013 (implicit EFLAGS)]>, OpSize;
5014 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5015 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5016 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5018 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5019 (implicit EFLAGS)]>, OpSize;
5023 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5024 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5026 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5028 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5030 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5032 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5034 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5038 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5039 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5040 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5041 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5042 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5043 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5045 //===----------------------------------------------------------------------===//
5046 // SSE4.2 - CRC Instructions
5047 //===----------------------------------------------------------------------===//
5049 // No CRC instructions have AVX equivalents
5051 // crc intrinsic instruction
5052 // This set of instructions are only rm, the only difference is the size
5054 let Constraints = "$src1 = $dst" in {
5055 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5056 (ins GR32:$src1, i8mem:$src2),
5057 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5059 (int_x86_sse42_crc32_8 GR32:$src1,
5060 (load addr:$src2)))]>;
5061 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5062 (ins GR32:$src1, GR8:$src2),
5063 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5065 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5066 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5067 (ins GR32:$src1, i16mem:$src2),
5068 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5070 (int_x86_sse42_crc32_16 GR32:$src1,
5071 (load addr:$src2)))]>,
5073 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5074 (ins GR32:$src1, GR16:$src2),
5075 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5077 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5079 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5080 (ins GR32:$src1, i32mem:$src2),
5081 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5083 (int_x86_sse42_crc32_32 GR32:$src1,
5084 (load addr:$src2)))]>;
5085 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5086 (ins GR32:$src1, GR32:$src2),
5087 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5089 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5090 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5091 (ins GR64:$src1, i8mem:$src2),
5092 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5094 (int_x86_sse42_crc64_8 GR64:$src1,
5095 (load addr:$src2)))]>,
5097 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5098 (ins GR64:$src1, GR8:$src2),
5099 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5101 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5103 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5104 (ins GR64:$src1, i64mem:$src2),
5105 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5107 (int_x86_sse42_crc64_64 GR64:$src1,
5108 (load addr:$src2)))]>,
5110 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5111 (ins GR64:$src1, GR64:$src2),
5112 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5114 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5118 //===----------------------------------------------------------------------===//
5119 // AES-NI Instructions
5120 //===----------------------------------------------------------------------===//
5122 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5123 Intrinsic IntId128, bit Is2Addr = 1> {
5124 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5125 (ins VR128:$src1, VR128:$src2),
5127 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5128 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5129 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5131 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5132 (ins VR128:$src1, i128mem:$src2),
5134 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5135 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5137 (IntId128 VR128:$src1,
5138 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5141 // Perform One Round of an AES Encryption/Decryption Flow
5142 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5143 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5144 int_x86_aesni_aesenc, 0>, VEX_4V;
5145 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5146 int_x86_aesni_aesenclast, 0>, VEX_4V;
5147 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5148 int_x86_aesni_aesdec, 0>, VEX_4V;
5149 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5150 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5153 let Constraints = "$src1 = $dst" in {
5154 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5155 int_x86_aesni_aesenc>;
5156 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5157 int_x86_aesni_aesenclast>;
5158 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5159 int_x86_aesni_aesdec>;
5160 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5161 int_x86_aesni_aesdeclast>;
5164 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5165 (AESENCrr VR128:$src1, VR128:$src2)>;
5166 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5167 (AESENCrm VR128:$src1, addr:$src2)>;
5168 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5169 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5170 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5171 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5172 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5173 (AESDECrr VR128:$src1, VR128:$src2)>;
5174 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5175 (AESDECrm VR128:$src1, addr:$src2)>;
5176 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5177 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5178 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5179 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5181 // Perform the AES InvMixColumn Transformation
5182 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5183 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5185 "vaesimc\t{$src1, $dst|$dst, $src1}",
5187 (int_x86_aesni_aesimc VR128:$src1))]>,
5189 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5190 (ins i128mem:$src1),
5191 "vaesimc\t{$src1, $dst|$dst, $src1}",
5193 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5196 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5198 "aesimc\t{$src1, $dst|$dst, $src1}",
5200 (int_x86_aesni_aesimc VR128:$src1))]>,
5202 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5203 (ins i128mem:$src1),
5204 "aesimc\t{$src1, $dst|$dst, $src1}",
5206 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5209 // AES Round Key Generation Assist
5210 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5211 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5212 (ins VR128:$src1, i8imm:$src2),
5213 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5215 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5217 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5218 (ins i128mem:$src1, i8imm:$src2),
5219 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5221 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5225 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5226 (ins VR128:$src1, i8imm:$src2),
5227 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5229 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5231 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5232 (ins i128mem:$src1, i8imm:$src2),
5233 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5235 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5239 //===----------------------------------------------------------------------===//
5240 // CLMUL Instructions
5241 //===----------------------------------------------------------------------===//
5243 // Only the AVX version of CLMUL instructions are described here.
5245 // Carry-less Multiplication instructions
5246 let isAsmParserOnly = 1 in {
5247 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5248 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5249 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5252 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5253 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5254 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5258 multiclass avx_vpclmul<string asm> {
5259 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5260 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5263 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5264 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5267 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5268 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5269 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5270 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5272 } // isAsmParserOnly
5274 //===----------------------------------------------------------------------===//
5276 //===----------------------------------------------------------------------===//
5278 let isAsmParserOnly = 1 in {
5280 // Load from memory and broadcast to all elements of the destination operand
5281 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5282 X86MemOperand x86memop, Intrinsic Int> :
5283 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5284 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5285 [(set RC:$dst, (Int addr:$src))]>, VEX;
5287 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5288 int_x86_avx_vbroadcastss>;
5289 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5290 int_x86_avx_vbroadcastss_256>;
5291 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5292 int_x86_avx_vbroadcast_sd_256>;
5293 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5294 int_x86_avx_vbroadcastf128_pd_256>;
5296 // Insert packed floating-point values
5297 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5298 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5299 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5301 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5302 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5303 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5306 // Extract packed floating-point values
5307 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5308 (ins VR256:$src1, i8imm:$src2),
5309 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5311 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5312 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5313 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5316 // Conditional SIMD Packed Loads and Stores
5317 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5318 Intrinsic IntLd, Intrinsic IntLd256,
5319 Intrinsic IntSt, Intrinsic IntSt256,
5320 PatFrag pf128, PatFrag pf256> {
5321 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5322 (ins VR128:$src1, f128mem:$src2),
5323 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5324 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5326 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5327 (ins VR256:$src1, f256mem:$src2),
5328 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5329 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5331 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5332 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5333 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5334 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5335 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5336 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5337 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5338 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5341 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5342 int_x86_avx_maskload_ps,
5343 int_x86_avx_maskload_ps_256,
5344 int_x86_avx_maskstore_ps,
5345 int_x86_avx_maskstore_ps_256,
5346 memopv4f32, memopv8f32>;
5347 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5348 int_x86_avx_maskload_pd,
5349 int_x86_avx_maskload_pd_256,
5350 int_x86_avx_maskstore_pd,
5351 int_x86_avx_maskstore_pd_256,
5352 memopv2f64, memopv4f64>;
5354 // Permute Floating-Point Values
5355 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5356 RegisterClass RC, X86MemOperand x86memop_f,
5357 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5358 Intrinsic IntVar, Intrinsic IntImm> {
5359 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5360 (ins RC:$src1, RC:$src2),
5361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5362 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5363 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5364 (ins RC:$src1, x86memop_i:$src2),
5365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5366 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5368 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5369 (ins RC:$src1, i8imm:$src2),
5370 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5371 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5372 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5373 (ins x86memop_f:$src1, i8imm:$src2),
5374 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5375 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5378 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5379 memopv4f32, memopv4i32,
5380 int_x86_avx_vpermilvar_ps,
5381 int_x86_avx_vpermil_ps>;
5382 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5383 memopv8f32, memopv8i32,
5384 int_x86_avx_vpermilvar_ps_256,
5385 int_x86_avx_vpermil_ps_256>;
5386 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5387 memopv2f64, memopv2i64,
5388 int_x86_avx_vpermilvar_pd,
5389 int_x86_avx_vpermil_pd>;
5390 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5391 memopv4f64, memopv4i64,
5392 int_x86_avx_vpermilvar_pd_256,
5393 int_x86_avx_vpermil_pd_256>;
5395 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5396 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5397 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5399 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5400 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5401 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5404 // Zero All YMM registers
5405 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5406 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5408 // Zero Upper bits of YMM registers
5409 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5410 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5412 } // isAsmParserOnly
5414 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5415 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5416 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5417 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5418 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5419 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5421 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5422 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5423 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5424 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5425 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5426 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5428 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5429 (VBROADCASTF128 addr:$src)>;
5431 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5432 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5433 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5434 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5435 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5436 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5438 def : Pat<(int_x86_avx_vperm2f128_ps_256
5439 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5440 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5441 def : Pat<(int_x86_avx_vperm2f128_pd_256
5442 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5443 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5444 def : Pat<(int_x86_avx_vperm2f128_si_256
5445 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5446 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5448 //===----------------------------------------------------------------------===//
5449 // SSE Shuffle pattern fragments
5450 //===----------------------------------------------------------------------===//
5452 // This is part of a "work in progress" refactoring. The idea is that all
5453 // vector shuffles are going to be translated into target specific nodes and
5454 // directly matched by the patterns below (which can be changed along the way)
5455 // The AVX version of some but not all of them are described here, and more
5456 // should come in a near future.
5458 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5459 // SSE2 loads, which are always promoted to v2i64. The last one should match
5460 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5461 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5462 // we investigate further.
5463 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5465 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5466 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5468 (PSHUFDmi addr:$src1, imm:$imm)>;
5469 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5471 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5473 // Shuffle with PSHUFD instruction.
5474 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5475 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5476 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5477 (PSHUFDri VR128:$src1, imm:$imm)>;
5479 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5480 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5481 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5482 (PSHUFDri VR128:$src1, imm:$imm)>;
5484 // Shuffle with SHUFPD instruction.
5485 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5486 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5487 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5488 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5489 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5490 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5492 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5493 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5494 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5495 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5497 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5498 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5499 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5500 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5502 // Shuffle with SHUFPS instruction.
5503 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5504 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5505 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5506 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5507 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5508 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5510 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5511 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5512 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5513 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5515 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5516 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5517 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5518 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5519 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5520 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5522 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5523 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5524 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5525 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5527 // Shuffle with MOVHLPS instruction
5528 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5529 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5530 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5531 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5533 // Shuffle with MOVDDUP instruction
5534 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5535 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5536 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5537 (MOVDDUPrm addr:$src)>;
5539 def : Pat<(X86Movddup (bc_v4f32 (memopv2f64 addr:$src))),
5540 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5541 def : Pat<(X86Movddup (bc_v4f32 (memopv2f64 addr:$src))),
5542 (MOVDDUPrm addr:$src)>;
5544 def : Pat<(X86Movddup (memopv2i64 addr:$src)),
5545 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5546 def : Pat<(X86Movddup (memopv2i64 addr:$src)),
5547 (MOVDDUPrm addr:$src)>;
5549 def : Pat<(X86Movddup (bc_v4i32 (memopv2i64 addr:$src))),
5550 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5551 def : Pat<(X86Movddup (bc_v4i32 (memopv2i64 addr:$src))),
5552 (MOVDDUPrm addr:$src)>;
5554 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5555 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5556 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5557 (MOVDDUPrm addr:$src)>;
5559 def : Pat<(X86Movddup (bc_v2f64
5560 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5561 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5562 def : Pat<(X86Movddup (bc_v2f64
5563 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5564 (MOVDDUPrm addr:$src)>;
5566 // Shuffle with UNPCKLPS
5567 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5568 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5569 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5570 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5572 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5573 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5574 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5575 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5577 // Shuffle with UNPCKHPS
5578 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5579 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5580 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5581 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5583 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5584 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5585 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5586 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5588 // Shuffle with UNPCKLPD
5589 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5590 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5591 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5592 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5594 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5595 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5596 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5597 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5599 // Shuffle with UNPCKHPD
5600 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5601 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5602 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5603 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5605 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5606 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5607 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5608 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5610 // Shuffle with PUNPCKLBW
5611 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5612 (bc_v16i8 (memopv2i64 addr:$src2)))),
5613 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5614 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5615 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5617 // Shuffle with PUNPCKLWD
5618 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5619 (bc_v8i16 (memopv2i64 addr:$src2)))),
5620 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5621 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5622 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5624 // Shuffle with PUNPCKLDQ
5625 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5626 (bc_v4i32 (memopv2i64 addr:$src2)))),
5627 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5628 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5629 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5631 // Shuffle with PUNPCKLQDQ
5632 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5633 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5634 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5635 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5637 // Shuffle with PUNPCKHBW
5638 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5639 (bc_v16i8 (memopv2i64 addr:$src2)))),
5640 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5641 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5642 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5644 // Shuffle with PUNPCKHWD
5645 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5646 (bc_v8i16 (memopv2i64 addr:$src2)))),
5647 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5648 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5649 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5651 // Shuffle with PUNPCKHDQ
5652 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5653 (bc_v4i32 (memopv2i64 addr:$src2)))),
5654 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5655 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5656 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5658 // Shuffle with PUNPCKHQDQ
5659 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5660 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5661 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5662 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5664 // Shuffle with MOVLHPS
5665 def : Pat<(X86Movlhps VR128:$src1,
5666 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5667 (MOVHPSrm VR128:$src1, addr:$src2)>;
5668 def : Pat<(X86Movlhps VR128:$src1,
5669 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5670 (MOVHPSrm VR128:$src1, addr:$src2)>;
5671 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5672 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5673 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5674 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5675 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5676 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5677 // FIXME: Instead of X86Movddup, there should be a X86Movlhps here, the problem
5678 // is during lowering, where it's not possible to recognize the load fold cause
5679 // it has two uses through a bitcast. One use disappears at isel time and the
5680 // fold opportunity reappears.
5681 def : Pat<(v2i64 (X86Movddup VR128:$src)),
5682 (MOVLHPSrr VR128:$src, VR128:$src)>;
5683 def : Pat<(v4f32 (X86Movddup VR128:$src)),
5684 (MOVLHPSrr VR128:$src, VR128:$src)>;
5685 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5686 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5688 // Shuffle with MOVLHPD
5689 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5690 (scalar_to_vector (loadf64 addr:$src2)))),
5691 (MOVHPDrm VR128:$src1, addr:$src2)>;
5692 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5693 // is during lowering, where it's not possible to recognize the load fold cause
5694 // it has two uses through a bitcast. One use disappears at isel time and the
5695 // fold opportunity reappears.
5696 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5697 (scalar_to_vector (loadf64 addr:$src2)))),
5698 (MOVHPDrm VR128:$src1, addr:$src2)>;
5700 // Shuffle with MOVSS
5701 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5702 (MOVSSrr VR128:$src1, FR32:$src2)>;
5703 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5704 (MOVSSrr (v4i32 VR128:$src1),
5705 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5706 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5707 (MOVSSrr (v4f32 VR128:$src1),
5708 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5709 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5710 // is during lowering, where it's not possible to recognize the load fold cause
5711 // it has two uses through a bitcast. One use disappears at isel time and the
5712 // fold opportunity reappears.
5713 def : Pat<(X86Movss VR128:$src1,
5714 (bc_v4i32 (v2i64 (load addr:$src2)))),
5715 (MOVLPSrm VR128:$src1, addr:$src2)>;
5717 // Shuffle with MOVSD
5718 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5719 (MOVSDrr VR128:$src1, FR64:$src2)>;
5720 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5721 (MOVSDrr (v2i64 VR128:$src1),
5722 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5723 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5724 (MOVSDrr (v2f64 VR128:$src1),
5725 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5726 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5727 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5728 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5729 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5731 // Shuffle with MOVSHDUP
5732 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5733 (MOVSHDUPrr VR128:$src)>;
5734 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5735 (MOVSHDUPrm addr:$src)>;
5737 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5738 (MOVSHDUPrr VR128:$src)>;
5739 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5740 (MOVSHDUPrm addr:$src)>;
5742 // Shuffle with MOVSLDUP
5743 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5744 (MOVSLDUPrr VR128:$src)>;
5745 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5746 (MOVSLDUPrm addr:$src)>;
5748 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5749 (MOVSLDUPrr VR128:$src)>;
5750 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5751 (MOVSLDUPrm addr:$src)>;
5753 // Shuffle with PSHUFHW
5754 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5755 (PSHUFHWri VR128:$src, imm:$imm)>;
5756 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5757 (PSHUFHWmi addr:$src, imm:$imm)>;
5759 // Shuffle with PSHUFLW
5760 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5761 (PSHUFLWri VR128:$src, imm:$imm)>;
5762 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5763 (PSHUFLWmi addr:$src, imm:$imm)>;
5765 // Shuffle with PALIGN
5766 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5767 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5768 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5769 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5770 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5771 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5772 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5773 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5775 // Shuffle with MOVLPS
5776 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5777 (MOVLPSrm VR128:$src1, addr:$src2)>;
5778 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5779 (MOVLPSrm VR128:$src1, addr:$src2)>;
5780 def : Pat<(X86Movlps VR128:$src1,
5781 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5782 (MOVLPSrm VR128:$src1, addr:$src2)>;
5784 // Shuffle with MOVLPD
5785 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5786 (MOVLPDrm VR128:$src1, addr:$src2)>;
5787 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5788 (MOVLPDrm VR128:$src1, addr:$src2)>;
5789 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5790 (scalar_to_vector (loadf64 addr:$src2)))),
5791 (MOVLPDrm VR128:$src1, addr:$src2)>;
5793 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5794 def : Pat<(store (f64 (vector_extract
5795 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5796 (MOVHPSmr addr:$dst, VR128:$src)>;
5797 def : Pat<(store (f64 (vector_extract
5798 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5799 (MOVHPDmr addr:$dst, VR128:$src)>;
5801 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5802 (MOVLPSmr addr:$src1, VR128:$src2)>;
5803 def : Pat<(store (v4i32 (X86Movlps
5804 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5805 (MOVLPSmr addr:$src1, VR128:$src2)>;
5807 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5808 (MOVLPDmr addr:$src1, VR128:$src2)>;
5809 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5810 (MOVLPDmr addr:$src1, VR128:$src2)>;