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 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // SSE 1 & 2 - Move Instructions
120 //===----------------------------------------------------------------------===//
122 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
123 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
124 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
126 // Loading from memory automatically zeroing upper bits.
127 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
128 PatFrag mem_pat, string OpcodeStr> :
129 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
131 [(set RC:$dst, (mem_pat addr:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
139 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
140 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
141 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
143 let canFoldAsLoad = 1, isReMaterializable = 1 in {
144 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
146 let AddedComplexity = 20 in
147 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
150 let Constraints = "$src1 = $dst" in {
151 def MOVSSrr : sse12_move_rr<FR32, v4f32,
152 "movss\t{$src2, $dst|$dst, $src2}">, XS;
153 def MOVSDrr : sse12_move_rr<FR64, v2f64,
154 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
157 let canFoldAsLoad = 1, isReMaterializable = 1 in {
158 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
160 let AddedComplexity = 20 in
161 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
164 let AddedComplexity = 15 in {
165 // Extract the low 32-bit value from one vector and insert it into another.
166 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
167 (MOVSSrr (v4f32 VR128:$src1),
168 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
169 // Extract the low 64-bit value from one vector and insert it into another.
170 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
171 (MOVSDrr (v2f64 VR128:$src1),
172 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
175 // Implicitly promote a 32-bit scalar to a vector.
176 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
177 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
178 // Implicitly promote a 64-bit scalar to a vector.
179 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
180 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
181 // Implicitly promote a 32-bit scalar to a vector.
182 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
183 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
184 // Implicitly promote a 64-bit scalar to a vector.
185 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
186 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
188 let AddedComplexity = 20 in {
189 // MOVSSrm zeros the high parts of the register; represent this
190 // with SUBREG_TO_REG.
191 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
192 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
193 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 // MOVSDrm zeros the high parts of the register; represent this
198 // with SUBREG_TO_REG.
199 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
200 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
201 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzload addr:$src)),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
211 // Store scalar value to memory.
212 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
213 "movss\t{$src, $dst|$dst, $src}",
214 [(store FR32:$src, addr:$dst)]>;
215 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
216 "movsd\t{$src, $dst|$dst, $src}",
217 [(store FR64:$src, addr:$dst)]>;
219 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
220 "movss\t{$src, $dst|$dst, $src}",
221 [(store FR32:$src, addr:$dst)]>, XS, VEX;
222 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
223 "movsd\t{$src, $dst|$dst, $src}",
224 [(store FR64:$src, addr:$dst)]>, XD, VEX;
226 // Extract and store.
227 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
230 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
231 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
236 // Move Aligned/Unaligned floating point values
237 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
238 X86MemOperand x86memop, PatFrag ld_frag,
239 string asm, Domain d,
240 bit IsReMaterializable = 1> {
241 let neverHasSideEffects = 1 in
242 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
243 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
244 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
245 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
246 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
247 [(set RC:$dst, (ld_frag addr:$src))], d>;
250 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
251 "movaps", SSEPackedSingle>, VEX;
252 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
253 "movapd", SSEPackedDouble>, OpSize, VEX;
254 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
255 "movups", SSEPackedSingle>, VEX;
256 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
257 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
259 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
260 "movaps", SSEPackedSingle>, VEX;
261 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
262 "movapd", SSEPackedDouble>, OpSize, VEX;
263 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
264 "movups", SSEPackedSingle>, VEX;
265 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
266 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
267 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
268 "movaps", SSEPackedSingle>, TB;
269 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
270 "movapd", SSEPackedDouble>, TB, OpSize;
271 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
272 "movups", SSEPackedSingle>, TB;
273 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
274 "movupd", SSEPackedDouble, 0>, TB, OpSize;
276 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
277 "movaps\t{$src, $dst|$dst, $src}",
278 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
279 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
280 "movapd\t{$src, $dst|$dst, $src}",
281 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
282 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
283 "movups\t{$src, $dst|$dst, $src}",
284 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
285 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
286 "movupd\t{$src, $dst|$dst, $src}",
287 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
288 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
289 "movaps\t{$src, $dst|$dst, $src}",
290 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
291 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
292 "movapd\t{$src, $dst|$dst, $src}",
293 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
294 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
295 "movups\t{$src, $dst|$dst, $src}",
296 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
297 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
298 "movupd\t{$src, $dst|$dst, $src}",
299 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
302 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
303 (VMOVUPSYmr addr:$dst, VR256:$src)>;
305 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
306 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
307 (VMOVUPDYmr addr:$dst, VR256:$src)>;
309 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
310 "movaps\t{$src, $dst|$dst, $src}",
311 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
312 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
313 "movapd\t{$src, $dst|$dst, $src}",
314 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
315 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
316 "movups\t{$src, $dst|$dst, $src}",
317 [(store (v4f32 VR128:$src), addr:$dst)]>;
318 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movupd\t{$src, $dst|$dst, $src}",
320 [(store (v2f64 VR128:$src), addr:$dst)]>;
322 // Intrinsic forms of MOVUPS/D load and store
323 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
324 (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
327 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
328 (ins f128mem:$dst, VR128:$src),
329 "movupd\t{$src, $dst|$dst, $src}",
330 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
332 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
333 "movups\t{$src, $dst|$dst, $src}",
334 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
335 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
336 "movupd\t{$src, $dst|$dst, $src}",
337 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
339 // Move Low/High packed floating point values
340 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
341 PatFrag mov_frag, string base_opc,
343 def PSrm : PI<opc, MRMSrcMem,
344 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
345 !strconcat(base_opc, "s", asm_opr),
348 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
349 SSEPackedSingle>, TB;
351 def PDrm : PI<opc, MRMSrcMem,
352 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
353 !strconcat(base_opc, "d", asm_opr),
354 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
355 (scalar_to_vector (loadf64 addr:$src2)))))],
356 SSEPackedDouble>, TB, OpSize;
359 let AddedComplexity = 20 in {
360 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
361 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
362 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
363 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
365 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
366 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
367 "\t{$src2, $dst|$dst, $src2}">;
368 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
369 "\t{$src2, $dst|$dst, $src2}">;
372 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
373 "movlps\t{$src, $dst|$dst, $src}",
374 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
375 (iPTR 0))), addr:$dst)]>, VEX;
376 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
377 "movlpd\t{$src, $dst|$dst, $src}",
378 [(store (f64 (vector_extract (v2f64 VR128:$src),
379 (iPTR 0))), addr:$dst)]>, VEX;
380 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
381 "movlps\t{$src, $dst|$dst, $src}",
382 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
383 (iPTR 0))), addr:$dst)]>;
384 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
385 "movlpd\t{$src, $dst|$dst, $src}",
386 [(store (f64 (vector_extract (v2f64 VR128:$src),
387 (iPTR 0))), addr:$dst)]>;
389 // v2f64 extract element 1 is always custom lowered to unpack high to low
390 // and extract element 0 so the non-store version isn't too horrible.
391 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
392 "movhps\t{$src, $dst|$dst, $src}",
393 [(store (f64 (vector_extract
394 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
395 (undef)), (iPTR 0))), addr:$dst)]>,
397 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
398 "movhpd\t{$src, $dst|$dst, $src}",
399 [(store (f64 (vector_extract
400 (v2f64 (unpckh VR128:$src, (undef))),
401 (iPTR 0))), addr:$dst)]>,
403 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movhps\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract
406 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
407 (undef)), (iPTR 0))), addr:$dst)]>;
408 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movhpd\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract
411 (v2f64 (unpckh VR128:$src, (undef))),
412 (iPTR 0))), addr:$dst)]>;
414 let AddedComplexity = 20 in {
415 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
416 (ins VR128:$src1, VR128:$src2),
417 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
419 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
421 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
422 (ins VR128:$src1, VR128:$src2),
423 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
425 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
428 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
429 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
430 (ins VR128:$src1, VR128:$src2),
431 "movlhps\t{$src2, $dst|$dst, $src2}",
433 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
434 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
435 (ins VR128:$src1, VR128:$src2),
436 "movhlps\t{$src2, $dst|$dst, $src2}",
438 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
441 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
442 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
443 let AddedComplexity = 20 in {
444 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
445 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
446 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
447 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
450 //===----------------------------------------------------------------------===//
451 // SSE 1 & 2 - Conversion Instructions
452 //===----------------------------------------------------------------------===//
454 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
455 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
457 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
458 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
460 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
463 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
464 X86MemOperand x86memop, string asm> {
465 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
467 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
471 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
472 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
473 string asm, Domain d> {
474 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
475 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
476 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
477 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
480 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
481 X86MemOperand x86memop, string asm> {
482 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
483 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
484 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
485 (ins DstRC:$src1, x86memop:$src),
486 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
489 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
490 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
491 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
492 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
494 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
495 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
496 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
497 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
500 // The assembler can recognize rr 64-bit instructions by seeing a rxx
501 // register, but the same isn't true when only using memory operands,
502 // provide other assembly "l" and "q" forms to address this explicitly
503 // where appropriate to do so.
504 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
506 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
508 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
510 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
512 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
515 let Predicates = [HasAVX] in {
516 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
517 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
518 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
519 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
520 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
521 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
522 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
523 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
525 def : Pat<(f32 (sint_to_fp GR32:$src)),
526 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
527 def : Pat<(f32 (sint_to_fp GR64:$src)),
528 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
529 def : Pat<(f64 (sint_to_fp GR32:$src)),
530 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
531 def : Pat<(f64 (sint_to_fp GR64:$src)),
532 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
535 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
536 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
537 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
538 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
539 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
540 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
541 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
542 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
543 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
544 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
545 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
546 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
547 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
548 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
549 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
550 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
552 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
553 // and/or XMM operand(s).
555 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
556 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
558 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
559 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
560 [(set DstRC:$dst, (Int SrcRC:$src))]>;
561 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
562 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
563 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
566 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
567 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
568 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
569 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
571 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
572 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
573 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
574 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
575 (ins DstRC:$src1, x86memop:$src2),
577 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
578 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
579 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
582 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
583 f32mem, load, "cvtss2si">, XS, VEX;
584 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
585 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
587 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
588 f128mem, load, "cvtsd2si">, XD, VEX;
589 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
590 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
593 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
594 // Get rid of this hack or rename the intrinsics, there are several
595 // intructions that only match with the intrinsic form, why create duplicates
596 // to let them be recognized by the assembler?
597 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
598 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
599 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
600 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
601 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
602 f32mem, load, "cvtss2si">, XS;
603 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
604 f32mem, load, "cvtss2si{q}">, XS, REX_W;
605 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
606 f128mem, load, "cvtsd2si{l}">, XD;
607 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
608 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
611 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
612 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
613 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
614 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
616 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
617 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
618 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
619 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
622 let Constraints = "$src1 = $dst" in {
623 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
624 int_x86_sse_cvtsi2ss, i32mem, loadi32,
626 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
627 int_x86_sse_cvtsi642ss, i64mem, loadi64,
628 "cvtsi2ss{q}">, XS, REX_W;
629 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
630 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
632 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
633 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
634 "cvtsi2sd">, XD, REX_W;
639 // Aliases for intrinsics
640 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
641 f32mem, load, "cvttss2si">, XS, VEX;
642 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
643 int_x86_sse_cvttss2si64, f32mem, load,
644 "cvttss2si">, XS, VEX, VEX_W;
645 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
646 f128mem, load, "cvttsd2si">, XD, VEX;
647 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
648 int_x86_sse2_cvttsd2si64, f128mem, load,
649 "cvttsd2si">, XD, VEX, VEX_W;
650 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
651 f32mem, load, "cvttss2si">, XS;
652 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
653 int_x86_sse_cvttss2si64, f32mem, load,
654 "cvttss2si{q}">, XS, REX_W;
655 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
656 f128mem, load, "cvttsd2si">, XD;
657 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
658 int_x86_sse2_cvttsd2si64, f128mem, load,
659 "cvttsd2si{q}">, XD, REX_W;
661 let Pattern = []<dag> in {
662 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
663 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
664 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
665 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
667 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
668 "cvtdq2ps\t{$src, $dst|$dst, $src}",
669 SSEPackedSingle>, TB, VEX;
670 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
671 "cvtdq2ps\t{$src, $dst|$dst, $src}",
672 SSEPackedSingle>, TB, VEX;
674 let Pattern = []<dag> in {
675 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
676 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
677 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
678 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
679 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
680 "cvtdq2ps\t{$src, $dst|$dst, $src}",
681 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
686 // Convert scalar double to scalar single
687 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
688 (ins FR64:$src1, FR64:$src2),
689 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
691 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
692 (ins FR64:$src1, f64mem:$src2),
693 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
694 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
695 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
698 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
699 "cvtsd2ss\t{$src, $dst|$dst, $src}",
700 [(set FR32:$dst, (fround FR64:$src))]>;
701 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
702 "cvtsd2ss\t{$src, $dst|$dst, $src}",
703 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
704 Requires<[HasSSE2, OptForSize]>;
706 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
707 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
709 let Constraints = "$src1 = $dst" in
710 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
711 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
713 // Convert scalar single to scalar double
714 // SSE2 instructions with XS prefix
715 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
716 (ins FR32:$src1, FR32:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 []>, XS, Requires<[HasAVX]>, VEX_4V;
719 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
720 (ins FR32:$src1, f32mem:$src2),
721 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
722 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
724 let Predicates = [HasAVX] in {
725 def : Pat<(f64 (fextend FR32:$src)),
726 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
727 def : Pat<(fextend (loadf32 addr:$src)),
728 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
729 def : Pat<(extloadf32 addr:$src),
730 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
733 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
734 "cvtss2sd\t{$src, $dst|$dst, $src}",
735 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
737 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
738 "cvtss2sd\t{$src, $dst|$dst, $src}",
739 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
740 Requires<[HasSSE2, OptForSize]>;
742 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
743 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
744 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
745 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
746 VR128:$src2))]>, XS, VEX_4V,
748 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
749 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
750 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
751 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
752 (load addr:$src2)))]>, XS, VEX_4V,
754 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
755 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
756 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
757 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
758 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
761 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
762 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
763 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
764 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
765 (load addr:$src2)))]>, XS,
769 def : Pat<(extloadf32 addr:$src),
770 (CVTSS2SDrr (MOVSSrm addr:$src))>,
771 Requires<[HasSSE2, OptForSpeed]>;
773 // Convert doubleword to packed single/double fp
774 // SSE2 instructions without OpSize prefix
775 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
776 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
777 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
778 TB, VEX, Requires<[HasAVX]>;
779 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
780 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
781 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
782 (bitconvert (memopv2i64 addr:$src))))]>,
783 TB, VEX, Requires<[HasAVX]>;
784 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
785 "cvtdq2ps\t{$src, $dst|$dst, $src}",
786 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
787 TB, Requires<[HasSSE2]>;
788 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
789 "cvtdq2ps\t{$src, $dst|$dst, $src}",
790 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
791 (bitconvert (memopv2i64 addr:$src))))]>,
792 TB, Requires<[HasSSE2]>;
794 // FIXME: why the non-intrinsic version is described as SSE3?
795 // SSE2 instructions with XS prefix
796 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
797 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
798 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
799 XS, VEX, Requires<[HasAVX]>;
800 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
801 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
802 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
803 (bitconvert (memopv2i64 addr:$src))))]>,
804 XS, VEX, Requires<[HasAVX]>;
805 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
806 "cvtdq2pd\t{$src, $dst|$dst, $src}",
807 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
808 XS, Requires<[HasSSE2]>;
809 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
810 "cvtdq2pd\t{$src, $dst|$dst, $src}",
811 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
812 (bitconvert (memopv2i64 addr:$src))))]>,
813 XS, Requires<[HasSSE2]>;
816 // Convert packed single/double fp to doubleword
817 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
818 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
819 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
820 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
821 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
822 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
823 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
824 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
825 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
826 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
827 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
828 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
830 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
831 "cvtps2dq\t{$src, $dst|$dst, $src}",
832 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
834 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
836 "cvtps2dq\t{$src, $dst|$dst, $src}",
837 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
838 (memop addr:$src)))]>, VEX;
839 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
840 "cvtps2dq\t{$src, $dst|$dst, $src}",
841 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
842 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
843 "cvtps2dq\t{$src, $dst|$dst, $src}",
844 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
845 (memop addr:$src)))]>;
847 // SSE2 packed instructions with XD prefix
848 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
849 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
850 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
851 XD, VEX, Requires<[HasAVX]>;
852 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
853 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
854 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
855 (memop addr:$src)))]>,
856 XD, VEX, Requires<[HasAVX]>;
857 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
858 "cvtpd2dq\t{$src, $dst|$dst, $src}",
859 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
860 XD, Requires<[HasSSE2]>;
861 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
862 "cvtpd2dq\t{$src, $dst|$dst, $src}",
863 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
864 (memop addr:$src)))]>,
865 XD, Requires<[HasSSE2]>;
868 // Convert with truncation packed single/double fp to doubleword
869 // SSE2 packed instructions with XS prefix
870 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
871 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
872 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
873 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
874 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
875 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
876 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
877 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
878 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
879 "cvttps2dq\t{$src, $dst|$dst, $src}",
881 (int_x86_sse2_cvttps2dq VR128:$src))]>;
882 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
883 "cvttps2dq\t{$src, $dst|$dst, $src}",
885 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
887 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
888 "vcvttps2dq\t{$src, $dst|$dst, $src}",
890 (int_x86_sse2_cvttps2dq VR128:$src))]>,
891 XS, VEX, Requires<[HasAVX]>;
892 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
893 "vcvttps2dq\t{$src, $dst|$dst, $src}",
894 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
895 (memop addr:$src)))]>,
896 XS, VEX, Requires<[HasAVX]>;
898 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
899 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
900 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
901 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
903 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
904 (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
905 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
906 (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
908 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
910 "cvttpd2dq\t{$src, $dst|$dst, $src}",
911 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
913 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
915 "cvttpd2dq\t{$src, $dst|$dst, $src}",
916 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
917 (memop addr:$src)))]>, VEX;
918 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
919 "cvttpd2dq\t{$src, $dst|$dst, $src}",
920 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
921 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
922 "cvttpd2dq\t{$src, $dst|$dst, $src}",
923 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
924 (memop addr:$src)))]>;
926 // The assembler can recognize rr 256-bit instructions by seeing a ymm
927 // register, but the same isn't true when using memory operands instead.
928 // Provide other assembly rr and rm forms to address this explicitly.
929 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
930 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
931 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
932 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
935 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
936 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
937 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
938 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
941 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
942 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
943 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
944 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
946 // Convert packed single to packed double
947 let Predicates = [HasAVX] in {
948 // SSE2 instructions without OpSize prefix
949 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
950 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
951 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
952 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
953 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
954 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
955 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
956 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
958 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
959 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
960 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
961 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
963 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
964 "vcvtps2pd\t{$src, $dst|$dst, $src}",
965 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
966 VEX, Requires<[HasAVX]>;
967 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
968 "vcvtps2pd\t{$src, $dst|$dst, $src}",
969 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
970 (load addr:$src)))]>,
971 VEX, Requires<[HasAVX]>;
972 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
973 "cvtps2pd\t{$src, $dst|$dst, $src}",
974 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
975 TB, Requires<[HasSSE2]>;
976 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
977 "cvtps2pd\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
979 (load addr:$src)))]>,
980 TB, Requires<[HasSSE2]>;
982 // Convert packed double to packed single
983 // The assembler can recognize rr 256-bit instructions by seeing a ymm
984 // register, but the same isn't true when using memory operands instead.
985 // Provide other assembly rr and rm forms to address this explicitly.
986 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
987 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
988 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
989 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
992 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
993 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
994 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
995 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
998 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
999 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1000 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1001 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1002 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1003 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1004 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1005 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1008 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1009 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1010 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1011 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1013 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1014 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1015 (memop addr:$src)))]>;
1016 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1017 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1018 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1019 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1020 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1021 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1022 (memop addr:$src)))]>;
1024 // AVX 256-bit register conversion intrinsics
1025 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1026 // whenever possible to avoid declaring two versions of each one.
1027 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1028 (VCVTDQ2PSYrr VR256:$src)>;
1029 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1030 (VCVTDQ2PSYrm addr:$src)>;
1032 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1033 (VCVTPD2PSYrr VR256:$src)>;
1034 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1035 (VCVTPD2PSYrm addr:$src)>;
1037 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1038 (VCVTPS2DQYrr VR256:$src)>;
1039 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1040 (VCVTPS2DQYrm addr:$src)>;
1042 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1043 (VCVTPS2PDYrr VR128:$src)>;
1044 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1045 (VCVTPS2PDYrm addr:$src)>;
1047 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1048 (VCVTTPD2DQYrr VR256:$src)>;
1049 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1050 (VCVTTPD2DQYrm addr:$src)>;
1052 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1053 (VCVTTPS2DQYrr VR256:$src)>;
1054 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1055 (VCVTTPS2DQYrm addr:$src)>;
1057 //===----------------------------------------------------------------------===//
1058 // SSE 1 & 2 - Compare Instructions
1059 //===----------------------------------------------------------------------===//
1061 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1062 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1063 string asm, string asm_alt> {
1064 let isAsmParserOnly = 1 in {
1065 def rr : SIi8<0xC2, MRMSrcReg,
1066 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1069 def rm : SIi8<0xC2, MRMSrcMem,
1070 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1074 // Accept explicit immediate argument form instead of comparison code.
1075 def rr_alt : SIi8<0xC2, MRMSrcReg,
1076 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1079 def rm_alt : SIi8<0xC2, MRMSrcMem,
1080 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1084 let neverHasSideEffects = 1 in {
1085 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1086 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1087 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1089 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1090 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1091 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1095 let Constraints = "$src1 = $dst" in {
1096 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1097 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1098 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1099 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1100 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1101 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1102 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1103 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1104 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1105 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1106 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1107 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1108 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1109 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1110 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1111 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1113 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1114 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1115 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1116 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1117 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1118 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1119 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1120 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1121 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1122 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1123 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1124 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1125 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1128 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1129 Intrinsic Int, string asm> {
1130 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1131 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1132 [(set VR128:$dst, (Int VR128:$src1,
1133 VR128:$src, imm:$cc))]>;
1134 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1135 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1136 [(set VR128:$dst, (Int VR128:$src1,
1137 (load addr:$src), imm:$cc))]>;
1140 // Aliases to match intrinsics which expect XMM operand(s).
1141 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1142 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1144 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1145 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1147 let Constraints = "$src1 = $dst" in {
1148 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1149 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1150 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1151 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1155 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1156 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1157 ValueType vt, X86MemOperand x86memop,
1158 PatFrag ld_frag, string OpcodeStr, Domain d> {
1159 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1160 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1161 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1162 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1163 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1164 [(set EFLAGS, (OpNode (vt RC:$src1),
1165 (ld_frag addr:$src2)))], d>;
1168 let Defs = [EFLAGS] in {
1169 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1170 "ucomiss", SSEPackedSingle>, VEX;
1171 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1172 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1173 let Pattern = []<dag> in {
1174 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1175 "comiss", SSEPackedSingle>, VEX;
1176 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1177 "comisd", SSEPackedDouble>, OpSize, VEX;
1180 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1181 load, "ucomiss", SSEPackedSingle>, VEX;
1182 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1183 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1185 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1186 load, "comiss", SSEPackedSingle>, VEX;
1187 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1188 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1189 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1190 "ucomiss", SSEPackedSingle>, TB;
1191 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1192 "ucomisd", SSEPackedDouble>, TB, OpSize;
1194 let Pattern = []<dag> in {
1195 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1196 "comiss", SSEPackedSingle>, TB;
1197 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1198 "comisd", SSEPackedDouble>, TB, OpSize;
1201 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1202 load, "ucomiss", SSEPackedSingle>, TB;
1203 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1204 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1206 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1207 "comiss", SSEPackedSingle>, TB;
1208 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1209 "comisd", SSEPackedDouble>, TB, OpSize;
1210 } // Defs = [EFLAGS]
1212 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1213 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1214 Intrinsic Int, string asm, string asm_alt,
1216 let isAsmParserOnly = 1 in {
1217 def rri : PIi8<0xC2, MRMSrcReg,
1218 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1219 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1220 def rmi : PIi8<0xC2, MRMSrcMem,
1221 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1222 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1225 // Accept explicit immediate argument form instead of comparison code.
1226 def rri_alt : PIi8<0xC2, MRMSrcReg,
1227 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1229 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1230 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1234 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1235 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1236 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1237 SSEPackedSingle>, VEX_4V;
1238 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1239 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1240 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1241 SSEPackedDouble>, OpSize, VEX_4V;
1242 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1243 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1244 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1245 SSEPackedSingle>, VEX_4V;
1246 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1247 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1248 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1249 SSEPackedDouble>, OpSize, VEX_4V;
1250 let Constraints = "$src1 = $dst" in {
1251 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1252 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1253 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1254 SSEPackedSingle>, TB;
1255 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1256 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1257 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1258 SSEPackedDouble>, TB, OpSize;
1261 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1262 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1263 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1264 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1265 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1266 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1267 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1268 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1270 //===----------------------------------------------------------------------===//
1271 // SSE 1 & 2 - Shuffle Instructions
1272 //===----------------------------------------------------------------------===//
1274 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1275 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1276 ValueType vt, string asm, PatFrag mem_frag,
1277 Domain d, bit IsConvertibleToThreeAddress = 0> {
1278 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1279 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1280 [(set RC:$dst, (vt (shufp:$src3
1281 RC:$src1, (mem_frag addr:$src2))))], d>;
1282 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1283 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1284 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1286 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1289 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1290 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1291 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1292 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1293 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1294 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1295 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1296 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1297 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1298 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1299 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1300 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1302 let Constraints = "$src1 = $dst" in {
1303 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1304 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1305 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1307 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1308 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1309 memopv2f64, SSEPackedDouble>, TB, OpSize;
1312 //===----------------------------------------------------------------------===//
1313 // SSE 1 & 2 - Unpack Instructions
1314 //===----------------------------------------------------------------------===//
1316 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1317 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1318 PatFrag mem_frag, RegisterClass RC,
1319 X86MemOperand x86memop, string asm,
1321 def rr : PI<opc, MRMSrcReg,
1322 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1324 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1325 def rm : PI<opc, MRMSrcMem,
1326 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1328 (vt (OpNode RC:$src1,
1329 (mem_frag addr:$src2))))], d>;
1332 let AddedComplexity = 10 in {
1333 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1334 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1335 SSEPackedSingle>, VEX_4V;
1336 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1337 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1338 SSEPackedDouble>, OpSize, VEX_4V;
1339 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1340 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1341 SSEPackedSingle>, VEX_4V;
1342 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1343 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1344 SSEPackedDouble>, OpSize, VEX_4V;
1346 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1347 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1348 SSEPackedSingle>, VEX_4V;
1349 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1350 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1351 SSEPackedDouble>, OpSize, VEX_4V;
1352 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1353 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1354 SSEPackedSingle>, VEX_4V;
1355 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1356 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1357 SSEPackedDouble>, OpSize, VEX_4V;
1359 let Constraints = "$src1 = $dst" in {
1360 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1361 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1362 SSEPackedSingle>, TB;
1363 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1364 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1365 SSEPackedDouble>, TB, OpSize;
1366 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1367 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1368 SSEPackedSingle>, TB;
1369 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1370 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1371 SSEPackedDouble>, TB, OpSize;
1372 } // Constraints = "$src1 = $dst"
1373 } // AddedComplexity
1375 //===----------------------------------------------------------------------===//
1376 // SSE 1 & 2 - Extract Floating-Point Sign mask
1377 //===----------------------------------------------------------------------===//
1379 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1380 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1382 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1383 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1384 [(set GR32:$dst, (Int RC:$src))], d>;
1385 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1386 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1390 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1391 "movmskps", SSEPackedSingle>, VEX;
1392 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1393 "movmskpd", SSEPackedDouble>, OpSize,
1395 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1396 "movmskps", SSEPackedSingle>, VEX;
1397 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1398 "movmskpd", SSEPackedDouble>, OpSize,
1400 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1401 SSEPackedSingle>, TB;
1402 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1403 SSEPackedDouble>, TB, OpSize;
1406 def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
1407 "movmskpd\t{$src, $dst|$dst, $src}",
1408 [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1409 def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1410 "movmskpd\t{$src, $dst|$dst, $src}",
1411 [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1412 def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
1413 "movmskps\t{$src, $dst|$dst, $src}",
1414 [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1415 def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1416 "movmskps\t{$src, $dst|$dst, $src}",
1417 [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1420 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1421 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1422 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1423 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1425 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1426 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1427 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1428 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1431 //===----------------------------------------------------------------------===//
1432 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1433 //===----------------------------------------------------------------------===//
1435 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1436 // names that start with 'Fs'.
1438 // Alias instructions that map fld0 to pxor for sse.
1439 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1440 canFoldAsLoad = 1 in {
1441 // FIXME: Set encoding to pseudo!
1442 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1443 [(set FR32:$dst, fp32imm0)]>,
1444 Requires<[HasSSE1]>, TB, OpSize;
1445 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1446 [(set FR64:$dst, fpimm0)]>,
1447 Requires<[HasSSE2]>, TB, OpSize;
1448 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1449 [(set FR32:$dst, fp32imm0)]>,
1450 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1451 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1452 [(set FR64:$dst, fpimm0)]>,
1453 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1456 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1457 // bits are disregarded.
1458 let neverHasSideEffects = 1 in {
1459 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1460 "movaps\t{$src, $dst|$dst, $src}", []>;
1461 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1462 "movapd\t{$src, $dst|$dst, $src}", []>;
1465 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1466 // bits are disregarded.
1467 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1468 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1469 "movaps\t{$src, $dst|$dst, $src}",
1470 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1471 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1472 "movapd\t{$src, $dst|$dst, $src}",
1473 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1476 //===----------------------------------------------------------------------===//
1477 // SSE 1 & 2 - Logical Instructions
1478 //===----------------------------------------------------------------------===//
1480 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1482 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1484 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1485 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1487 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1488 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1490 let Constraints = "$src1 = $dst" in {
1491 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1492 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1494 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1495 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1499 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1500 let mayLoad = 0 in {
1501 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1502 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1503 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1506 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1507 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1509 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1511 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1513 let Pattern = []<dag> in {
1514 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1515 !strconcat(OpcodeStr, "ps"), f128mem,
1516 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1517 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1518 (memopv2i64 addr:$src2)))], 0>, VEX_4V;
1520 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1521 !strconcat(OpcodeStr, "pd"), f128mem,
1522 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1523 (bc_v2i64 (v2f64 VR128:$src2))))],
1524 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1525 (memopv2i64 addr:$src2)))], 0>,
1528 let Constraints = "$src1 = $dst" in {
1529 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1530 !strconcat(OpcodeStr, "ps"), f128mem,
1531 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1532 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1533 (memopv2i64 addr:$src2)))]>, TB;
1535 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1536 !strconcat(OpcodeStr, "pd"), f128mem,
1537 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1538 (bc_v2i64 (v2f64 VR128:$src2))))],
1539 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1540 (memopv2i64 addr:$src2)))]>, TB, OpSize;
1544 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1546 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1548 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1549 !strconcat(OpcodeStr, "ps"), f256mem,
1550 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
1551 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1552 (memopv4i64 addr:$src2)))], 0>, VEX_4V;
1554 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1555 !strconcat(OpcodeStr, "pd"), f256mem,
1556 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1557 (bc_v4i64 (v4f64 VR256:$src2))))],
1558 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1559 (memopv4i64 addr:$src2)))], 0>,
1563 // AVX 256-bit packed logical ops forms
1564 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1565 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1566 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1567 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
1569 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1570 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1571 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1572 let isCommutable = 0 in
1573 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
1575 //===----------------------------------------------------------------------===//
1576 // SSE 1 & 2 - Arithmetic Instructions
1577 //===----------------------------------------------------------------------===//
1579 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1582 /// In addition, we also have a special variant of the scalar form here to
1583 /// represent the associated intrinsic operation. This form is unlike the
1584 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1585 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1587 /// These three forms can each be reg+reg or reg+mem.
1590 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1592 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1594 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1595 OpNode, FR32, f32mem, Is2Addr>, XS;
1596 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1597 OpNode, FR64, f64mem, Is2Addr>, XD;
1600 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1602 let mayLoad = 0 in {
1603 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1604 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1605 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1606 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1610 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1612 let mayLoad = 0 in {
1613 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1614 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1615 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1616 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1620 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1622 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1623 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1624 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1625 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1628 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1630 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1631 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1632 SSEPackedSingle, Is2Addr>, TB;
1634 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1635 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1636 SSEPackedDouble, Is2Addr>, TB, OpSize;
1639 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1640 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1641 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1642 SSEPackedSingle, 0>, TB;
1644 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1645 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1646 SSEPackedDouble, 0>, TB, OpSize;
1649 // Binary Arithmetic instructions
1650 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1651 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1652 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1653 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1654 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1655 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1656 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1657 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1659 let isCommutable = 0 in {
1660 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1661 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1662 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1663 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1664 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1665 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1666 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1667 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1668 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1669 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1670 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1671 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1672 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1673 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1674 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1675 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1676 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1677 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1678 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1679 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1682 let Constraints = "$src1 = $dst" in {
1683 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1684 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1685 basic_sse12_fp_binop_s_int<0x58, "add">;
1686 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1687 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1688 basic_sse12_fp_binop_s_int<0x59, "mul">;
1690 let isCommutable = 0 in {
1691 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1692 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1693 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1694 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1695 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1696 basic_sse12_fp_binop_s_int<0x5E, "div">;
1697 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1698 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1699 basic_sse12_fp_binop_s_int<0x5F, "max">,
1700 basic_sse12_fp_binop_p_int<0x5F, "max">;
1701 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1702 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1703 basic_sse12_fp_binop_s_int<0x5D, "min">,
1704 basic_sse12_fp_binop_p_int<0x5D, "min">;
1709 /// In addition, we also have a special variant of the scalar form here to
1710 /// represent the associated intrinsic operation. This form is unlike the
1711 /// plain scalar form, in that it takes an entire vector (instead of a
1712 /// scalar) and leaves the top elements undefined.
1714 /// And, we have a special variant form for a full-vector intrinsic form.
1716 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1717 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1718 SDNode OpNode, Intrinsic F32Int> {
1719 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1720 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1721 [(set FR32:$dst, (OpNode FR32:$src))]>;
1722 // For scalar unary operations, fold a load into the operation
1723 // only in OptForSize mode. It eliminates an instruction, but it also
1724 // eliminates a whole-register clobber (the load), so it introduces a
1725 // partial register update condition.
1726 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1727 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1728 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1729 Requires<[HasSSE1, OptForSize]>;
1730 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1731 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1732 [(set VR128:$dst, (F32Int VR128:$src))]>;
1733 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1734 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1735 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1738 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1739 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1740 SDNode OpNode, Intrinsic F32Int> {
1741 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1742 !strconcat(OpcodeStr,
1743 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1744 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1745 !strconcat(OpcodeStr,
1746 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1747 []>, XS, Requires<[HasAVX, OptForSize]>;
1748 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1749 !strconcat(OpcodeStr,
1750 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1751 [(set VR128:$dst, (F32Int VR128:$src))]>;
1752 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1753 !strconcat(OpcodeStr,
1754 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1755 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1758 /// sse1_fp_unop_p - SSE1 unops in packed form.
1759 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1760 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1761 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1762 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1763 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1764 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1765 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1768 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1769 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1770 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1771 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1772 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1773 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1774 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1775 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1778 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1779 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1780 Intrinsic V4F32Int> {
1781 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1782 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1783 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1784 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1785 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1786 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1789 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1790 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1791 Intrinsic V4F32Int> {
1792 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1793 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1794 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1795 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1796 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1797 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1800 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1801 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1802 SDNode OpNode, Intrinsic F64Int> {
1803 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1804 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1805 [(set FR64:$dst, (OpNode FR64:$src))]>;
1806 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1807 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1808 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1809 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1810 Requires<[HasSSE2, OptForSize]>;
1811 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1812 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1813 [(set VR128:$dst, (F64Int VR128:$src))]>;
1814 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1815 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1816 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1819 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1820 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1821 SDNode OpNode, Intrinsic F64Int> {
1822 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1823 !strconcat(OpcodeStr,
1824 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1825 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1826 (ins FR64:$src1, f64mem:$src2),
1827 !strconcat(OpcodeStr,
1828 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1829 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1830 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1831 [(set VR128:$dst, (F64Int VR128:$src))]>;
1832 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1833 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1834 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1837 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1838 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1840 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1841 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1842 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1843 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1844 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1845 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1848 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1849 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1850 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1851 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1852 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1853 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1854 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1855 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1858 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1859 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1860 Intrinsic V2F64Int> {
1861 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1862 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1863 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1864 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1865 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1866 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1869 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1870 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1871 Intrinsic V2F64Int> {
1872 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1873 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1874 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1875 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1876 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1877 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1880 let Predicates = [HasAVX] in {
1882 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1883 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1886 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1887 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1888 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1889 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1890 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1891 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1892 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1893 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1896 // Reciprocal approximations. Note that these typically require refinement
1897 // in order to obtain suitable precision.
1898 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1899 int_x86_sse_rsqrt_ss>, VEX_4V;
1900 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1901 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1902 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1903 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1905 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1907 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1908 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1909 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1910 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1913 def : Pat<(f32 (fsqrt FR32:$src)),
1914 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
1915 def : Pat<(f64 (fsqrt FR64:$src)),
1916 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
1917 def : Pat<(f64 (fsqrt (load addr:$src))),
1918 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
1919 Requires<[HasAVX, OptForSize]>;
1920 def : Pat<(f32 (fsqrt (load addr:$src))),
1921 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
1922 Requires<[HasAVX, OptForSize]>;
1925 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1926 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1927 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1928 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1929 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1930 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1932 // Reciprocal approximations. Note that these typically require refinement
1933 // in order to obtain suitable precision.
1934 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1935 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1936 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1937 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1938 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1939 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1941 // There is no f64 version of the reciprocal approximation instructions.
1943 //===----------------------------------------------------------------------===//
1944 // SSE 1 & 2 - Non-temporal stores
1945 //===----------------------------------------------------------------------===//
1947 let AddedComplexity = 400 in { // Prefer non-temporal versions
1948 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1949 (ins f128mem:$dst, VR128:$src),
1950 "movntps\t{$src, $dst|$dst, $src}",
1951 [(alignednontemporalstore (v4f32 VR128:$src),
1953 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1954 (ins f128mem:$dst, VR128:$src),
1955 "movntpd\t{$src, $dst|$dst, $src}",
1956 [(alignednontemporalstore (v2f64 VR128:$src),
1958 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1959 (ins f128mem:$dst, VR128:$src),
1960 "movntdq\t{$src, $dst|$dst, $src}",
1961 [(alignednontemporalstore (v2f64 VR128:$src),
1964 let ExeDomain = SSEPackedInt in
1965 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1966 (ins f128mem:$dst, VR128:$src),
1967 "movntdq\t{$src, $dst|$dst, $src}",
1968 [(alignednontemporalstore (v4f32 VR128:$src),
1971 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1972 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
1974 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1975 (ins f256mem:$dst, VR256:$src),
1976 "movntps\t{$src, $dst|$dst, $src}",
1977 [(alignednontemporalstore (v8f32 VR256:$src),
1979 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1980 (ins f256mem:$dst, VR256:$src),
1981 "movntpd\t{$src, $dst|$dst, $src}",
1982 [(alignednontemporalstore (v4f64 VR256:$src),
1984 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1985 (ins f256mem:$dst, VR256:$src),
1986 "movntdq\t{$src, $dst|$dst, $src}",
1987 [(alignednontemporalstore (v4f64 VR256:$src),
1989 let ExeDomain = SSEPackedInt in
1990 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1991 (ins f256mem:$dst, VR256:$src),
1992 "movntdq\t{$src, $dst|$dst, $src}",
1993 [(alignednontemporalstore (v8f32 VR256:$src),
1997 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1998 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1999 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2000 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2001 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2002 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2004 let AddedComplexity = 400 in { // Prefer non-temporal versions
2005 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2006 "movntps\t{$src, $dst|$dst, $src}",
2007 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2008 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2009 "movntpd\t{$src, $dst|$dst, $src}",
2010 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2012 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2013 "movntdq\t{$src, $dst|$dst, $src}",
2014 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2016 let ExeDomain = SSEPackedInt in
2017 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2018 "movntdq\t{$src, $dst|$dst, $src}",
2019 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2021 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2022 (MOVNTDQmr addr:$dst, VR128:$src)>;
2024 // There is no AVX form for instructions below this point
2025 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2026 "movnti{l}\t{$src, $dst|$dst, $src}",
2027 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2028 TB, Requires<[HasSSE2]>;
2029 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2030 "movnti{q}\t{$src, $dst|$dst, $src}",
2031 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2032 TB, Requires<[HasSSE2]>;
2035 //===----------------------------------------------------------------------===//
2036 // SSE 1 & 2 - Misc Instructions (No AVX form)
2037 //===----------------------------------------------------------------------===//
2039 // Prefetch intrinsic.
2040 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2041 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2042 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2043 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2044 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2045 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2046 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2047 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2049 // Load, store, and memory fence
2050 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2051 TB, Requires<[HasSSE1]>;
2052 def : Pat<(X86SFence), (SFENCE)>;
2054 // Alias instructions that map zero vector to pxor / xorp* for sse.
2055 // We set canFoldAsLoad because this can be converted to a constant-pool
2056 // load of an all-zeros value if folding it would be beneficial.
2057 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2058 // JIT implementation, it does not expand the instructions below like
2059 // X86MCInstLower does.
2060 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2061 isCodeGenOnly = 1 in {
2062 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2063 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2064 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2065 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2066 let ExeDomain = SSEPackedInt in
2067 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2068 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2071 // The same as done above but for AVX. The 128-bit versions are the
2072 // same, but re-encoded. The 256-bit does not support PI version, and
2073 // doesn't need it because on sandy bridge the register is set to zero
2074 // at the rename stage without using any execution unit, so SET0PSY
2075 // and SET0PDY can be used for vector int instructions without penalty
2076 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2077 // JIT implementatioan, it does not expand the instructions below like
2078 // X86MCInstLower does.
2079 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2080 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2081 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2082 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2083 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2084 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2085 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2086 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2087 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2088 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2089 let ExeDomain = SSEPackedInt in
2090 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2091 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2094 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2095 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2096 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2098 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2099 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2101 //===----------------------------------------------------------------------===//
2102 // SSE 1 & 2 - Load/Store XCSR register
2103 //===----------------------------------------------------------------------===//
2105 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2106 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2107 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2108 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2110 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2111 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2112 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2113 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2115 //===---------------------------------------------------------------------===//
2116 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2117 //===---------------------------------------------------------------------===//
2119 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2121 let neverHasSideEffects = 1 in {
2122 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2123 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2124 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2125 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2127 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2128 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2129 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2130 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2132 let canFoldAsLoad = 1, mayLoad = 1 in {
2133 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2134 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2135 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2136 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2137 let Predicates = [HasAVX] in {
2138 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2139 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2140 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2141 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2145 let mayStore = 1 in {
2146 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2147 (ins i128mem:$dst, VR128:$src),
2148 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2149 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2150 (ins i256mem:$dst, VR256:$src),
2151 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2152 let Predicates = [HasAVX] in {
2153 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2154 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2155 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2156 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2160 let neverHasSideEffects = 1 in
2161 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2162 "movdqa\t{$src, $dst|$dst, $src}", []>;
2164 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2165 "movdqu\t{$src, $dst|$dst, $src}",
2166 []>, XS, Requires<[HasSSE2]>;
2168 let canFoldAsLoad = 1, mayLoad = 1 in {
2169 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2170 "movdqa\t{$src, $dst|$dst, $src}",
2171 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2172 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2173 "movdqu\t{$src, $dst|$dst, $src}",
2174 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2175 XS, Requires<[HasSSE2]>;
2178 let mayStore = 1 in {
2179 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2180 "movdqa\t{$src, $dst|$dst, $src}",
2181 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2182 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2183 "movdqu\t{$src, $dst|$dst, $src}",
2184 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2185 XS, Requires<[HasSSE2]>;
2188 // Intrinsic forms of MOVDQU load and store
2189 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2190 "vmovdqu\t{$src, $dst|$dst, $src}",
2191 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2192 XS, VEX, Requires<[HasAVX]>;
2194 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2195 "movdqu\t{$src, $dst|$dst, $src}",
2196 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2197 XS, Requires<[HasSSE2]>;
2199 } // ExeDomain = SSEPackedInt
2201 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2202 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2203 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2205 //===---------------------------------------------------------------------===//
2206 // SSE2 - Packed Integer Arithmetic Instructions
2207 //===---------------------------------------------------------------------===//
2209 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2211 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2212 bit IsCommutable = 0, bit Is2Addr = 1> {
2213 let isCommutable = IsCommutable in
2214 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2215 (ins VR128:$src1, VR128:$src2),
2217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2219 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2220 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2221 (ins VR128:$src1, i128mem:$src2),
2223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2225 [(set VR128:$dst, (IntId VR128:$src1,
2226 (bitconvert (memopv2i64 addr:$src2))))]>;
2229 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2230 string OpcodeStr, Intrinsic IntId,
2231 Intrinsic IntId2, bit Is2Addr = 1> {
2232 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2233 (ins VR128:$src1, VR128:$src2),
2235 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2236 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2237 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2238 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2239 (ins VR128:$src1, i128mem:$src2),
2241 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2242 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2243 [(set VR128:$dst, (IntId VR128:$src1,
2244 (bitconvert (memopv2i64 addr:$src2))))]>;
2245 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2246 (ins VR128:$src1, i32i8imm:$src2),
2248 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2249 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2250 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2253 /// PDI_binop_rm - Simple SSE2 binary operator.
2254 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2255 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2256 let isCommutable = IsCommutable in
2257 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2258 (ins VR128:$src1, VR128:$src2),
2260 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2261 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2262 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2263 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2264 (ins VR128:$src1, i128mem:$src2),
2266 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2267 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2268 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2269 (bitconvert (memopv2i64 addr:$src2)))))]>;
2272 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2274 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2275 /// to collapse (bitconvert VT to VT) into its operand.
2277 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2278 bit IsCommutable = 0, bit Is2Addr = 1> {
2279 let isCommutable = IsCommutable in
2280 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2281 (ins VR128:$src1, VR128:$src2),
2283 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2284 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2285 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2286 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2287 (ins VR128:$src1, i128mem:$src2),
2289 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2290 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2291 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2294 } // ExeDomain = SSEPackedInt
2296 // 128-bit Integer Arithmetic
2298 let Predicates = [HasAVX] in {
2299 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2300 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2301 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2302 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2303 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2304 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2305 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2306 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2307 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2310 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2312 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2314 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2316 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2318 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2320 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2322 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2324 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2326 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2328 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2330 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2332 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2334 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2336 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2338 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2340 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2342 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2344 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2346 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2350 let Constraints = "$src1 = $dst" in {
2351 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2352 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2353 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2354 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2355 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2356 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2357 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2358 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2359 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2362 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2363 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2364 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2365 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2366 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2367 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2368 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2369 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2370 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2371 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2372 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2373 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2374 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2375 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2376 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2377 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2378 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2379 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2380 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2382 } // Constraints = "$src1 = $dst"
2384 //===---------------------------------------------------------------------===//
2385 // SSE2 - Packed Integer Logical Instructions
2386 //===---------------------------------------------------------------------===//
2388 let Predicates = [HasAVX] in {
2389 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2390 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2392 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2393 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2395 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2396 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2399 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2400 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2402 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2403 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2405 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2406 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2409 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2410 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2412 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2413 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2416 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2417 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2418 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2420 let ExeDomain = SSEPackedInt in {
2421 let neverHasSideEffects = 1 in {
2422 // 128-bit logical shifts.
2423 def VPSLLDQri : PDIi8<0x73, MRM7r,
2424 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2425 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2427 def VPSRLDQri : PDIi8<0x73, MRM3r,
2428 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2429 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2431 // PSRADQri doesn't exist in SSE[1-3].
2433 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2434 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2435 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2436 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2437 VR128:$src2)))]>, VEX_4V;
2439 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2440 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2441 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2442 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2443 (memopv2i64 addr:$src2))))]>,
2448 let Constraints = "$src1 = $dst" in {
2449 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2450 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2451 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2452 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2453 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2454 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2456 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2457 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2458 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2459 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2460 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2461 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2463 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2464 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2465 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2466 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2468 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2469 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2470 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2472 let ExeDomain = SSEPackedInt in {
2473 let neverHasSideEffects = 1 in {
2474 // 128-bit logical shifts.
2475 def PSLLDQri : PDIi8<0x73, MRM7r,
2476 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2477 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2478 def PSRLDQri : PDIi8<0x73, MRM3r,
2479 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2480 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2481 // PSRADQri doesn't exist in SSE[1-3].
2483 def PANDNrr : PDI<0xDF, MRMSrcReg,
2484 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2485 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2487 def PANDNrm : PDI<0xDF, MRMSrcMem,
2488 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2489 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2491 } // Constraints = "$src1 = $dst"
2493 let Predicates = [HasAVX] in {
2494 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2495 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2496 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2497 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2498 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2499 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2500 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2501 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2502 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2503 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2505 // Shift up / down and insert zero's.
2506 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2507 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2508 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2509 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2512 let Predicates = [HasSSE2] in {
2513 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2514 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2515 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2516 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2517 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2518 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2519 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2520 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2521 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2522 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2524 // Shift up / down and insert zero's.
2525 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2526 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2527 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2528 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2531 //===---------------------------------------------------------------------===//
2532 // SSE2 - Packed Integer Comparison Instructions
2533 //===---------------------------------------------------------------------===//
2535 let Predicates = [HasAVX] in {
2536 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2538 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2540 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2542 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2544 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2546 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2550 let Constraints = "$src1 = $dst" in {
2551 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2552 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2553 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2554 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2555 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2556 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2557 } // Constraints = "$src1 = $dst"
2559 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2560 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2561 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2562 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2563 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2564 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2565 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2566 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2567 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2568 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2569 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2570 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2572 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2573 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2574 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2575 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2576 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2577 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2578 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2579 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2580 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2581 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2582 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2583 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2585 //===---------------------------------------------------------------------===//
2586 // SSE2 - Packed Integer Pack Instructions
2587 //===---------------------------------------------------------------------===//
2589 let Predicates = [HasAVX] in {
2590 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2592 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2594 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2598 let Constraints = "$src1 = $dst" in {
2599 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2600 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2601 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2602 } // Constraints = "$src1 = $dst"
2604 //===---------------------------------------------------------------------===//
2605 // SSE2 - Packed Integer Shuffle Instructions
2606 //===---------------------------------------------------------------------===//
2608 let ExeDomain = SSEPackedInt in {
2609 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2611 def ri : Ii8<0x70, MRMSrcReg,
2612 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2613 !strconcat(OpcodeStr,
2614 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2615 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2617 def mi : Ii8<0x70, MRMSrcMem,
2618 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2619 !strconcat(OpcodeStr,
2620 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2621 [(set VR128:$dst, (vt (pshuf_frag:$src2
2622 (bc_frag (memopv2i64 addr:$src1)),
2625 } // ExeDomain = SSEPackedInt
2627 let Predicates = [HasAVX] in {
2628 let AddedComplexity = 5 in
2629 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2632 // SSE2 with ImmT == Imm8 and XS prefix.
2633 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2636 // SSE2 with ImmT == Imm8 and XD prefix.
2637 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2641 let Predicates = [HasSSE2] in {
2642 let AddedComplexity = 5 in
2643 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2645 // SSE2 with ImmT == Imm8 and XS prefix.
2646 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2648 // SSE2 with ImmT == Imm8 and XD prefix.
2649 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2652 //===---------------------------------------------------------------------===//
2653 // SSE2 - Packed Integer Unpack Instructions
2654 //===---------------------------------------------------------------------===//
2656 let ExeDomain = SSEPackedInt in {
2657 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2658 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
2659 def rr : PDI<opc, MRMSrcReg,
2660 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2662 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2663 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2664 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
2665 def rm : PDI<opc, MRMSrcMem,
2666 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2668 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2669 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2670 [(set VR128:$dst, (OpNode VR128:$src1,
2671 (bc_frag (memopv2i64
2675 let Predicates = [HasAVX] in {
2676 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
2677 bc_v16i8, 0>, VEX_4V;
2678 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
2679 bc_v8i16, 0>, VEX_4V;
2680 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
2681 bc_v4i32, 0>, VEX_4V;
2683 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2684 /// knew to collapse (bitconvert VT to VT) into its operand.
2685 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2686 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2687 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2688 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
2689 VR128:$src2)))]>, VEX_4V;
2690 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2691 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2692 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2693 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
2694 (memopv2i64 addr:$src2))))]>, VEX_4V;
2696 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
2697 bc_v16i8, 0>, VEX_4V;
2698 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
2699 bc_v8i16, 0>, VEX_4V;
2700 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
2701 bc_v4i32, 0>, VEX_4V;
2703 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2704 /// knew to collapse (bitconvert VT to VT) into its operand.
2705 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2706 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2707 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2708 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
2709 VR128:$src2)))]>, VEX_4V;
2710 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2711 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2712 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2713 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
2714 (memopv2i64 addr:$src2))))]>, VEX_4V;
2717 let Constraints = "$src1 = $dst" in {
2718 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
2719 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
2720 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
2722 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2723 /// knew to collapse (bitconvert VT to VT) into its operand.
2724 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2725 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2726 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2728 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
2729 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2730 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2731 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2733 (v2i64 (X86Punpcklqdq VR128:$src1,
2734 (memopv2i64 addr:$src2))))]>;
2736 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
2737 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
2738 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
2740 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2741 /// knew to collapse (bitconvert VT to VT) into its operand.
2742 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2743 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2744 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2746 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
2747 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2748 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2749 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2751 (v2i64 (X86Punpckhqdq VR128:$src1,
2752 (memopv2i64 addr:$src2))))]>;
2755 } // ExeDomain = SSEPackedInt
2757 //===---------------------------------------------------------------------===//
2758 // SSE2 - Packed Integer Extract and Insert
2759 //===---------------------------------------------------------------------===//
2761 let ExeDomain = SSEPackedInt in {
2762 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2763 def rri : Ii8<0xC4, MRMSrcReg,
2764 (outs VR128:$dst), (ins VR128:$src1,
2765 GR32:$src2, i32i8imm:$src3),
2767 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2768 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2770 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2771 def rmi : Ii8<0xC4, MRMSrcMem,
2772 (outs VR128:$dst), (ins VR128:$src1,
2773 i16mem:$src2, i32i8imm:$src3),
2775 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2776 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2778 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2783 let Predicates = [HasAVX] in
2784 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2785 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2786 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2787 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2788 imm:$src2))]>, OpSize, VEX;
2789 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2790 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2791 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2792 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2796 let Predicates = [HasAVX] in {
2797 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2798 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2799 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2800 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2801 []>, OpSize, VEX_4V;
2804 let Constraints = "$src1 = $dst" in
2805 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2807 } // ExeDomain = SSEPackedInt
2809 //===---------------------------------------------------------------------===//
2810 // SSE2 - Packed Mask Creation
2811 //===---------------------------------------------------------------------===//
2813 let ExeDomain = SSEPackedInt in {
2815 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2816 "pmovmskb\t{$src, $dst|$dst, $src}",
2817 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2818 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2819 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2820 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2821 "pmovmskb\t{$src, $dst|$dst, $src}",
2822 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2824 } // ExeDomain = SSEPackedInt
2826 //===---------------------------------------------------------------------===//
2827 // SSE2 - Conditional Store
2828 //===---------------------------------------------------------------------===//
2830 let ExeDomain = SSEPackedInt in {
2833 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2834 (ins VR128:$src, VR128:$mask),
2835 "maskmovdqu\t{$mask, $src|$src, $mask}",
2836 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2838 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2839 (ins VR128:$src, VR128:$mask),
2840 "maskmovdqu\t{$mask, $src|$src, $mask}",
2841 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2844 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2845 "maskmovdqu\t{$mask, $src|$src, $mask}",
2846 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2848 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2849 "maskmovdqu\t{$mask, $src|$src, $mask}",
2850 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2852 } // ExeDomain = SSEPackedInt
2854 //===---------------------------------------------------------------------===//
2855 // SSE2 - Move Doubleword
2856 //===---------------------------------------------------------------------===//
2858 // Move Int Doubleword to Packed Double Int
2859 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2860 "movd\t{$src, $dst|$dst, $src}",
2862 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2863 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2864 "movd\t{$src, $dst|$dst, $src}",
2866 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2868 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2869 "mov{d|q}\t{$src, $dst|$dst, $src}",
2871 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
2872 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2873 "mov{d|q}\t{$src, $dst|$dst, $src}",
2874 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
2876 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2877 "movd\t{$src, $dst|$dst, $src}",
2879 (v4i32 (scalar_to_vector GR32:$src)))]>;
2880 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2881 "movd\t{$src, $dst|$dst, $src}",
2883 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2884 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2885 "mov{d|q}\t{$src, $dst|$dst, $src}",
2887 (v2i64 (scalar_to_vector GR64:$src)))]>;
2888 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2889 "mov{d|q}\t{$src, $dst|$dst, $src}",
2890 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2893 // Move Int Doubleword to Single Scalar
2894 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2895 "movd\t{$src, $dst|$dst, $src}",
2896 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2898 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2899 "movd\t{$src, $dst|$dst, $src}",
2900 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2902 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2903 "movd\t{$src, $dst|$dst, $src}",
2904 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2906 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2907 "movd\t{$src, $dst|$dst, $src}",
2908 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2910 // Move Packed Doubleword Int to Packed Double Int
2911 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2912 "movd\t{$src, $dst|$dst, $src}",
2913 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2915 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2916 (ins i32mem:$dst, VR128:$src),
2917 "movd\t{$src, $dst|$dst, $src}",
2918 [(store (i32 (vector_extract (v4i32 VR128:$src),
2919 (iPTR 0))), addr:$dst)]>, VEX;
2920 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2921 "movd\t{$src, $dst|$dst, $src}",
2922 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2924 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2925 "movd\t{$src, $dst|$dst, $src}",
2926 [(store (i32 (vector_extract (v4i32 VR128:$src),
2927 (iPTR 0))), addr:$dst)]>;
2929 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2930 "mov{d|q}\t{$src, $dst|$dst, $src}",
2931 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2933 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2934 "movq\t{$src, $dst|$dst, $src}",
2935 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2937 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2938 "mov{d|q}\t{$src, $dst|$dst, $src}",
2939 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2940 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2941 "movq\t{$src, $dst|$dst, $src}",
2942 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2944 // Move Scalar Single to Double Int
2945 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2946 "movd\t{$src, $dst|$dst, $src}",
2947 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2948 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2949 "movd\t{$src, $dst|$dst, $src}",
2950 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2951 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2953 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2954 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2955 "movd\t{$src, $dst|$dst, $src}",
2956 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2958 // movd / movq to XMM register zero-extends
2959 let AddedComplexity = 15 in {
2960 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2961 "movd\t{$src, $dst|$dst, $src}",
2962 [(set VR128:$dst, (v4i32 (X86vzmovl
2963 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2965 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2966 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2967 [(set VR128:$dst, (v2i64 (X86vzmovl
2968 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2971 let AddedComplexity = 15 in {
2972 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2973 "movd\t{$src, $dst|$dst, $src}",
2974 [(set VR128:$dst, (v4i32 (X86vzmovl
2975 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2976 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2977 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2978 [(set VR128:$dst, (v2i64 (X86vzmovl
2979 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2982 let AddedComplexity = 20 in {
2983 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2984 "movd\t{$src, $dst|$dst, $src}",
2986 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2987 (loadi32 addr:$src))))))]>,
2989 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2990 "movd\t{$src, $dst|$dst, $src}",
2992 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2993 (loadi32 addr:$src))))))]>;
2995 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2996 (MOVZDI2PDIrm addr:$src)>;
2997 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2998 (MOVZDI2PDIrm addr:$src)>;
2999 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3000 (MOVZDI2PDIrm addr:$src)>;
3003 // These are the correct encodings of the instructions so that we know how to
3004 // read correct assembly, even though we continue to emit the wrong ones for
3005 // compatibility with Darwin's buggy assembler.
3006 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3007 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
3008 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3009 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
3010 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3011 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
3012 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3013 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
3014 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3015 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3016 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3017 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3019 //===---------------------------------------------------------------------===//
3020 // SSE2 - Move Quadword
3021 //===---------------------------------------------------------------------===//
3023 // Move Quadword Int to Packed Quadword Int
3024 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3025 "vmovq\t{$src, $dst|$dst, $src}",
3027 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3028 VEX, Requires<[HasAVX]>;
3029 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3030 "movq\t{$src, $dst|$dst, $src}",
3032 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3033 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3035 // Move Packed Quadword Int to Quadword Int
3036 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3037 "movq\t{$src, $dst|$dst, $src}",
3038 [(store (i64 (vector_extract (v2i64 VR128:$src),
3039 (iPTR 0))), addr:$dst)]>, VEX;
3040 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3041 "movq\t{$src, $dst|$dst, $src}",
3042 [(store (i64 (vector_extract (v2i64 VR128:$src),
3043 (iPTR 0))), addr:$dst)]>;
3045 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3046 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3048 // Store / copy lower 64-bits of a XMM register.
3049 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3050 "movq\t{$src, $dst|$dst, $src}",
3051 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3052 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3053 "movq\t{$src, $dst|$dst, $src}",
3054 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3056 let AddedComplexity = 20 in
3057 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3058 "vmovq\t{$src, $dst|$dst, $src}",
3060 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3061 (loadi64 addr:$src))))))]>,
3062 XS, VEX, Requires<[HasAVX]>;
3064 let AddedComplexity = 20 in {
3065 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3066 "movq\t{$src, $dst|$dst, $src}",
3068 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3069 (loadi64 addr:$src))))))]>,
3070 XS, Requires<[HasSSE2]>;
3072 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3073 (MOVZQI2PQIrm addr:$src)>;
3074 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3075 (MOVZQI2PQIrm addr:$src)>;
3076 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3079 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3080 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3081 let AddedComplexity = 15 in
3082 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3083 "vmovq\t{$src, $dst|$dst, $src}",
3084 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3085 XS, VEX, Requires<[HasAVX]>;
3086 let AddedComplexity = 15 in
3087 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3088 "movq\t{$src, $dst|$dst, $src}",
3089 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3090 XS, Requires<[HasSSE2]>;
3092 let AddedComplexity = 20 in
3093 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3094 "vmovq\t{$src, $dst|$dst, $src}",
3095 [(set VR128:$dst, (v2i64 (X86vzmovl
3096 (loadv2i64 addr:$src))))]>,
3097 XS, VEX, Requires<[HasAVX]>;
3098 let AddedComplexity = 20 in {
3099 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3100 "movq\t{$src, $dst|$dst, $src}",
3101 [(set VR128:$dst, (v2i64 (X86vzmovl
3102 (loadv2i64 addr:$src))))]>,
3103 XS, Requires<[HasSSE2]>;
3105 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3106 (MOVZPQILo2PQIrm addr:$src)>;
3109 // Instructions to match in the assembler
3110 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3111 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3112 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3113 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3114 // Recognize "movd" with GR64 destination, but encode as a "movq"
3115 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3116 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3118 // Instructions for the disassembler
3119 // xr = XMM register
3122 let Predicates = [HasAVX] in
3123 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3124 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3125 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3126 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3128 //===---------------------------------------------------------------------===//
3129 // SSE2 - Misc Instructions
3130 //===---------------------------------------------------------------------===//
3133 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3134 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3135 TB, Requires<[HasSSE2]>;
3137 // Load, store, and memory fence
3138 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3139 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3140 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3141 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3142 def : Pat<(X86LFence), (LFENCE)>;
3143 def : Pat<(X86MFence), (MFENCE)>;
3146 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3147 // was introduced with SSE2, it's backward compatible.
3148 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3150 // Alias instructions that map zero vector to pxor / xorp* for sse.
3151 // We set canFoldAsLoad because this can be converted to a constant-pool
3152 // load of an all-ones value if folding it would be beneficial.
3153 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
3154 // JIT implementation, it does not expand the instructions below like
3155 // X86MCInstLower does.
3156 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3157 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3158 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3159 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3160 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3161 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
3162 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3163 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
3165 //===---------------------------------------------------------------------===//
3166 // SSE3 - Conversion Instructions
3167 //===---------------------------------------------------------------------===//
3169 // Convert Packed Double FP to Packed DW Integers
3170 let Predicates = [HasAVX] in {
3171 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3172 // register, but the same isn't true when using memory operands instead.
3173 // Provide other assembly rr and rm forms to address this explicitly.
3174 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3175 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3176 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3177 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3180 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3181 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3182 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3183 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3186 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3187 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3188 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3189 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3192 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3193 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3194 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3195 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3197 // Convert Packed DW Integers to Packed Double FP
3198 let Predicates = [HasAVX] in {
3199 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3200 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3201 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3202 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3203 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3204 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3205 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3206 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3209 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3210 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3211 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3212 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3214 // AVX 256-bit register conversion intrinsics
3215 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3216 (VCVTDQ2PDYrr VR128:$src)>;
3217 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3218 (VCVTDQ2PDYrm addr:$src)>;
3220 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3221 (VCVTPD2DQYrr VR256:$src)>;
3222 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3223 (VCVTPD2DQYrm addr:$src)>;
3225 //===---------------------------------------------------------------------===//
3226 // SSE3 - Move Instructions
3227 //===---------------------------------------------------------------------===//
3229 //===---------------------------------------------------------------------===//
3230 // Replicate Single FP - MOVSHDUP and MOVSLDUP
3232 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
3233 ValueType vt, RegisterClass RC, PatFrag mem_frag,
3234 X86MemOperand x86memop> {
3235 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
3236 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3237 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
3238 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
3239 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3240 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
3243 let Predicates = [HasAVX] in {
3244 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3245 v4f32, VR128, memopv4f32, f128mem>, VEX;
3246 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3247 v4f32, VR128, memopv4f32, f128mem>, VEX;
3248 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3249 v8f32, VR256, memopv8f32, f256mem>, VEX;
3250 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3251 v8f32, VR256, memopv8f32, f256mem>, VEX;
3253 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
3254 memopv4f32, f128mem>;
3255 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
3256 memopv4f32, f128mem>;
3258 let Predicates = [HasSSE3] in {
3259 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3260 (MOVSHDUPrr VR128:$src)>;
3261 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3262 (MOVSHDUPrm addr:$src)>;
3263 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3264 (MOVSLDUPrr VR128:$src)>;
3265 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3266 (MOVSLDUPrm addr:$src)>;
3269 let Predicates = [HasAVX] in {
3270 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3271 (VMOVSHDUPrr VR128:$src)>;
3272 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3273 (VMOVSHDUPrm addr:$src)>;
3274 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3275 (VMOVSLDUPrr VR128:$src)>;
3276 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3277 (VMOVSLDUPrm addr:$src)>;
3278 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
3279 (VMOVSHDUPYrr VR256:$src)>;
3280 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
3281 (VMOVSHDUPYrm addr:$src)>;
3282 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
3283 (VMOVSLDUPYrr VR256:$src)>;
3284 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
3285 (VMOVSLDUPYrm addr:$src)>;
3288 //===---------------------------------------------------------------------===//
3289 // Replicate Double FP - MOVDDUP
3291 multiclass sse3_replicate_dfp<string OpcodeStr> {
3292 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3293 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3294 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3295 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3296 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3298 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3302 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3303 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3304 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3306 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3307 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3311 let Predicates = [HasAVX] in {
3312 // FIXME: Merge above classes when we have patterns for the ymm version
3313 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3314 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3316 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3318 // Move Unaligned Integer
3319 let Predicates = [HasAVX] in {
3320 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3321 "vlddqu\t{$src, $dst|$dst, $src}",
3322 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3323 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3324 "vlddqu\t{$src, $dst|$dst, $src}",
3325 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3327 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3328 "lddqu\t{$src, $dst|$dst, $src}",
3329 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3331 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3333 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3335 // Several Move patterns
3336 let AddedComplexity = 5 in {
3337 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3338 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3339 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3340 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3341 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3342 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3343 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3344 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3347 //===---------------------------------------------------------------------===//
3348 // SSE3 - Arithmetic
3349 //===---------------------------------------------------------------------===//
3351 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3352 X86MemOperand x86memop, bit Is2Addr = 1> {
3353 def rr : I<0xD0, MRMSrcReg,
3354 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3356 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3357 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3358 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3359 def rm : I<0xD0, MRMSrcMem,
3360 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3362 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3363 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3364 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3367 let Predicates = [HasAVX],
3368 ExeDomain = SSEPackedDouble in {
3369 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3370 f128mem, 0>, TB, XD, VEX_4V;
3371 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3372 f128mem, 0>, TB, OpSize, VEX_4V;
3373 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3374 f256mem, 0>, TB, XD, VEX_4V;
3375 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3376 f256mem, 0>, TB, OpSize, VEX_4V;
3378 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3379 ExeDomain = SSEPackedDouble in {
3380 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3382 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3383 f128mem>, TB, OpSize;
3386 //===---------------------------------------------------------------------===//
3387 // SSE3 Instructions
3388 //===---------------------------------------------------------------------===//
3391 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3392 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3393 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3395 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3396 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3397 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3399 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3401 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3402 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3403 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3405 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3406 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3407 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3409 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3410 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3411 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3413 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3415 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3416 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3417 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3420 let Predicates = [HasAVX] in {
3421 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3422 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3423 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3424 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3425 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3426 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3427 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3428 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3429 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3430 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3431 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3432 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3433 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3434 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3435 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3436 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3439 let Constraints = "$src1 = $dst" in {
3440 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3441 int_x86_sse3_hadd_ps>;
3442 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3443 int_x86_sse3_hadd_pd>;
3444 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3445 int_x86_sse3_hsub_ps>;
3446 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3447 int_x86_sse3_hsub_pd>;
3450 //===---------------------------------------------------------------------===//
3451 // SSSE3 - Packed Absolute Instructions
3452 //===---------------------------------------------------------------------===//
3455 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3456 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3457 PatFrag mem_frag128, Intrinsic IntId128> {
3458 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3460 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3461 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3464 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3466 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3469 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3472 let Predicates = [HasAVX] in {
3473 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3474 int_x86_ssse3_pabs_b_128>, VEX;
3475 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3476 int_x86_ssse3_pabs_w_128>, VEX;
3477 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3478 int_x86_ssse3_pabs_d_128>, VEX;
3481 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3482 int_x86_ssse3_pabs_b_128>;
3483 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3484 int_x86_ssse3_pabs_w_128>;
3485 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3486 int_x86_ssse3_pabs_d_128>;
3488 //===---------------------------------------------------------------------===//
3489 // SSSE3 - Packed Binary Operator Instructions
3490 //===---------------------------------------------------------------------===//
3492 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3493 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3494 PatFrag mem_frag128, Intrinsic IntId128,
3496 let isCommutable = 1 in
3497 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3498 (ins VR128:$src1, VR128:$src2),
3500 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3501 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3502 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3504 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3505 (ins VR128:$src1, i128mem:$src2),
3507 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3508 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3510 (IntId128 VR128:$src1,
3511 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3514 let Predicates = [HasAVX] in {
3515 let isCommutable = 0 in {
3516 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3517 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3518 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3519 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3520 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3521 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3522 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3523 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3524 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3525 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3526 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3527 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3528 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3529 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3530 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3531 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3532 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3533 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3534 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3535 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3536 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3537 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3539 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3540 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3543 // None of these have i8 immediate fields.
3544 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3545 let isCommutable = 0 in {
3546 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3547 int_x86_ssse3_phadd_w_128>;
3548 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3549 int_x86_ssse3_phadd_d_128>;
3550 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3551 int_x86_ssse3_phadd_sw_128>;
3552 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3553 int_x86_ssse3_phsub_w_128>;
3554 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3555 int_x86_ssse3_phsub_d_128>;
3556 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3557 int_x86_ssse3_phsub_sw_128>;
3558 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3559 int_x86_ssse3_pmadd_ub_sw_128>;
3560 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3561 int_x86_ssse3_pshuf_b_128>;
3562 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3563 int_x86_ssse3_psign_b_128>;
3564 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3565 int_x86_ssse3_psign_w_128>;
3566 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3567 int_x86_ssse3_psign_d_128>;
3569 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3570 int_x86_ssse3_pmul_hr_sw_128>;
3573 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3574 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3575 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3576 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3578 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3579 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3580 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3581 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3582 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3583 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3585 //===---------------------------------------------------------------------===//
3586 // SSSE3 - Packed Align Instruction Patterns
3587 //===---------------------------------------------------------------------===//
3589 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3590 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3591 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3593 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3595 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3597 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3598 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3600 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3602 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3606 let Predicates = [HasAVX] in
3607 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3608 let Constraints = "$src1 = $dst" in
3609 defm PALIGN : ssse3_palign<"palignr">;
3611 let AddedComplexity = 5 in {
3612 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3613 (PALIGNR128rr VR128:$src2, VR128:$src1,
3614 (SHUFFLE_get_palign_imm VR128:$src3))>,
3615 Requires<[HasSSSE3]>;
3616 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3617 (PALIGNR128rr VR128:$src2, VR128:$src1,
3618 (SHUFFLE_get_palign_imm VR128:$src3))>,
3619 Requires<[HasSSSE3]>;
3620 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3621 (PALIGNR128rr VR128:$src2, VR128:$src1,
3622 (SHUFFLE_get_palign_imm VR128:$src3))>,
3623 Requires<[HasSSSE3]>;
3624 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3625 (PALIGNR128rr VR128:$src2, VR128:$src1,
3626 (SHUFFLE_get_palign_imm VR128:$src3))>,
3627 Requires<[HasSSSE3]>;
3630 //===---------------------------------------------------------------------===//
3631 // SSSE3 Misc Instructions
3632 //===---------------------------------------------------------------------===//
3634 // Thread synchronization
3635 let usesCustomInserter = 1 in {
3636 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3637 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3638 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3639 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3642 let Uses = [EAX, ECX, EDX] in
3643 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3644 Requires<[HasSSE3]>;
3645 let Uses = [ECX, EAX] in
3646 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3647 Requires<[HasSSE3]>;
3649 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3650 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3652 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3653 Requires<[In32BitMode]>;
3654 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3655 Requires<[In64BitMode]>;
3657 //===---------------------------------------------------------------------===//
3658 // Non-Instruction Patterns
3659 //===---------------------------------------------------------------------===//
3661 // extload f32 -> f64. This matches load+fextend because we have a hack in
3662 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3664 // Since these loads aren't folded into the fextend, we have to match it
3666 let Predicates = [HasSSE2] in
3667 def : Pat<(fextend (loadf32 addr:$src)),
3668 (CVTSS2SDrm addr:$src)>;
3670 // Bitcasts between 128-bit vector types. Return the original type since
3671 // no instruction is needed for the conversion
3672 let Predicates = [HasXMMInt] in {
3673 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3674 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3675 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3676 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3677 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3678 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3679 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3680 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3681 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3682 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3683 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3684 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3685 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3686 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3687 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3688 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3689 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3690 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3691 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3692 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3693 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3694 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3695 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3696 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3697 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3698 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3699 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3700 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3701 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3702 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3705 // Bitcasts between 256-bit vector types. Return the original type since
3706 // no instruction is needed for the conversion
3707 let Predicates = [HasAVX] in {
3708 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3709 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
3710 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
3711 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
3712 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
3713 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
3714 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
3715 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
3716 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
3717 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
3718 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
3719 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
3720 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
3721 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
3722 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
3723 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
3724 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
3725 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
3726 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
3727 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
3728 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
3729 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
3730 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
3731 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
3732 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
3733 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
3734 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
3735 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
3736 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
3737 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
3740 // Move scalar to XMM zero-extended
3741 // movd to XMM register zero-extends
3742 let AddedComplexity = 15 in {
3743 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3744 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3745 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3746 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3747 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3748 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3749 (MOVSSrr (v4f32 (V_SET0PS)),
3750 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3751 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3752 (MOVSSrr (v4i32 (V_SET0PI)),
3753 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3756 // Splat v2f64 / v2i64
3757 let AddedComplexity = 10 in {
3758 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3759 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3760 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3761 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3764 // Special unary SHUFPSrri case.
3765 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3766 (SHUFPSrri VR128:$src1, VR128:$src1,
3767 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3768 let AddedComplexity = 5 in
3769 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3770 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3771 Requires<[HasSSE2]>;
3772 // Special unary SHUFPDrri case.
3773 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3774 (SHUFPDrri VR128:$src1, VR128:$src1,
3775 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3776 Requires<[HasSSE2]>;
3777 // Special unary SHUFPDrri case.
3778 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3779 (SHUFPDrri VR128:$src1, VR128:$src1,
3780 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3781 Requires<[HasSSE2]>;
3782 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3783 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3784 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3785 Requires<[HasSSE2]>;
3787 // Special binary v4i32 shuffle cases with SHUFPS.
3788 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3789 (SHUFPSrri VR128:$src1, VR128:$src2,
3790 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3791 Requires<[HasSSE2]>;
3792 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3793 (SHUFPSrmi VR128:$src1, addr:$src2,
3794 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3795 Requires<[HasSSE2]>;
3796 // Special binary v2i64 shuffle cases using SHUFPDrri.
3797 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3798 (SHUFPDrri VR128:$src1, VR128:$src2,
3799 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3800 Requires<[HasSSE2]>;
3802 let AddedComplexity = 20 in {
3803 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3804 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3805 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3807 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3808 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3809 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3811 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3812 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3813 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3814 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3815 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3818 let AddedComplexity = 20 in {
3819 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3820 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3821 (MOVLPSrm VR128:$src1, addr:$src2)>;
3822 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3823 (MOVLPDrm VR128:$src1, addr:$src2)>;
3824 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3825 (MOVLPSrm VR128:$src1, addr:$src2)>;
3826 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3827 (MOVLPDrm VR128:$src1, addr:$src2)>;
3830 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3831 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3832 (MOVLPSmr addr:$src1, VR128:$src2)>;
3833 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3834 (MOVLPDmr addr:$src1, VR128:$src2)>;
3835 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3837 (MOVLPSmr addr:$src1, VR128:$src2)>;
3838 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3839 (MOVLPDmr addr:$src1, VR128:$src2)>;
3841 let AddedComplexity = 15 in {
3842 // Setting the lowest element in the vector.
3843 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3844 (MOVSSrr (v4i32 VR128:$src1),
3845 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3846 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3847 (MOVSDrr (v2i64 VR128:$src1),
3848 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3850 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3851 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3852 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3853 Requires<[HasSSE2]>;
3854 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3855 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3856 Requires<[HasSSE2]>;
3859 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3860 // fall back to this for SSE1)
3861 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3862 (SHUFPSrri VR128:$src2, VR128:$src1,
3863 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3865 // Set lowest element and zero upper elements.
3866 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3867 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3869 // Use movaps / movups for SSE integer load / store (one byte shorter).
3870 // The instructions selected below are then converted to MOVDQA/MOVDQU
3871 // during the SSE domain pass.
3872 let Predicates = [HasSSE1] in {
3873 def : Pat<(alignedloadv4i32 addr:$src),
3874 (MOVAPSrm addr:$src)>;
3875 def : Pat<(loadv4i32 addr:$src),
3876 (MOVUPSrm addr:$src)>;
3877 def : Pat<(alignedloadv2i64 addr:$src),
3878 (MOVAPSrm addr:$src)>;
3879 def : Pat<(loadv2i64 addr:$src),
3880 (MOVUPSrm addr:$src)>;
3882 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3883 (MOVAPSmr addr:$dst, VR128:$src)>;
3884 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3885 (MOVAPSmr addr:$dst, VR128:$src)>;
3886 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3887 (MOVAPSmr addr:$dst, VR128:$src)>;
3888 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3889 (MOVAPSmr addr:$dst, VR128:$src)>;
3890 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3891 (MOVUPSmr addr:$dst, VR128:$src)>;
3892 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3893 (MOVUPSmr addr:$dst, VR128:$src)>;
3894 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3895 (MOVUPSmr addr:$dst, VR128:$src)>;
3896 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3897 (MOVUPSmr addr:$dst, VR128:$src)>;
3900 // Use vmovaps/vmovups for AVX integer load/store.
3901 let Predicates = [HasAVX] in {
3902 // 128-bit load/store
3903 def : Pat<(alignedloadv4i32 addr:$src),
3904 (VMOVAPSrm addr:$src)>;
3905 def : Pat<(loadv4i32 addr:$src),
3906 (VMOVUPSrm addr:$src)>;
3907 def : Pat<(alignedloadv2i64 addr:$src),
3908 (VMOVAPSrm addr:$src)>;
3909 def : Pat<(loadv2i64 addr:$src),
3910 (VMOVUPSrm addr:$src)>;
3912 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3913 (VMOVAPSmr addr:$dst, VR128:$src)>;
3914 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3915 (VMOVAPSmr addr:$dst, VR128:$src)>;
3916 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3917 (VMOVAPSmr addr:$dst, VR128:$src)>;
3918 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3919 (VMOVAPSmr addr:$dst, VR128:$src)>;
3920 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3921 (VMOVUPSmr addr:$dst, VR128:$src)>;
3922 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3923 (VMOVUPSmr addr:$dst, VR128:$src)>;
3924 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3925 (VMOVUPSmr addr:$dst, VR128:$src)>;
3926 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3927 (VMOVUPSmr addr:$dst, VR128:$src)>;
3929 // 256-bit load/store
3930 def : Pat<(alignedloadv4i64 addr:$src),
3931 (VMOVAPSYrm addr:$src)>;
3932 def : Pat<(loadv4i64 addr:$src),
3933 (VMOVUPSYrm addr:$src)>;
3934 def : Pat<(alignedloadv8i32 addr:$src),
3935 (VMOVAPSYrm addr:$src)>;
3936 def : Pat<(loadv8i32 addr:$src),
3937 (VMOVUPSYrm addr:$src)>;
3938 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
3939 (VMOVAPSYmr addr:$dst, VR256:$src)>;
3940 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
3941 (VMOVAPSYmr addr:$dst, VR256:$src)>;
3942 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
3943 (VMOVUPSYmr addr:$dst, VR256:$src)>;
3944 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
3945 (VMOVUPSYmr addr:$dst, VR256:$src)>;
3948 //===----------------------------------------------------------------------===//
3949 // SSE4.1 - Packed Move with Sign/Zero Extend
3950 //===----------------------------------------------------------------------===//
3952 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3953 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3954 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3955 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3957 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3958 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3960 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3964 let Predicates = [HasAVX] in {
3965 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3967 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3969 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3971 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3973 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3975 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3979 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3980 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3981 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3982 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3983 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3984 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3986 // Common patterns involving scalar load.
3987 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3988 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3989 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3990 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3992 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3993 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3994 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3995 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3997 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3998 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3999 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4000 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4002 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4003 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4004 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4005 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4007 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4008 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4009 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4010 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4012 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4013 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4014 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4015 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4018 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4019 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4020 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4021 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4023 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4024 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4026 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4030 let Predicates = [HasAVX] in {
4031 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4033 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4035 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4037 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4041 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4042 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4043 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4044 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4046 // Common patterns involving scalar load
4047 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4048 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4049 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4050 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4052 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4053 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4054 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4055 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4058 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4059 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4060 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4061 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4063 // Expecting a i16 load any extended to i32 value.
4064 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4065 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4066 [(set VR128:$dst, (IntId (bitconvert
4067 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4071 let Predicates = [HasAVX] in {
4072 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4074 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4077 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4078 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4080 // Common patterns involving scalar load
4081 def : Pat<(int_x86_sse41_pmovsxbq
4082 (bitconvert (v4i32 (X86vzmovl
4083 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4084 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4086 def : Pat<(int_x86_sse41_pmovzxbq
4087 (bitconvert (v4i32 (X86vzmovl
4088 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4089 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4091 //===----------------------------------------------------------------------===//
4092 // SSE4.1 - Extract Instructions
4093 //===----------------------------------------------------------------------===//
4095 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4096 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4097 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4098 (ins VR128:$src1, i32i8imm:$src2),
4099 !strconcat(OpcodeStr,
4100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4101 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4103 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4104 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4105 !strconcat(OpcodeStr,
4106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4109 // There's an AssertZext in the way of writing the store pattern
4110 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4113 let Predicates = [HasAVX] in {
4114 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4115 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4116 (ins VR128:$src1, i32i8imm:$src2),
4117 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4120 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4123 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4124 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4125 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4126 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4127 !strconcat(OpcodeStr,
4128 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4131 // There's an AssertZext in the way of writing the store pattern
4132 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4135 let Predicates = [HasAVX] in
4136 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4138 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4141 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4142 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4143 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4144 (ins VR128:$src1, i32i8imm:$src2),
4145 !strconcat(OpcodeStr,
4146 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4148 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4149 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4150 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4151 !strconcat(OpcodeStr,
4152 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4153 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4154 addr:$dst)]>, OpSize;
4157 let Predicates = [HasAVX] in
4158 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4160 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4162 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4163 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4164 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4165 (ins VR128:$src1, i32i8imm:$src2),
4166 !strconcat(OpcodeStr,
4167 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4169 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4170 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4171 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4172 !strconcat(OpcodeStr,
4173 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4174 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4175 addr:$dst)]>, OpSize, REX_W;
4178 let Predicates = [HasAVX] in
4179 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4181 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4183 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4185 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4186 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4187 (ins VR128:$src1, i32i8imm:$src2),
4188 !strconcat(OpcodeStr,
4189 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4191 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4193 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4194 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4195 !strconcat(OpcodeStr,
4196 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4197 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4198 addr:$dst)]>, OpSize;
4201 let Predicates = [HasAVX] in {
4202 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4203 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4204 (ins VR128:$src1, i32i8imm:$src2),
4205 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4208 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4210 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4211 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4214 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4215 Requires<[HasSSE41]>;
4217 //===----------------------------------------------------------------------===//
4218 // SSE4.1 - Insert Instructions
4219 //===----------------------------------------------------------------------===//
4221 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4222 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4223 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4225 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4227 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4229 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4230 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4231 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4233 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4235 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4237 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4238 imm:$src3))]>, OpSize;
4241 let Predicates = [HasAVX] in
4242 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4243 let Constraints = "$src1 = $dst" in
4244 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4246 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4247 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4248 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4250 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4252 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4254 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4256 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4257 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4259 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4261 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4263 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4264 imm:$src3)))]>, OpSize;
4267 let Predicates = [HasAVX] in
4268 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4269 let Constraints = "$src1 = $dst" in
4270 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4272 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4273 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4274 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4276 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4278 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4280 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4282 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4283 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4285 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4287 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4289 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4290 imm:$src3)))]>, OpSize;
4293 let Predicates = [HasAVX] in
4294 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4295 let Constraints = "$src1 = $dst" in
4296 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4298 // insertps has a few different modes, there's the first two here below which
4299 // are optimized inserts that won't zero arbitrary elements in the destination
4300 // vector. The next one matches the intrinsic and could zero arbitrary elements
4301 // in the target vector.
4302 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4303 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4304 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
4306 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4308 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4310 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4312 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4313 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
4315 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4317 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4319 (X86insrtps VR128:$src1,
4320 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4321 imm:$src3))]>, OpSize;
4324 let Constraints = "$src1 = $dst" in
4325 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4326 let Predicates = [HasAVX] in
4327 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4329 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4330 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4332 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4333 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4334 Requires<[HasSSE41]>;
4336 //===----------------------------------------------------------------------===//
4337 // SSE4.1 - Round Instructions
4338 //===----------------------------------------------------------------------===//
4340 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4341 X86MemOperand x86memop, RegisterClass RC,
4342 PatFrag mem_frag32, PatFrag mem_frag64,
4343 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4344 // Intrinsic operation, reg.
4345 // Vector intrinsic operation, reg
4346 def PSr : SS4AIi8<opcps, MRMSrcReg,
4347 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4348 !strconcat(OpcodeStr,
4349 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4350 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4353 // Vector intrinsic operation, mem
4354 def PSm : Ii8<opcps, MRMSrcMem,
4355 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4356 !strconcat(OpcodeStr,
4357 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4359 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4361 Requires<[HasSSE41]>;
4363 // Vector intrinsic operation, reg
4364 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4365 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4366 !strconcat(OpcodeStr,
4367 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4368 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4371 // Vector intrinsic operation, mem
4372 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4373 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4374 !strconcat(OpcodeStr,
4375 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4377 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4381 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4382 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4383 // Intrinsic operation, reg.
4384 // Vector intrinsic operation, reg
4385 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4386 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4387 !strconcat(OpcodeStr,
4388 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4391 // Vector intrinsic operation, mem
4392 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4393 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4394 !strconcat(OpcodeStr,
4395 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4396 []>, TA, OpSize, Requires<[HasSSE41]>;
4398 // Vector intrinsic operation, reg
4399 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4400 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4401 !strconcat(OpcodeStr,
4402 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4405 // Vector intrinsic operation, mem
4406 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4407 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4408 !strconcat(OpcodeStr,
4409 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4413 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4416 Intrinsic F64Int, bit Is2Addr = 1> {
4417 // Intrinsic operation, reg.
4418 def SSr : SS4AIi8<opcss, MRMSrcReg,
4419 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4421 !strconcat(OpcodeStr,
4422 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4423 !strconcat(OpcodeStr,
4424 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4425 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4428 // Intrinsic operation, mem.
4429 def SSm : SS4AIi8<opcss, MRMSrcMem,
4430 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4432 !strconcat(OpcodeStr,
4433 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4434 !strconcat(OpcodeStr,
4435 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4437 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4440 // Intrinsic operation, reg.
4441 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4442 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4444 !strconcat(OpcodeStr,
4445 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4446 !strconcat(OpcodeStr,
4447 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4448 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4451 // Intrinsic operation, mem.
4452 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4453 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4455 !strconcat(OpcodeStr,
4456 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4457 !strconcat(OpcodeStr,
4458 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4460 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4464 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4466 // Intrinsic operation, reg.
4467 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4468 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4469 !strconcat(OpcodeStr,
4470 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4473 // Intrinsic operation, mem.
4474 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4475 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4476 !strconcat(OpcodeStr,
4477 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4480 // Intrinsic operation, reg.
4481 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4482 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4483 !strconcat(OpcodeStr,
4484 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4487 // Intrinsic operation, mem.
4488 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4489 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4490 !strconcat(OpcodeStr,
4491 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4495 // FP round - roundss, roundps, roundsd, roundpd
4496 let Predicates = [HasAVX] in {
4498 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4499 memopv4f32, memopv2f64,
4500 int_x86_sse41_round_ps,
4501 int_x86_sse41_round_pd>, VEX;
4502 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4503 memopv8f32, memopv4f64,
4504 int_x86_avx_round_ps_256,
4505 int_x86_avx_round_pd_256>, VEX;
4506 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4507 int_x86_sse41_round_ss,
4508 int_x86_sse41_round_sd, 0>, VEX_4V;
4510 // Instructions for the assembler
4511 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4513 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4515 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4518 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4519 memopv4f32, memopv2f64,
4520 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4521 let Constraints = "$src1 = $dst" in
4522 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4523 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4525 //===----------------------------------------------------------------------===//
4526 // SSE4.1 - Packed Bit Test
4527 //===----------------------------------------------------------------------===//
4529 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4530 // the intel intrinsic that corresponds to this.
4531 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4532 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4533 "vptest\t{$src2, $src1|$src1, $src2}",
4534 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4536 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4537 "vptest\t{$src2, $src1|$src1, $src2}",
4538 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4541 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4542 "vptest\t{$src2, $src1|$src1, $src2}",
4543 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4545 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4546 "vptest\t{$src2, $src1|$src1, $src2}",
4547 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4551 let Defs = [EFLAGS] in {
4552 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4553 "ptest \t{$src2, $src1|$src1, $src2}",
4554 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4556 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4557 "ptest \t{$src2, $src1|$src1, $src2}",
4558 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4562 // The bit test instructions below are AVX only
4563 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4564 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4565 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4566 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4567 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4568 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4569 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4570 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4574 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4575 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4576 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4577 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4578 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4581 //===----------------------------------------------------------------------===//
4582 // SSE4.1 - Misc Instructions
4583 //===----------------------------------------------------------------------===//
4585 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4586 "popcnt{w}\t{$src, $dst|$dst, $src}",
4587 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4588 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4589 "popcnt{w}\t{$src, $dst|$dst, $src}",
4590 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4592 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4593 "popcnt{l}\t{$src, $dst|$dst, $src}",
4594 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4595 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4596 "popcnt{l}\t{$src, $dst|$dst, $src}",
4597 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4599 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4600 "popcnt{q}\t{$src, $dst|$dst, $src}",
4601 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4602 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4603 "popcnt{q}\t{$src, $dst|$dst, $src}",
4604 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4608 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4609 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4610 Intrinsic IntId128> {
4611 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4614 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4615 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4617 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4620 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4623 let Predicates = [HasAVX] in
4624 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4625 int_x86_sse41_phminposuw>, VEX;
4626 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4627 int_x86_sse41_phminposuw>;
4629 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4630 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4631 Intrinsic IntId128, bit Is2Addr = 1> {
4632 let isCommutable = 1 in
4633 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4634 (ins VR128:$src1, VR128:$src2),
4636 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4637 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4638 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4639 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4640 (ins VR128:$src1, i128mem:$src2),
4642 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4643 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4645 (IntId128 VR128:$src1,
4646 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4649 let Predicates = [HasAVX] in {
4650 let isCommutable = 0 in
4651 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4653 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4655 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4657 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4659 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4661 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4663 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4665 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4667 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4669 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4671 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4675 let Constraints = "$src1 = $dst" in {
4676 let isCommutable = 0 in
4677 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4678 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4679 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4680 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4681 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4682 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4683 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4684 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4685 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4686 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4687 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4690 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4691 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4692 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4693 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4695 /// SS48I_binop_rm - Simple SSE41 binary operator.
4696 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4697 ValueType OpVT, bit Is2Addr = 1> {
4698 let isCommutable = 1 in
4699 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4700 (ins VR128:$src1, VR128:$src2),
4702 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4703 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4704 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4706 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4707 (ins VR128:$src1, i128mem:$src2),
4709 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4710 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4711 [(set VR128:$dst, (OpNode VR128:$src1,
4712 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4716 let Predicates = [HasAVX] in
4717 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4718 let Constraints = "$src1 = $dst" in
4719 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4721 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4722 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4723 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4724 X86MemOperand x86memop, bit Is2Addr = 1> {
4725 let isCommutable = 1 in
4726 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4727 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
4729 !strconcat(OpcodeStr,
4730 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4731 !strconcat(OpcodeStr,
4732 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4733 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4735 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4736 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
4738 !strconcat(OpcodeStr,
4739 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4740 !strconcat(OpcodeStr,
4741 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4744 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4748 let Predicates = [HasAVX] in {
4749 let isCommutable = 0 in {
4750 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4751 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4752 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4753 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4754 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4755 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4756 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4757 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4758 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4759 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4760 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4761 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4763 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4764 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4765 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4766 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4767 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4768 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4771 let Constraints = "$src1 = $dst" in {
4772 let isCommutable = 0 in {
4773 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4774 VR128, memopv16i8, i128mem>;
4775 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4776 VR128, memopv16i8, i128mem>;
4777 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4778 VR128, memopv16i8, i128mem>;
4779 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4780 VR128, memopv16i8, i128mem>;
4782 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4783 VR128, memopv16i8, i128mem>;
4784 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4785 VR128, memopv16i8, i128mem>;
4788 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4789 let Predicates = [HasAVX] in {
4790 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4791 RegisterClass RC, X86MemOperand x86memop,
4792 PatFrag mem_frag, Intrinsic IntId> {
4793 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4794 (ins RC:$src1, RC:$src2, RC:$src3),
4795 !strconcat(OpcodeStr,
4796 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4797 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4798 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4800 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4801 (ins RC:$src1, x86memop:$src2, RC:$src3),
4802 !strconcat(OpcodeStr,
4803 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4805 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4807 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4811 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4812 memopv16i8, int_x86_sse41_blendvpd>;
4813 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4814 memopv16i8, int_x86_sse41_blendvps>;
4815 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4816 memopv16i8, int_x86_sse41_pblendvb>;
4817 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4818 memopv32i8, int_x86_avx_blendv_pd_256>;
4819 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4820 memopv32i8, int_x86_avx_blendv_ps_256>;
4822 /// SS41I_ternary_int - SSE 4.1 ternary operator
4823 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4824 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4825 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4826 (ins VR128:$src1, VR128:$src2),
4827 !strconcat(OpcodeStr,
4828 "\t{$src2, $dst|$dst, $src2}"),
4829 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4832 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4833 (ins VR128:$src1, i128mem:$src2),
4834 !strconcat(OpcodeStr,
4835 "\t{$src2, $dst|$dst, $src2}"),
4838 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4842 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4843 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4844 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4846 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4847 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4849 let Predicates = [HasAVX] in
4850 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4851 "vmovntdqa\t{$src, $dst|$dst, $src}",
4852 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4854 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4855 "movntdqa\t{$src, $dst|$dst, $src}",
4856 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4859 //===----------------------------------------------------------------------===//
4860 // SSE4.2 - Compare Instructions
4861 //===----------------------------------------------------------------------===//
4863 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4864 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4865 Intrinsic IntId128, bit Is2Addr = 1> {
4866 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4867 (ins VR128:$src1, VR128:$src2),
4869 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4871 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4873 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4874 (ins VR128:$src1, i128mem:$src2),
4876 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4877 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4879 (IntId128 VR128:$src1,
4880 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4883 let Predicates = [HasAVX] in
4884 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4886 let Constraints = "$src1 = $dst" in
4887 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4889 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4890 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4891 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4892 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4894 //===----------------------------------------------------------------------===//
4895 // SSE4.2 - String/text Processing Instructions
4896 //===----------------------------------------------------------------------===//
4898 // Packed Compare Implicit Length Strings, Return Mask
4899 multiclass pseudo_pcmpistrm<string asm> {
4900 def REG : PseudoI<(outs VR128:$dst),
4901 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4902 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4904 def MEM : PseudoI<(outs VR128:$dst),
4905 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4906 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4907 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4910 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4911 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4912 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4915 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4916 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4917 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4918 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4919 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4920 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4921 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4924 let Defs = [XMM0, EFLAGS] in {
4925 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4926 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4927 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4928 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4929 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4930 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4933 // Packed Compare Explicit Length Strings, Return Mask
4934 multiclass pseudo_pcmpestrm<string asm> {
4935 def REG : PseudoI<(outs VR128:$dst),
4936 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4937 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4938 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4939 def MEM : PseudoI<(outs VR128:$dst),
4940 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4941 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4942 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4945 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4946 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4947 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4950 let Predicates = [HasAVX],
4951 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4952 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4953 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4954 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4955 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4956 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4957 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4960 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4961 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4962 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4963 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4964 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4965 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4966 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4969 // Packed Compare Implicit Length Strings, Return Index
4970 let Defs = [ECX, EFLAGS] in {
4971 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4972 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4973 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4974 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4975 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4976 (implicit EFLAGS)]>, OpSize;
4977 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4978 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4979 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4980 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4981 (implicit EFLAGS)]>, OpSize;
4985 let Predicates = [HasAVX] in {
4986 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4988 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4990 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4992 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4994 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4996 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5000 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5001 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5002 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5003 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5004 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5005 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5007 // Packed Compare Explicit Length Strings, Return Index
5008 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5009 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5010 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5011 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5012 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5013 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5014 (implicit EFLAGS)]>, OpSize;
5015 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5016 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5017 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5019 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5020 (implicit EFLAGS)]>, OpSize;
5024 let Predicates = [HasAVX] in {
5025 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5027 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5029 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5031 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5033 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5035 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5039 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5040 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5041 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5042 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5043 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5044 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5046 //===----------------------------------------------------------------------===//
5047 // SSE4.2 - CRC Instructions
5048 //===----------------------------------------------------------------------===//
5050 // No CRC instructions have AVX equivalents
5052 // crc intrinsic instruction
5053 // This set of instructions are only rm, the only difference is the size
5055 let Constraints = "$src1 = $dst" in {
5056 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5057 (ins GR32:$src1, i8mem:$src2),
5058 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5060 (int_x86_sse42_crc32_32_8 GR32:$src1,
5061 (load addr:$src2)))]>;
5062 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5063 (ins GR32:$src1, GR8:$src2),
5064 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5066 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5067 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5068 (ins GR32:$src1, i16mem:$src2),
5069 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5071 (int_x86_sse42_crc32_32_16 GR32:$src1,
5072 (load addr:$src2)))]>,
5074 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5075 (ins GR32:$src1, GR16:$src2),
5076 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5078 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5080 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5081 (ins GR32:$src1, i32mem:$src2),
5082 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5084 (int_x86_sse42_crc32_32_32 GR32:$src1,
5085 (load addr:$src2)))]>;
5086 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5087 (ins GR32:$src1, GR32:$src2),
5088 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5090 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5091 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5092 (ins GR64:$src1, i8mem:$src2),
5093 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5095 (int_x86_sse42_crc32_64_8 GR64:$src1,
5096 (load addr:$src2)))]>,
5098 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5099 (ins GR64:$src1, GR8:$src2),
5100 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5102 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5104 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5105 (ins GR64:$src1, i64mem:$src2),
5106 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5108 (int_x86_sse42_crc32_64_64 GR64:$src1,
5109 (load addr:$src2)))]>,
5111 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5112 (ins GR64:$src1, GR64:$src2),
5113 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5115 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5119 //===----------------------------------------------------------------------===//
5120 // AES-NI Instructions
5121 //===----------------------------------------------------------------------===//
5123 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5124 Intrinsic IntId128, bit Is2Addr = 1> {
5125 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5126 (ins VR128:$src1, VR128:$src2),
5128 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5129 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5130 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5132 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5133 (ins VR128:$src1, i128mem:$src2),
5135 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5136 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5138 (IntId128 VR128:$src1,
5139 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5142 // Perform One Round of an AES Encryption/Decryption Flow
5143 let Predicates = [HasAVX, HasAES] in {
5144 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5145 int_x86_aesni_aesenc, 0>, VEX_4V;
5146 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5147 int_x86_aesni_aesenclast, 0>, VEX_4V;
5148 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5149 int_x86_aesni_aesdec, 0>, VEX_4V;
5150 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5151 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5154 let Constraints = "$src1 = $dst" in {
5155 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5156 int_x86_aesni_aesenc>;
5157 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5158 int_x86_aesni_aesenclast>;
5159 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5160 int_x86_aesni_aesdec>;
5161 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5162 int_x86_aesni_aesdeclast>;
5165 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5166 (AESENCrr VR128:$src1, VR128:$src2)>;
5167 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5168 (AESENCrm VR128:$src1, addr:$src2)>;
5169 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5170 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5171 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5172 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5173 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5174 (AESDECrr VR128:$src1, VR128:$src2)>;
5175 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5176 (AESDECrm VR128:$src1, addr:$src2)>;
5177 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5178 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5179 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5180 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5182 // Perform the AES InvMixColumn Transformation
5183 let Predicates = [HasAVX, HasAES] in {
5184 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5186 "vaesimc\t{$src1, $dst|$dst, $src1}",
5188 (int_x86_aesni_aesimc VR128:$src1))]>,
5190 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5191 (ins i128mem:$src1),
5192 "vaesimc\t{$src1, $dst|$dst, $src1}",
5194 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5197 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5199 "aesimc\t{$src1, $dst|$dst, $src1}",
5201 (int_x86_aesni_aesimc VR128:$src1))]>,
5203 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5204 (ins i128mem:$src1),
5205 "aesimc\t{$src1, $dst|$dst, $src1}",
5207 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5210 // AES Round Key Generation Assist
5211 let Predicates = [HasAVX, HasAES] in {
5212 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5213 (ins VR128:$src1, i8imm:$src2),
5214 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5216 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5218 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5219 (ins i128mem:$src1, i8imm:$src2),
5220 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5222 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5226 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5227 (ins VR128:$src1, i8imm:$src2),
5228 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5230 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5232 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5233 (ins i128mem:$src1, i8imm:$src2),
5234 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5236 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5240 //===----------------------------------------------------------------------===//
5241 // CLMUL Instructions
5242 //===----------------------------------------------------------------------===//
5244 // Carry-less Multiplication instructions
5245 let Constraints = "$src1 = $dst" in {
5246 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5247 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5248 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5251 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5252 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5253 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5257 // AVX carry-less Multiplication instructions
5258 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5259 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5260 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5263 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5264 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5265 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5269 multiclass pclmul_alias<string asm, int immop> {
5270 def : InstAlias<!strconcat("pclmul", asm,
5271 "dq {$src, $dst|$dst, $src}"),
5272 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5274 def : InstAlias<!strconcat("pclmul", asm,
5275 "dq {$src, $dst|$dst, $src}"),
5276 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5278 def : InstAlias<!strconcat("vpclmul", asm,
5279 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5280 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5282 def : InstAlias<!strconcat("vpclmul", asm,
5283 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5284 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5286 defm : pclmul_alias<"hqhq", 0x11>;
5287 defm : pclmul_alias<"hqlq", 0x01>;
5288 defm : pclmul_alias<"lqhq", 0x10>;
5289 defm : pclmul_alias<"lqlq", 0x00>;
5291 //===----------------------------------------------------------------------===//
5293 //===----------------------------------------------------------------------===//
5295 //===----------------------------------------------------------------------===//
5296 // VBROADCAST - Load from memory and broadcast to all elements of the
5297 // destination operand
5299 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5300 X86MemOperand x86memop, Intrinsic Int> :
5301 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5302 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5303 [(set RC:$dst, (Int addr:$src))]>, VEX;
5305 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5306 int_x86_avx_vbroadcastss>;
5307 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5308 int_x86_avx_vbroadcastss_256>;
5309 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5310 int_x86_avx_vbroadcast_sd_256>;
5311 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5312 int_x86_avx_vbroadcastf128_pd_256>;
5314 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5315 (VBROADCASTF128 addr:$src)>;
5317 //===----------------------------------------------------------------------===//
5318 // VINSERTF128 - Insert packed floating-point values
5320 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5321 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5322 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5324 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5325 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5326 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5329 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5330 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5331 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5332 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5333 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5334 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5336 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5338 (VINSERTF128rr VR256:$src1, VR128:$src2,
5339 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5340 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5342 (VINSERTF128rr VR256:$src1, VR128:$src2,
5343 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5344 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5346 (VINSERTF128rr VR256:$src1, VR128:$src2,
5347 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5348 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5350 (VINSERTF128rr VR256:$src1, VR128:$src2,
5351 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5352 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
5354 (VINSERTF128rr VR256:$src1, VR128:$src2,
5355 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5356 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
5358 (VINSERTF128rr VR256:$src1, VR128:$src2,
5359 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5361 // Special COPY patterns
5362 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
5363 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5364 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
5365 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5366 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
5367 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5368 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
5369 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5370 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
5371 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5372 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
5373 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5375 //===----------------------------------------------------------------------===//
5376 // VEXTRACTF128 - Extract packed floating-point values
5378 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5379 (ins VR256:$src1, i8imm:$src2),
5380 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5382 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5383 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5384 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5387 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5388 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5389 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5390 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5391 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5392 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5394 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5395 (v4f32 (VEXTRACTF128rr
5396 (v8f32 VR256:$src1),
5397 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5398 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5399 (v2f64 (VEXTRACTF128rr
5400 (v4f64 VR256:$src1),
5401 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5402 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5403 (v4i32 (VEXTRACTF128rr
5404 (v8i32 VR256:$src1),
5405 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5406 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5407 (v2i64 (VEXTRACTF128rr
5408 (v4i64 VR256:$src1),
5409 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5410 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5411 (v8i16 (VEXTRACTF128rr
5412 (v16i16 VR256:$src1),
5413 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5414 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5415 (v16i8 (VEXTRACTF128rr
5416 (v32i8 VR256:$src1),
5417 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5419 //===----------------------------------------------------------------------===//
5420 // VMASKMOV - Conditional SIMD Packed Loads and Stores
5422 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5423 Intrinsic IntLd, Intrinsic IntLd256,
5424 Intrinsic IntSt, Intrinsic IntSt256,
5425 PatFrag pf128, PatFrag pf256> {
5426 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5427 (ins VR128:$src1, f128mem:$src2),
5428 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5429 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5431 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5432 (ins VR256:$src1, f256mem:$src2),
5433 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5434 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5436 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5437 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5438 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5439 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5440 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5441 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5443 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5446 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5447 int_x86_avx_maskload_ps,
5448 int_x86_avx_maskload_ps_256,
5449 int_x86_avx_maskstore_ps,
5450 int_x86_avx_maskstore_ps_256,
5451 memopv4f32, memopv8f32>;
5452 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5453 int_x86_avx_maskload_pd,
5454 int_x86_avx_maskload_pd_256,
5455 int_x86_avx_maskstore_pd,
5456 int_x86_avx_maskstore_pd_256,
5457 memopv2f64, memopv4f64>;
5459 //===----------------------------------------------------------------------===//
5460 // VPERM - Permute Floating-Point Values
5462 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5463 RegisterClass RC, X86MemOperand x86memop_f,
5464 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5465 Intrinsic IntVar, Intrinsic IntImm> {
5466 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5467 (ins RC:$src1, RC:$src2),
5468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5469 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5470 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5471 (ins RC:$src1, x86memop_i:$src2),
5472 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5473 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5475 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5476 (ins RC:$src1, i8imm:$src2),
5477 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5478 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5479 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5480 (ins x86memop_f:$src1, i8imm:$src2),
5481 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5482 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5485 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5486 memopv4f32, memopv4i32,
5487 int_x86_avx_vpermilvar_ps,
5488 int_x86_avx_vpermil_ps>;
5489 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5490 memopv8f32, memopv8i32,
5491 int_x86_avx_vpermilvar_ps_256,
5492 int_x86_avx_vpermil_ps_256>;
5493 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5494 memopv2f64, memopv2i64,
5495 int_x86_avx_vpermilvar_pd,
5496 int_x86_avx_vpermil_pd>;
5497 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5498 memopv4f64, memopv4i64,
5499 int_x86_avx_vpermilvar_pd_256,
5500 int_x86_avx_vpermil_pd_256>;
5502 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5503 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5504 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5506 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5507 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5508 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5511 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5512 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5513 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5514 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5515 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5516 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5518 def : Pat<(int_x86_avx_vperm2f128_ps_256
5519 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5520 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5521 def : Pat<(int_x86_avx_vperm2f128_pd_256
5522 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5523 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5524 def : Pat<(int_x86_avx_vperm2f128_si_256
5525 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5526 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5528 // Shuffle with VPERMIL instructions
5529 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
5530 (VPERMILPSYri VR256:$src1, imm:$imm)>;
5531 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
5532 (VPERMILPDYri VR256:$src1, imm:$imm)>;
5534 //===----------------------------------------------------------------------===//
5535 // VZERO - Zero YMM registers
5537 // Zero All YMM registers
5538 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5539 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5541 // Zero Upper bits of YMM registers
5542 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5543 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5545 //===----------------------------------------------------------------------===//
5546 // SSE Shuffle pattern fragments
5547 //===----------------------------------------------------------------------===//
5549 // This is part of a "work in progress" refactoring. The idea is that all
5550 // vector shuffles are going to be translated into target specific nodes and
5551 // directly matched by the patterns below (which can be changed along the way)
5552 // The AVX version of some but not all of them are described here, and more
5553 // should come in a near future.
5555 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5556 // SSE2 loads, which are always promoted to v2i64. The last one should match
5557 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5558 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5559 // we investigate further.
5560 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5562 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5563 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5565 (PSHUFDmi addr:$src1, imm:$imm)>;
5566 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5568 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5570 // Shuffle with PSHUFD instruction.
5571 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5572 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5573 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5574 (PSHUFDri VR128:$src1, imm:$imm)>;
5576 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5577 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5578 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5579 (PSHUFDri VR128:$src1, imm:$imm)>;
5581 // Shuffle with SHUFPD instruction.
5582 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5583 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5584 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5585 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5586 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5587 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5589 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5590 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5591 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5592 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5594 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5595 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5596 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5597 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5599 // Shuffle with SHUFPS instruction.
5600 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5601 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5602 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5603 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5604 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5605 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5607 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5608 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5609 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5610 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5612 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5613 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5614 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5615 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5616 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5617 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5619 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5620 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5621 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5622 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5624 // Shuffle with MOVHLPS instruction
5625 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5626 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5627 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5628 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5630 // Shuffle with MOVDDUP instruction
5631 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5632 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5633 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5634 (MOVDDUPrm addr:$src)>;
5636 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5637 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5638 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5639 (MOVDDUPrm addr:$src)>;
5641 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5642 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5643 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5644 (MOVDDUPrm addr:$src)>;
5646 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5647 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5648 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5649 (MOVDDUPrm addr:$src)>;
5651 def : Pat<(X86Movddup (bc_v2f64
5652 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5653 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5654 def : Pat<(X86Movddup (bc_v2f64
5655 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5656 (MOVDDUPrm addr:$src)>;
5659 // Shuffle with UNPCKLPS
5660 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5661 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5662 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5663 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5664 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5665 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5667 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5668 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5669 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5670 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5671 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5672 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5674 // Shuffle with UNPCKHPS
5675 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5676 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5677 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5678 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5680 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5681 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5682 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5683 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5685 // Shuffle with VUNPCKHPSY
5686 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
5687 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5688 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
5689 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5691 // Shuffle with UNPCKLPD
5692 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5693 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5694 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5695 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5696 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5697 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5699 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5700 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5701 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5702 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5703 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5704 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5706 // Shuffle with UNPCKHPD
5707 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5708 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5709 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5710 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5712 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5713 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5714 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5715 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5717 // Shuffle with VUNPCKHPDY
5718 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
5719 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5720 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
5721 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5723 // Shuffle with MOVLHPS
5724 def : Pat<(X86Movlhps VR128:$src1,
5725 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5726 (MOVHPSrm VR128:$src1, addr:$src2)>;
5727 def : Pat<(X86Movlhps VR128:$src1,
5728 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5729 (MOVHPSrm VR128:$src1, addr:$src2)>;
5730 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5731 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5732 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5733 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5734 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5735 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5737 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5738 // is during lowering, where it's not possible to recognize the load fold cause
5739 // it has two uses through a bitcast. One use disappears at isel time and the
5740 // fold opportunity reappears.
5741 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5742 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5744 // Shuffle with MOVLHPD
5745 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5746 (scalar_to_vector (loadf64 addr:$src2)))),
5747 (MOVHPDrm VR128:$src1, addr:$src2)>;
5749 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5750 // is during lowering, where it's not possible to recognize the load fold cause
5751 // it has two uses through a bitcast. One use disappears at isel time and the
5752 // fold opportunity reappears.
5753 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5754 (scalar_to_vector (loadf64 addr:$src2)))),
5755 (MOVHPDrm VR128:$src1, addr:$src2)>;
5757 // Shuffle with MOVSS
5758 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5759 (MOVSSrr VR128:$src1, FR32:$src2)>;
5760 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5761 (MOVSSrr (v4i32 VR128:$src1),
5762 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5763 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5764 (MOVSSrr (v4f32 VR128:$src1),
5765 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5766 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5767 // is during lowering, where it's not possible to recognize the load fold cause
5768 // it has two uses through a bitcast. One use disappears at isel time and the
5769 // fold opportunity reappears.
5770 def : Pat<(X86Movss VR128:$src1,
5771 (bc_v4i32 (v2i64 (load addr:$src2)))),
5772 (MOVLPSrm VR128:$src1, addr:$src2)>;
5774 // Shuffle with MOVSD
5775 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5776 (MOVSDrr VR128:$src1, FR64:$src2)>;
5777 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5778 (MOVSDrr (v2i64 VR128:$src1),
5779 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5780 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5781 (MOVSDrr (v2f64 VR128:$src1),
5782 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5783 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5784 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5785 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5786 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5788 // Shuffle with PSHUFHW
5789 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5790 (PSHUFHWri VR128:$src, imm:$imm)>;
5791 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5792 (PSHUFHWmi addr:$src, imm:$imm)>;
5794 // Shuffle with PSHUFLW
5795 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5796 (PSHUFLWri VR128:$src, imm:$imm)>;
5797 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5798 (PSHUFLWmi addr:$src, imm:$imm)>;
5800 // Shuffle with PALIGN
5801 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5802 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5803 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5804 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5805 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5806 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5807 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5808 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5810 // Shuffle with MOVLPS
5811 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5812 (MOVLPSrm VR128:$src1, addr:$src2)>;
5813 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5814 (MOVLPSrm VR128:$src1, addr:$src2)>;
5815 def : Pat<(X86Movlps VR128:$src1,
5816 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5817 (MOVLPSrm VR128:$src1, addr:$src2)>;
5818 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5819 // is during lowering, where it's not possible to recognize the load fold cause
5820 // it has two uses through a bitcast. One use disappears at isel time and the
5821 // fold opportunity reappears.
5822 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5823 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5825 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5826 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5828 // Shuffle with MOVLPD
5829 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5830 (MOVLPDrm VR128:$src1, addr:$src2)>;
5831 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5832 (MOVLPDrm VR128:$src1, addr:$src2)>;
5833 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5834 (scalar_to_vector (loadf64 addr:$src2)))),
5835 (MOVLPDrm VR128:$src1, addr:$src2)>;
5837 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5838 def : Pat<(store (f64 (vector_extract
5839 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5840 (MOVHPSmr addr:$dst, VR128:$src)>;
5841 def : Pat<(store (f64 (vector_extract
5842 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5843 (MOVHPDmr addr:$dst, VR128:$src)>;
5845 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5846 (MOVLPSmr addr:$src1, VR128:$src2)>;
5847 def : Pat<(store (v4i32 (X86Movlps
5848 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5849 (MOVLPSmr addr:$src1, VR128:$src2)>;
5851 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5852 (MOVLPDmr addr:$src1, VR128:$src2)>;
5853 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5854 (MOVLPDmr addr:$src1, VR128:$src2)>;