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, $src}">, XD,
500 // The assembler can recognize rr 64-bit instructions by seeing a rxx
501 // register, but the same isn't true when only using memory operands,
502 // provide other assembly "l" and "q" forms to address this explicitly
503 // where appropriate to do so.
504 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
506 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
508 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
510 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
512 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
515 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
516 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
517 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
518 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
519 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
520 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
521 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
522 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
523 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
524 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
525 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
526 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
527 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
528 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
529 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
530 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
532 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
533 // and/or XMM operand(s).
535 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
536 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
538 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
539 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
540 [(set DstRC:$dst, (Int SrcRC:$src))]>;
541 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
542 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
543 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
546 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
547 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
548 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
549 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
551 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
552 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
553 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
554 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
555 (ins DstRC:$src1, x86memop:$src2),
557 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
558 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
559 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
562 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
563 f32mem, load, "cvtss2si">, XS, VEX;
564 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
565 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
567 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
568 f128mem, load, "cvtsd2si">, XD, VEX;
569 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
570 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
573 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
574 // Get rid of this hack or rename the intrinsics, there are several
575 // intructions that only match with the intrinsic form, why create duplicates
576 // to let them be recognized by the assembler?
577 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
578 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
579 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
580 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
581 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
582 f32mem, load, "cvtss2si">, XS;
583 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
584 f32mem, load, "cvtss2si{q}">, XS, REX_W;
585 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
586 f128mem, load, "cvtsd2si{l}">, XD;
587 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
588 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
591 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
592 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
593 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
594 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
596 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
597 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
598 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
599 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
602 let Constraints = "$src1 = $dst" in {
603 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
604 int_x86_sse_cvtsi2ss, i32mem, loadi32,
606 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
607 int_x86_sse_cvtsi642ss, i64mem, loadi64,
608 "cvtsi2ss{q}">, XS, REX_W;
609 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
610 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
612 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
613 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
614 "cvtsi2sd">, XD, REX_W;
619 // Aliases for intrinsics
620 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
621 f32mem, load, "cvttss2si">, XS, VEX;
622 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
623 int_x86_sse_cvttss2si64, f32mem, load,
624 "cvttss2si">, XS, VEX, VEX_W;
625 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
626 f128mem, load, "cvttsd2si">, XD, VEX;
627 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
628 int_x86_sse2_cvttsd2si64, f128mem, load,
629 "cvttsd2si">, XD, VEX, VEX_W;
630 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
631 f32mem, load, "cvttss2si">, XS;
632 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
633 int_x86_sse_cvttss2si64, f32mem, load,
634 "cvttss2si{q}">, XS, REX_W;
635 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
636 f128mem, load, "cvttsd2si">, XD;
637 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
638 int_x86_sse2_cvttsd2si64, f128mem, load,
639 "cvttsd2si{q}">, XD, REX_W;
641 let Pattern = []<dag> in {
642 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
643 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
644 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
645 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
647 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
648 "cvtdq2ps\t{$src, $dst|$dst, $src}",
649 SSEPackedSingle>, TB, VEX;
650 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
651 "cvtdq2ps\t{$src, $dst|$dst, $src}",
652 SSEPackedSingle>, TB, VEX;
654 let Pattern = []<dag> in {
655 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
656 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
657 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
658 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
659 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
660 "cvtdq2ps\t{$src, $dst|$dst, $src}",
661 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
666 // Convert scalar double to scalar single
667 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
668 (ins FR64:$src1, FR64:$src2),
669 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
671 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
672 (ins FR64:$src1, f64mem:$src2),
673 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
674 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
675 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
678 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
679 "cvtsd2ss\t{$src, $dst|$dst, $src}",
680 [(set FR32:$dst, (fround FR64:$src))]>;
681 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
682 "cvtsd2ss\t{$src, $dst|$dst, $src}",
683 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
684 Requires<[HasSSE2, OptForSize]>;
686 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
687 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
689 let Constraints = "$src1 = $dst" in
690 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
691 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
693 // Convert scalar single to scalar double
694 // SSE2 instructions with XS prefix
695 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
696 (ins FR32:$src1, FR32:$src2),
697 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
698 []>, XS, Requires<[HasAVX]>, VEX_4V;
699 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
700 (ins FR32:$src1, f32mem:$src2),
701 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
702 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
703 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
706 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
707 "cvtss2sd\t{$src, $dst|$dst, $src}",
708 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
710 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
711 "cvtss2sd\t{$src, $dst|$dst, $src}",
712 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
713 Requires<[HasSSE2, OptForSize]>;
715 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
719 VR128:$src2))]>, XS, VEX_4V,
721 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
722 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
725 (load addr:$src2)))]>, XS, VEX_4V,
727 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
728 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
729 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
730 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
731 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
734 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
735 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
738 (load addr:$src2)))]>, XS,
742 def : Pat<(extloadf32 addr:$src),
743 (CVTSS2SDrr (MOVSSrm addr:$src))>,
744 Requires<[HasSSE2, OptForSpeed]>;
746 // Convert doubleword to packed single/double fp
747 // SSE2 instructions without OpSize prefix
748 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
749 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
750 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
751 TB, VEX, Requires<[HasAVX]>;
752 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
753 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
754 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
755 (bitconvert (memopv2i64 addr:$src))))]>,
756 TB, VEX, Requires<[HasAVX]>;
757 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
758 "cvtdq2ps\t{$src, $dst|$dst, $src}",
759 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
760 TB, Requires<[HasSSE2]>;
761 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
762 "cvtdq2ps\t{$src, $dst|$dst, $src}",
763 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
764 (bitconvert (memopv2i64 addr:$src))))]>,
765 TB, Requires<[HasSSE2]>;
767 // FIXME: why the non-intrinsic version is described as SSE3?
768 // SSE2 instructions with XS prefix
769 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
770 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
771 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
772 XS, VEX, Requires<[HasAVX]>;
773 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
774 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
775 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
776 (bitconvert (memopv2i64 addr:$src))))]>,
777 XS, VEX, Requires<[HasAVX]>;
778 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
779 "cvtdq2pd\t{$src, $dst|$dst, $src}",
780 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
781 XS, Requires<[HasSSE2]>;
782 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
783 "cvtdq2pd\t{$src, $dst|$dst, $src}",
784 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
785 (bitconvert (memopv2i64 addr:$src))))]>,
786 XS, Requires<[HasSSE2]>;
789 // Convert packed single/double fp to doubleword
790 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
791 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
792 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
794 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
796 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
798 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
800 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
803 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
804 "cvtps2dq\t{$src, $dst|$dst, $src}",
805 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
807 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
809 "cvtps2dq\t{$src, $dst|$dst, $src}",
810 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
811 (memop addr:$src)))]>, VEX;
812 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
813 "cvtps2dq\t{$src, $dst|$dst, $src}",
814 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
815 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
816 "cvtps2dq\t{$src, $dst|$dst, $src}",
817 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
818 (memop addr:$src)))]>;
820 // SSE2 packed instructions with XD prefix
821 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
822 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
823 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
824 XD, VEX, Requires<[HasAVX]>;
825 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
826 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
827 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
828 (memop addr:$src)))]>,
829 XD, VEX, Requires<[HasAVX]>;
830 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
831 "cvtpd2dq\t{$src, $dst|$dst, $src}",
832 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
833 XD, Requires<[HasSSE2]>;
834 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
835 "cvtpd2dq\t{$src, $dst|$dst, $src}",
836 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
837 (memop addr:$src)))]>,
838 XD, Requires<[HasSSE2]>;
841 // Convert with truncation packed single/double fp to doubleword
842 // SSE2 packed instructions with XS prefix
843 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
844 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
845 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
847 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
848 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
849 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
850 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
851 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
852 "cvttps2dq\t{$src, $dst|$dst, $src}",
854 (int_x86_sse2_cvttps2dq VR128:$src))]>;
855 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
856 "cvttps2dq\t{$src, $dst|$dst, $src}",
858 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
861 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
862 "vcvttps2dq\t{$src, $dst|$dst, $src}",
864 (int_x86_sse2_cvttps2dq VR128:$src))]>,
865 XS, VEX, Requires<[HasAVX]>;
866 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
867 "vcvttps2dq\t{$src, $dst|$dst, $src}",
868 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
869 (memop addr:$src)))]>,
870 XS, VEX, Requires<[HasAVX]>;
872 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
874 "cvttpd2dq\t{$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
877 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
879 "cvttpd2dq\t{$src, $dst|$dst, $src}",
880 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
881 (memop addr:$src)))]>, VEX;
882 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
883 "cvttpd2dq\t{$src, $dst|$dst, $src}",
884 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
885 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
886 "cvttpd2dq\t{$src, $dst|$dst, $src}",
887 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
888 (memop addr:$src)))]>;
890 // The assembler can recognize rr 256-bit instructions by seeing a ymm
891 // register, but the same isn't true when using memory operands instead.
892 // Provide other assembly rr and rm forms to address this explicitly.
893 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
894 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
895 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
896 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
899 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
900 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
901 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
902 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
905 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
906 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
907 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
908 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
910 // Convert packed single to packed double
911 let Predicates = [HasAVX] in {
912 // SSE2 instructions without OpSize prefix
913 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
914 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
915 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
916 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
917 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
918 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
919 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
920 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
922 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
923 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
924 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
925 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
927 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
928 "vcvtps2pd\t{$src, $dst|$dst, $src}",
929 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
930 VEX, Requires<[HasAVX]>;
931 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
932 "vcvtps2pd\t{$src, $dst|$dst, $src}",
933 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
934 (load addr:$src)))]>,
935 VEX, Requires<[HasAVX]>;
936 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
937 "cvtps2pd\t{$src, $dst|$dst, $src}",
938 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
939 TB, Requires<[HasSSE2]>;
940 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
941 "cvtps2pd\t{$src, $dst|$dst, $src}",
942 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
943 (load addr:$src)))]>,
944 TB, Requires<[HasSSE2]>;
946 // Convert packed double to packed single
947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
948 // register, but the same isn't true when using memory operands instead.
949 // Provide other assembly rr and rm forms to address this explicitly.
950 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
952 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
953 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
963 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
964 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
965 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
966 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
967 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
968 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
969 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
972 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
973 "cvtpd2ps\t{$src, $dst|$dst, $src}",
974 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
975 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
977 "cvtpd2ps\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
979 (memop addr:$src)))]>;
980 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
981 "cvtpd2ps\t{$src, $dst|$dst, $src}",
982 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
983 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
984 "cvtpd2ps\t{$src, $dst|$dst, $src}",
985 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
986 (memop addr:$src)))]>;
988 // AVX 256-bit register conversion intrinsics
989 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
990 // whenever possible to avoid declaring two versions of each one.
991 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
992 (VCVTDQ2PSYrr VR256:$src)>;
993 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
994 (VCVTDQ2PSYrm addr:$src)>;
996 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
997 (VCVTPD2PSYrr VR256:$src)>;
998 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
999 (VCVTPD2PSYrm addr:$src)>;
1001 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1002 (VCVTPS2DQYrr VR256:$src)>;
1003 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1004 (VCVTPS2DQYrm addr:$src)>;
1006 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1007 (VCVTPS2PDYrr VR128:$src)>;
1008 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1009 (VCVTPS2PDYrm addr:$src)>;
1011 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1012 (VCVTTPD2DQYrr VR256:$src)>;
1013 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1014 (VCVTTPD2DQYrm addr:$src)>;
1016 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1017 (VCVTTPS2DQYrr VR256:$src)>;
1018 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1019 (VCVTTPS2DQYrm addr:$src)>;
1021 //===----------------------------------------------------------------------===//
1022 // SSE 1 & 2 - Compare Instructions
1023 //===----------------------------------------------------------------------===//
1025 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1026 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1027 string asm, string asm_alt> {
1028 let isAsmParserOnly = 1 in {
1029 def rr : SIi8<0xC2, MRMSrcReg,
1030 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1033 def rm : SIi8<0xC2, MRMSrcMem,
1034 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1038 // Accept explicit immediate argument form instead of comparison code.
1039 def rr_alt : SIi8<0xC2, MRMSrcReg,
1040 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1043 def rm_alt : SIi8<0xC2, MRMSrcMem,
1044 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1048 let neverHasSideEffects = 1 in {
1049 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1050 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1051 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1053 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1054 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1055 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1059 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1060 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1061 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1062 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1063 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1064 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1065 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1068 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1069 Intrinsic Int, string asm> {
1070 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1071 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1072 [(set VR128:$dst, (Int VR128:$src1,
1073 VR128:$src, imm:$cc))]>;
1074 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1075 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1076 [(set VR128:$dst, (Int VR128:$src1,
1077 (load addr:$src), imm:$cc))]>;
1080 // Aliases to match intrinsics which expect XMM operand(s).
1081 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1082 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1084 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1085 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1087 let Constraints = "$src1 = $dst" in {
1088 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1089 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1090 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1091 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1095 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1096 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1097 ValueType vt, X86MemOperand x86memop,
1098 PatFrag ld_frag, string OpcodeStr, Domain d> {
1099 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1100 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1101 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1102 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1103 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1104 [(set EFLAGS, (OpNode (vt RC:$src1),
1105 (ld_frag addr:$src2)))], d>;
1108 let Defs = [EFLAGS] in {
1109 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1110 "ucomiss", SSEPackedSingle>, VEX;
1111 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1112 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1113 let Pattern = []<dag> in {
1114 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1115 "comiss", SSEPackedSingle>, VEX;
1116 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1117 "comisd", SSEPackedDouble>, OpSize, VEX;
1120 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1121 load, "ucomiss", SSEPackedSingle>, VEX;
1122 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1123 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1125 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1126 load, "comiss", SSEPackedSingle>, VEX;
1127 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1128 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1129 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1130 "ucomiss", SSEPackedSingle>, TB;
1131 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1132 "ucomisd", SSEPackedDouble>, TB, OpSize;
1134 let Pattern = []<dag> in {
1135 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1136 "comiss", SSEPackedSingle>, TB;
1137 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1138 "comisd", SSEPackedDouble>, TB, OpSize;
1141 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1142 load, "ucomiss", SSEPackedSingle>, TB;
1143 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1144 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1146 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1147 "comiss", SSEPackedSingle>, TB;
1148 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1149 "comisd", SSEPackedDouble>, TB, OpSize;
1150 } // Defs = [EFLAGS]
1152 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1153 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1154 Intrinsic Int, string asm, string asm_alt,
1156 let isAsmParserOnly = 1 in {
1157 def rri : PIi8<0xC2, MRMSrcReg,
1158 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1159 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1160 def rmi : PIi8<0xC2, MRMSrcMem,
1161 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1162 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1165 // Accept explicit immediate argument form instead of comparison code.
1166 def rri_alt : PIi8<0xC2, MRMSrcReg,
1167 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1169 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1170 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1174 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1175 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1176 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1177 SSEPackedSingle>, VEX_4V;
1178 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1179 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1180 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1181 SSEPackedDouble>, OpSize, VEX_4V;
1182 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1183 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1184 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1185 SSEPackedSingle>, VEX_4V;
1186 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1187 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1188 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1189 SSEPackedDouble>, OpSize, VEX_4V;
1190 let Constraints = "$src1 = $dst" in {
1191 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1192 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1193 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1194 SSEPackedSingle>, TB;
1195 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1196 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1197 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1198 SSEPackedDouble>, TB, OpSize;
1201 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1202 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1203 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1204 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1205 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1206 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1207 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1208 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1210 //===----------------------------------------------------------------------===//
1211 // SSE 1 & 2 - Shuffle Instructions
1212 //===----------------------------------------------------------------------===//
1214 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1215 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1216 ValueType vt, string asm, PatFrag mem_frag,
1217 Domain d, bit IsConvertibleToThreeAddress = 0> {
1218 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1219 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1220 [(set RC:$dst, (vt (shufp:$src3
1221 RC:$src1, (mem_frag addr:$src2))))], d>;
1222 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1223 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1224 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1226 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1229 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1230 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1231 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1232 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1233 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1234 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1235 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1236 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1237 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1238 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1239 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1240 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1242 let Constraints = "$src1 = $dst" in {
1243 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1244 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1245 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1247 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1248 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1249 memopv2f64, SSEPackedDouble>, TB, OpSize;
1252 //===----------------------------------------------------------------------===//
1253 // SSE 1 & 2 - Unpack Instructions
1254 //===----------------------------------------------------------------------===//
1256 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1257 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1258 PatFrag mem_frag, RegisterClass RC,
1259 X86MemOperand x86memop, string asm,
1261 def rr : PI<opc, MRMSrcReg,
1262 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1264 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1265 def rm : PI<opc, MRMSrcMem,
1266 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1268 (vt (OpNode RC:$src1,
1269 (mem_frag addr:$src2))))], d>;
1272 let AddedComplexity = 10 in {
1273 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1274 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1275 SSEPackedSingle>, VEX_4V;
1276 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1277 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1278 SSEPackedDouble>, OpSize, VEX_4V;
1279 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1280 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1281 SSEPackedSingle>, VEX_4V;
1282 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1283 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1284 SSEPackedDouble>, OpSize, VEX_4V;
1286 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1287 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1288 SSEPackedSingle>, VEX_4V;
1289 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1290 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1291 SSEPackedDouble>, OpSize, VEX_4V;
1292 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1293 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1294 SSEPackedSingle>, VEX_4V;
1295 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1296 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1297 SSEPackedDouble>, OpSize, VEX_4V;
1299 let Constraints = "$src1 = $dst" in {
1300 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1301 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1302 SSEPackedSingle>, TB;
1303 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1304 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1305 SSEPackedDouble>, TB, OpSize;
1306 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1307 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1308 SSEPackedSingle>, TB;
1309 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1310 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1311 SSEPackedDouble>, TB, OpSize;
1312 } // Constraints = "$src1 = $dst"
1313 } // AddedComplexity
1315 //===----------------------------------------------------------------------===//
1316 // SSE 1 & 2 - Extract Floating-Point Sign mask
1317 //===----------------------------------------------------------------------===//
1319 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1320 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1322 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1323 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1324 [(set GR32:$dst, (Int RC:$src))], d>;
1325 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1326 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1330 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1331 SSEPackedSingle>, TB;
1332 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1333 SSEPackedDouble>, TB, OpSize;
1335 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1336 "movmskps", SSEPackedSingle>, VEX;
1337 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1338 "movmskpd", SSEPackedDouble>, OpSize,
1340 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1341 "movmskps", SSEPackedSingle>, VEX;
1342 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1343 "movmskpd", SSEPackedDouble>, OpSize,
1347 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1348 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1349 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1350 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1352 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1353 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1354 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1355 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1358 //===----------------------------------------------------------------------===//
1359 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1360 //===----------------------------------------------------------------------===//
1362 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1363 // names that start with 'Fs'.
1365 // Alias instructions that map fld0 to pxor for sse.
1366 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1367 canFoldAsLoad = 1 in {
1368 // FIXME: Set encoding to pseudo!
1369 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1370 [(set FR32:$dst, fp32imm0)]>,
1371 Requires<[HasSSE1]>, TB, OpSize;
1372 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1373 [(set FR64:$dst, fpimm0)]>,
1374 Requires<[HasSSE2]>, TB, OpSize;
1375 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1376 [(set FR32:$dst, fp32imm0)]>,
1377 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1378 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1379 [(set FR64:$dst, fpimm0)]>,
1380 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1383 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1384 // bits are disregarded.
1385 let neverHasSideEffects = 1 in {
1386 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1387 "movaps\t{$src, $dst|$dst, $src}", []>;
1388 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1389 "movapd\t{$src, $dst|$dst, $src}", []>;
1392 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1393 // bits are disregarded.
1394 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1395 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1396 "movaps\t{$src, $dst|$dst, $src}",
1397 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1398 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1399 "movapd\t{$src, $dst|$dst, $src}",
1400 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1403 //===----------------------------------------------------------------------===//
1404 // SSE 1 & 2 - Logical Instructions
1405 //===----------------------------------------------------------------------===//
1407 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1409 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1411 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1412 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1414 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1415 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1417 let Constraints = "$src1 = $dst" in {
1418 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1419 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1421 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1422 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1426 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1427 let mayLoad = 0 in {
1428 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1429 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1430 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1433 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1434 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1436 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1438 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1439 SDNode OpNode, int HasPat = 0,
1440 list<list<dag>> Pattern = []> {
1441 let Pattern = []<dag> in {
1442 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1443 !strconcat(OpcodeStr, "ps"), f128mem,
1444 !if(HasPat, Pattern[0], // rr
1445 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1447 !if(HasPat, Pattern[2], // rm
1448 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1449 (memopv2i64 addr:$src2)))]), 0>,
1452 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1453 !strconcat(OpcodeStr, "pd"), f128mem,
1454 !if(HasPat, Pattern[1], // rr
1455 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1458 !if(HasPat, Pattern[3], // rm
1459 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1460 (memopv2i64 addr:$src2)))]), 0>,
1463 let Constraints = "$src1 = $dst" in {
1464 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1465 !strconcat(OpcodeStr, "ps"), f128mem,
1466 !if(HasPat, Pattern[0], // rr
1467 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1469 !if(HasPat, Pattern[2], // rm
1470 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1471 (memopv2i64 addr:$src2)))])>, TB;
1473 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1474 !strconcat(OpcodeStr, "pd"), f128mem,
1475 !if(HasPat, Pattern[1], // rr
1476 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1479 !if(HasPat, Pattern[3], // rm
1480 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1481 (memopv2i64 addr:$src2)))])>,
1486 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1488 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1489 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1490 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1492 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1493 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1496 // AVX 256-bit packed logical ops forms
1497 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1498 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1499 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1500 let isCommutable = 0 in
1501 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1503 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1504 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1505 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1506 let isCommutable = 0 in
1507 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1509 [(set VR128:$dst, (X86pandn VR128:$src1, VR128:$src2))],
1513 [(set VR128:$dst, (X86pandn VR128:$src1, (memopv2i64 addr:$src2)))],
1517 //===----------------------------------------------------------------------===//
1518 // SSE 1 & 2 - Arithmetic Instructions
1519 //===----------------------------------------------------------------------===//
1521 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1524 /// In addition, we also have a special variant of the scalar form here to
1525 /// represent the associated intrinsic operation. This form is unlike the
1526 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1527 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1529 /// These three forms can each be reg+reg or reg+mem.
1532 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1534 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1536 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1537 OpNode, FR32, f32mem, Is2Addr>, XS;
1538 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1539 OpNode, FR64, f64mem, Is2Addr>, XD;
1542 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1544 let mayLoad = 0 in {
1545 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1546 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1547 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1548 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1552 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1554 let mayLoad = 0 in {
1555 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1556 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1557 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1558 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1562 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1564 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1565 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1566 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1567 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1570 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1572 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1573 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1574 SSEPackedSingle, Is2Addr>, TB;
1576 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1577 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1578 SSEPackedDouble, Is2Addr>, TB, OpSize;
1581 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1582 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1583 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1584 SSEPackedSingle, 0>, TB;
1586 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1587 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1588 SSEPackedDouble, 0>, TB, OpSize;
1591 // Binary Arithmetic instructions
1592 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1593 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1594 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1595 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1596 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1597 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1598 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1599 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1601 let isCommutable = 0 in {
1602 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1603 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1604 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1605 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1606 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1607 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1608 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1609 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1610 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1611 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1612 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1613 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1614 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1615 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1616 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1617 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1618 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1619 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1620 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1621 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1624 let Constraints = "$src1 = $dst" in {
1625 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1626 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1627 basic_sse12_fp_binop_s_int<0x58, "add">;
1628 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1629 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1630 basic_sse12_fp_binop_s_int<0x59, "mul">;
1632 let isCommutable = 0 in {
1633 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1634 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1635 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1636 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1637 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1638 basic_sse12_fp_binop_s_int<0x5E, "div">;
1639 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1640 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1641 basic_sse12_fp_binop_s_int<0x5F, "max">,
1642 basic_sse12_fp_binop_p_int<0x5F, "max">;
1643 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1644 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1645 basic_sse12_fp_binop_s_int<0x5D, "min">,
1646 basic_sse12_fp_binop_p_int<0x5D, "min">;
1651 /// In addition, we also have a special variant of the scalar form here to
1652 /// represent the associated intrinsic operation. This form is unlike the
1653 /// plain scalar form, in that it takes an entire vector (instead of a
1654 /// scalar) and leaves the top elements undefined.
1656 /// And, we have a special variant form for a full-vector intrinsic form.
1658 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1659 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1660 SDNode OpNode, Intrinsic F32Int> {
1661 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1662 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1663 [(set FR32:$dst, (OpNode FR32:$src))]>;
1664 // For scalar unary operations, fold a load into the operation
1665 // only in OptForSize mode. It eliminates an instruction, but it also
1666 // eliminates a whole-register clobber (the load), so it introduces a
1667 // partial register update condition.
1668 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1669 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1670 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1671 Requires<[HasSSE1, OptForSize]>;
1672 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1673 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1674 [(set VR128:$dst, (F32Int VR128:$src))]>;
1675 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1676 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1677 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1680 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1681 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1682 SDNode OpNode, Intrinsic F32Int> {
1683 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1684 !strconcat(OpcodeStr,
1685 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1686 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1687 !strconcat(OpcodeStr,
1688 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1689 []>, XS, Requires<[HasAVX, OptForSize]>;
1690 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1691 !strconcat(OpcodeStr,
1692 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1693 [(set VR128:$dst, (F32Int VR128:$src))]>;
1694 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1695 !strconcat(OpcodeStr,
1696 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1697 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1700 /// sse1_fp_unop_p - SSE1 unops in packed form.
1701 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1702 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1703 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1704 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1705 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1706 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1707 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1710 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1711 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1712 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1713 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1714 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1715 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1716 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1717 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1720 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1721 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1722 Intrinsic V4F32Int> {
1723 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1724 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1725 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1726 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1727 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1728 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1731 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1732 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1733 Intrinsic V4F32Int> {
1734 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1735 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1736 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1737 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1738 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1739 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1742 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1743 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1744 SDNode OpNode, Intrinsic F64Int> {
1745 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1746 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1747 [(set FR64:$dst, (OpNode FR64:$src))]>;
1748 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1749 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1750 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1751 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1752 Requires<[HasSSE2, OptForSize]>;
1753 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1754 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1755 [(set VR128:$dst, (F64Int VR128:$src))]>;
1756 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1757 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1758 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1761 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1762 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1763 SDNode OpNode, Intrinsic F64Int> {
1764 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1765 !strconcat(OpcodeStr,
1766 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1767 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1768 (ins FR64:$src1, f64mem:$src2),
1769 !strconcat(OpcodeStr,
1770 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1771 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1772 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1773 [(set VR128:$dst, (F64Int VR128:$src))]>;
1774 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1775 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1776 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1779 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1780 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1782 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1783 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1784 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1785 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1786 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1787 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1790 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1791 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1792 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1793 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1794 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1795 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1796 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1797 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1800 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1801 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1802 Intrinsic V2F64Int> {
1803 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1804 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1805 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1806 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1807 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1808 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1811 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1812 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1813 Intrinsic V2F64Int> {
1814 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1815 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1816 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1817 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1818 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1819 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1822 let Predicates = [HasAVX] in {
1824 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1825 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1828 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1829 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1830 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1831 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1832 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1833 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1834 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1835 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1838 // Reciprocal approximations. Note that these typically require refinement
1839 // in order to obtain suitable precision.
1840 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1841 int_x86_sse_rsqrt_ss>, VEX_4V;
1842 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1843 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1844 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1845 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1847 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1849 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1850 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1851 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1852 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1856 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1857 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1858 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1859 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1860 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1861 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1863 // Reciprocal approximations. Note that these typically require refinement
1864 // in order to obtain suitable precision.
1865 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1866 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1867 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1868 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1869 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1870 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1872 // There is no f64 version of the reciprocal approximation instructions.
1874 //===----------------------------------------------------------------------===//
1875 // SSE 1 & 2 - Non-temporal stores
1876 //===----------------------------------------------------------------------===//
1878 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1879 (ins i128mem:$dst, VR128:$src),
1880 "movntps\t{$src, $dst|$dst, $src}",
1881 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1882 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1883 (ins i128mem:$dst, VR128:$src),
1884 "movntpd\t{$src, $dst|$dst, $src}",
1885 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1887 let ExeDomain = SSEPackedInt in
1888 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1889 (ins f128mem:$dst, VR128:$src),
1890 "movntdq\t{$src, $dst|$dst, $src}",
1891 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1893 let AddedComplexity = 400 in { // Prefer non-temporal versions
1894 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1895 (ins f128mem:$dst, VR128:$src),
1896 "movntps\t{$src, $dst|$dst, $src}",
1897 [(alignednontemporalstore (v4f32 VR128:$src),
1899 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1900 (ins f128mem:$dst, VR128:$src),
1901 "movntpd\t{$src, $dst|$dst, $src}",
1902 [(alignednontemporalstore (v2f64 VR128:$src),
1904 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1905 (ins f128mem:$dst, VR128:$src),
1906 "movntdq\t{$src, $dst|$dst, $src}",
1907 [(alignednontemporalstore (v2f64 VR128:$src),
1909 let ExeDomain = SSEPackedInt in
1910 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1911 (ins f128mem:$dst, VR128:$src),
1912 "movntdq\t{$src, $dst|$dst, $src}",
1913 [(alignednontemporalstore (v4f32 VR128:$src),
1916 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1917 (ins f256mem:$dst, VR256:$src),
1918 "movntps\t{$src, $dst|$dst, $src}",
1919 [(alignednontemporalstore (v8f32 VR256:$src),
1921 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1922 (ins f256mem:$dst, VR256:$src),
1923 "movntpd\t{$src, $dst|$dst, $src}",
1924 [(alignednontemporalstore (v4f64 VR256:$src),
1926 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1927 (ins f256mem:$dst, VR256:$src),
1928 "movntdq\t{$src, $dst|$dst, $src}",
1929 [(alignednontemporalstore (v4f64 VR256:$src),
1931 let ExeDomain = SSEPackedInt in
1932 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1933 (ins f256mem:$dst, VR256:$src),
1934 "movntdq\t{$src, $dst|$dst, $src}",
1935 [(alignednontemporalstore (v8f32 VR256:$src),
1939 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1940 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1941 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
1942 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
1943 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
1944 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
1946 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1947 "movntps\t{$src, $dst|$dst, $src}",
1948 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
1949 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
1950 "movntpd\t{$src, $dst|$dst, $src}",
1951 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
1953 let ExeDomain = SSEPackedInt in
1954 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1955 "movntdq\t{$src, $dst|$dst, $src}",
1956 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
1958 let AddedComplexity = 400 in { // Prefer non-temporal versions
1959 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1960 "movntps\t{$src, $dst|$dst, $src}",
1961 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1962 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1963 "movntpd\t{$src, $dst|$dst, $src}",
1964 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1966 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1967 "movntdq\t{$src, $dst|$dst, $src}",
1968 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1970 let ExeDomain = SSEPackedInt in
1971 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1972 "movntdq\t{$src, $dst|$dst, $src}",
1973 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1975 // There is no AVX form for instructions below this point
1976 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1977 "movnti\t{$src, $dst|$dst, $src}",
1978 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
1979 TB, Requires<[HasSSE2]>;
1981 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
1982 "movnti\t{$src, $dst|$dst, $src}",
1983 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
1984 TB, Requires<[HasSSE2]>;
1987 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1988 "movnti\t{$src, $dst|$dst, $src}",
1989 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
1990 TB, Requires<[HasSSE2]>;
1992 //===----------------------------------------------------------------------===//
1993 // SSE 1 & 2 - Misc Instructions (No AVX form)
1994 //===----------------------------------------------------------------------===//
1996 // Prefetch intrinsic.
1997 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
1998 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
1999 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2000 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2001 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2002 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2003 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2004 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2006 // Load, store, and memory fence
2007 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2008 TB, Requires<[HasSSE1]>;
2009 def : Pat<(X86SFence), (SFENCE)>;
2011 // Alias instructions that map zero vector to pxor / xorp* for sse.
2012 // We set canFoldAsLoad because this can be converted to a constant-pool
2013 // load of an all-zeros value if folding it would be beneficial.
2014 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2015 // JIT implementation, it does not expand the instructions below like
2016 // X86MCInstLower does.
2017 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2018 isCodeGenOnly = 1 in {
2019 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2020 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2021 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2022 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2023 let ExeDomain = SSEPackedInt in
2024 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2025 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2028 // The same as done above but for AVX. The 128-bit versions are the
2029 // same, but re-encoded. The 256-bit does not support PI version.
2030 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2031 // JIT implementatioan, it does not expand the instructions below like
2032 // X86MCInstLower does.
2033 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2034 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2035 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2036 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2037 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2038 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2039 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2040 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2041 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2042 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2043 let ExeDomain = SSEPackedInt in
2044 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2045 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2048 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2049 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2050 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2052 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2053 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2055 //===----------------------------------------------------------------------===//
2056 // SSE 1 & 2 - Load/Store XCSR register
2057 //===----------------------------------------------------------------------===//
2059 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2060 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2061 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2062 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2064 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2065 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2066 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2067 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2069 //===---------------------------------------------------------------------===//
2070 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2071 //===---------------------------------------------------------------------===//
2073 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2075 let neverHasSideEffects = 1 in {
2076 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2077 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2078 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2079 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2081 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2082 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2083 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2084 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2086 let canFoldAsLoad = 1, mayLoad = 1 in {
2087 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2088 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2089 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2090 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2091 let Predicates = [HasAVX] in {
2092 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2093 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2094 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2095 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2099 let mayStore = 1 in {
2100 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2101 (ins i128mem:$dst, VR128:$src),
2102 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2103 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2104 (ins i256mem:$dst, VR256:$src),
2105 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2106 let Predicates = [HasAVX] in {
2107 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2108 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2109 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2110 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2114 let neverHasSideEffects = 1 in
2115 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2116 "movdqa\t{$src, $dst|$dst, $src}", []>;
2118 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2119 "movdqu\t{$src, $dst|$dst, $src}",
2120 []>, XS, Requires<[HasSSE2]>;
2122 let canFoldAsLoad = 1, mayLoad = 1 in {
2123 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2124 "movdqa\t{$src, $dst|$dst, $src}",
2125 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2126 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2127 "movdqu\t{$src, $dst|$dst, $src}",
2128 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2129 XS, Requires<[HasSSE2]>;
2132 let mayStore = 1 in {
2133 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2134 "movdqa\t{$src, $dst|$dst, $src}",
2135 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2136 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2137 "movdqu\t{$src, $dst|$dst, $src}",
2138 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2139 XS, Requires<[HasSSE2]>;
2142 // Intrinsic forms of MOVDQU load and store
2143 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2144 "vmovdqu\t{$src, $dst|$dst, $src}",
2145 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2146 XS, VEX, Requires<[HasAVX]>;
2148 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2149 "movdqu\t{$src, $dst|$dst, $src}",
2150 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2151 XS, Requires<[HasSSE2]>;
2153 } // ExeDomain = SSEPackedInt
2155 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2156 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2157 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2159 //===---------------------------------------------------------------------===//
2160 // SSE2 - Packed Integer Arithmetic Instructions
2161 //===---------------------------------------------------------------------===//
2163 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2165 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2166 bit IsCommutable = 0, bit Is2Addr = 1> {
2167 let isCommutable = IsCommutable in
2168 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2169 (ins VR128:$src1, VR128:$src2),
2171 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2172 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2173 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2174 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2175 (ins VR128:$src1, i128mem:$src2),
2177 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2178 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2179 [(set VR128:$dst, (IntId VR128:$src1,
2180 (bitconvert (memopv2i64 addr:$src2))))]>;
2183 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2184 string OpcodeStr, Intrinsic IntId,
2185 Intrinsic IntId2, bit Is2Addr = 1> {
2186 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2187 (ins VR128:$src1, VR128:$src2),
2189 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2190 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2191 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2192 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2193 (ins VR128:$src1, i128mem:$src2),
2195 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2196 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2197 [(set VR128:$dst, (IntId VR128:$src1,
2198 (bitconvert (memopv2i64 addr:$src2))))]>;
2199 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2200 (ins VR128:$src1, i32i8imm:$src2),
2202 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2203 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2204 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2207 /// PDI_binop_rm - Simple SSE2 binary operator.
2208 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2209 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2210 let isCommutable = IsCommutable in
2211 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2212 (ins VR128:$src1, VR128:$src2),
2214 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2215 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2216 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2217 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2218 (ins VR128:$src1, i128mem:$src2),
2220 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2221 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2222 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2223 (bitconvert (memopv2i64 addr:$src2)))))]>;
2226 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2228 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2229 /// to collapse (bitconvert VT to VT) into its operand.
2231 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2232 bit IsCommutable = 0, bit Is2Addr = 1> {
2233 let isCommutable = IsCommutable in
2234 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2235 (ins VR128:$src1, VR128:$src2),
2237 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2239 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2240 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2241 (ins VR128:$src1, i128mem:$src2),
2243 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2245 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2248 } // ExeDomain = SSEPackedInt
2250 // 128-bit Integer Arithmetic
2252 let Predicates = [HasAVX] in {
2253 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2254 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2255 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2256 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2257 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2258 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2259 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2260 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2261 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2264 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2266 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2268 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2270 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2272 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2274 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2276 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2278 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2280 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2282 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2284 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2286 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2288 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2290 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2292 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2294 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2296 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2298 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2300 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2304 let Constraints = "$src1 = $dst" in {
2305 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2306 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2307 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2308 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2309 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2310 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2311 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2312 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2313 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2316 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2317 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2318 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2319 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2320 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2321 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2322 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2323 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2324 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2325 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2326 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2327 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2328 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2329 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2330 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2331 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2332 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2333 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2334 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2336 } // Constraints = "$src1 = $dst"
2338 //===---------------------------------------------------------------------===//
2339 // SSE2 - Packed Integer Logical Instructions
2340 //===---------------------------------------------------------------------===//
2342 let Predicates = [HasAVX] in {
2343 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2344 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2346 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2347 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2349 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2350 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2353 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2354 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2356 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2357 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2359 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2360 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2363 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2364 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2366 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2367 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2370 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2371 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2372 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2374 let ExeDomain = SSEPackedInt in {
2375 let neverHasSideEffects = 1 in {
2376 // 128-bit logical shifts.
2377 def VPSLLDQri : PDIi8<0x73, MRM7r,
2378 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2379 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2381 def VPSRLDQri : PDIi8<0x73, MRM3r,
2382 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2383 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2385 // PSRADQri doesn't exist in SSE[1-3].
2387 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2388 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2389 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2390 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2391 VR128:$src2)))]>, VEX_4V;
2393 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2394 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2395 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2396 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2397 (memopv2i64 addr:$src2))))]>,
2402 let Constraints = "$src1 = $dst" in {
2403 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2404 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2405 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2406 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2407 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2408 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2410 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2411 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2412 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2413 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2414 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2415 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2417 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2418 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2419 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2420 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2422 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2423 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2424 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2426 let ExeDomain = SSEPackedInt in {
2427 let neverHasSideEffects = 1 in {
2428 // 128-bit logical shifts.
2429 def PSLLDQri : PDIi8<0x73, MRM7r,
2430 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2431 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2432 def PSRLDQri : PDIi8<0x73, MRM3r,
2433 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2434 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2435 // PSRADQri doesn't exist in SSE[1-3].
2437 def PANDNrr : PDI<0xDF, MRMSrcReg,
2438 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2439 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2441 def PANDNrm : PDI<0xDF, MRMSrcMem,
2442 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2443 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2445 } // Constraints = "$src1 = $dst"
2447 let Predicates = [HasAVX] in {
2448 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2449 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2450 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2451 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2452 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2453 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2454 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2455 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2456 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2457 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2459 // Shift up / down and insert zero's.
2460 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2461 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2462 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2463 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2466 let Predicates = [HasSSE2] in {
2467 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2468 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2469 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2470 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2471 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2472 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2473 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2474 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2475 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2476 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2478 // Shift up / down and insert zero's.
2479 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2480 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2481 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2482 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2485 //===---------------------------------------------------------------------===//
2486 // SSE2 - Packed Integer Comparison Instructions
2487 //===---------------------------------------------------------------------===//
2489 let Predicates = [HasAVX] in {
2490 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2492 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2494 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2496 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2498 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2500 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2504 let Constraints = "$src1 = $dst" in {
2505 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2506 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2507 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2508 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2509 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2510 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2511 } // Constraints = "$src1 = $dst"
2513 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2514 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2515 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2516 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2517 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2518 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2519 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2520 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2521 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2522 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2523 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2524 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2526 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2527 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2528 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2529 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2530 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2531 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2532 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2533 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2534 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2535 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2536 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2537 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2539 //===---------------------------------------------------------------------===//
2540 // SSE2 - Packed Integer Pack Instructions
2541 //===---------------------------------------------------------------------===//
2543 let Predicates = [HasAVX] in {
2544 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2546 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2548 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2552 let Constraints = "$src1 = $dst" in {
2553 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2554 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2555 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2556 } // Constraints = "$src1 = $dst"
2558 //===---------------------------------------------------------------------===//
2559 // SSE2 - Packed Integer Shuffle Instructions
2560 //===---------------------------------------------------------------------===//
2562 let ExeDomain = SSEPackedInt in {
2563 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2565 def ri : Ii8<0x70, MRMSrcReg,
2566 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2567 !strconcat(OpcodeStr,
2568 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2569 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2571 def mi : Ii8<0x70, MRMSrcMem,
2572 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2573 !strconcat(OpcodeStr,
2574 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2575 [(set VR128:$dst, (vt (pshuf_frag:$src2
2576 (bc_frag (memopv2i64 addr:$src1)),
2579 } // ExeDomain = SSEPackedInt
2581 let Predicates = [HasAVX] in {
2582 let AddedComplexity = 5 in
2583 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2586 // SSE2 with ImmT == Imm8 and XS prefix.
2587 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2590 // SSE2 with ImmT == Imm8 and XD prefix.
2591 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2595 let Predicates = [HasSSE2] in {
2596 let AddedComplexity = 5 in
2597 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2599 // SSE2 with ImmT == Imm8 and XS prefix.
2600 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2602 // SSE2 with ImmT == Imm8 and XD prefix.
2603 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2606 //===---------------------------------------------------------------------===//
2607 // SSE2 - Packed Integer Unpack Instructions
2608 //===---------------------------------------------------------------------===//
2610 let ExeDomain = SSEPackedInt in {
2611 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2612 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2613 def rr : PDI<opc, MRMSrcReg,
2614 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2616 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2617 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2618 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2619 def rm : PDI<opc, MRMSrcMem,
2620 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2622 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2623 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2624 [(set VR128:$dst, (unp_frag VR128:$src1,
2625 (bc_frag (memopv2i64
2629 let Predicates = [HasAVX] in {
2630 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2632 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2634 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2637 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2638 /// knew to collapse (bitconvert VT to VT) into its operand.
2639 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2640 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2641 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2643 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2644 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2645 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2646 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2648 (v2i64 (unpckl VR128:$src1,
2649 (memopv2i64 addr:$src2))))]>, VEX_4V;
2651 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2653 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2655 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2658 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2659 /// knew to collapse (bitconvert VT to VT) into its operand.
2660 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2661 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2662 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2664 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2665 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2666 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2667 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2669 (v2i64 (unpckh VR128:$src1,
2670 (memopv2i64 addr:$src2))))]>, VEX_4V;
2673 let Constraints = "$src1 = $dst" in {
2674 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2675 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2676 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2678 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2679 /// knew to collapse (bitconvert VT to VT) into its operand.
2680 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2681 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2682 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2684 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2685 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2686 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2687 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2689 (v2i64 (unpckl VR128:$src1,
2690 (memopv2i64 addr:$src2))))]>;
2692 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2693 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2694 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2696 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2697 /// knew to collapse (bitconvert VT to VT) into its operand.
2698 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2699 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2700 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2702 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2703 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2704 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2705 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2707 (v2i64 (unpckh VR128:$src1,
2708 (memopv2i64 addr:$src2))))]>;
2711 } // ExeDomain = SSEPackedInt
2713 //===---------------------------------------------------------------------===//
2714 // SSE2 - Packed Integer Extract and Insert
2715 //===---------------------------------------------------------------------===//
2717 let ExeDomain = SSEPackedInt in {
2718 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2719 def rri : Ii8<0xC4, MRMSrcReg,
2720 (outs VR128:$dst), (ins VR128:$src1,
2721 GR32:$src2, i32i8imm:$src3),
2723 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2724 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2726 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2727 def rmi : Ii8<0xC4, MRMSrcMem,
2728 (outs VR128:$dst), (ins VR128:$src1,
2729 i16mem:$src2, i32i8imm:$src3),
2731 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2732 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2734 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2739 let Predicates = [HasAVX] in
2740 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2741 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2742 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2743 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2744 imm:$src2))]>, OpSize, VEX;
2745 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2746 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2747 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2748 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2752 let Predicates = [HasAVX] in {
2753 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2754 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2755 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2756 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2757 []>, OpSize, VEX_4V;
2760 let Constraints = "$src1 = $dst" in
2761 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2763 } // ExeDomain = SSEPackedInt
2765 //===---------------------------------------------------------------------===//
2766 // SSE2 - Packed Mask Creation
2767 //===---------------------------------------------------------------------===//
2769 let ExeDomain = SSEPackedInt in {
2771 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2772 "pmovmskb\t{$src, $dst|$dst, $src}",
2773 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2774 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2775 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2776 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2777 "pmovmskb\t{$src, $dst|$dst, $src}",
2778 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2780 } // ExeDomain = SSEPackedInt
2782 //===---------------------------------------------------------------------===//
2783 // SSE2 - Conditional Store
2784 //===---------------------------------------------------------------------===//
2786 let ExeDomain = SSEPackedInt in {
2789 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2790 (ins VR128:$src, VR128:$mask),
2791 "maskmovdqu\t{$mask, $src|$src, $mask}",
2792 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2794 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2795 (ins VR128:$src, VR128:$mask),
2796 "maskmovdqu\t{$mask, $src|$src, $mask}",
2797 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2800 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2801 "maskmovdqu\t{$mask, $src|$src, $mask}",
2802 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2804 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2805 "maskmovdqu\t{$mask, $src|$src, $mask}",
2806 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2808 } // ExeDomain = SSEPackedInt
2810 //===---------------------------------------------------------------------===//
2811 // SSE2 - Move Doubleword
2812 //===---------------------------------------------------------------------===//
2814 // Move Int Doubleword to Packed Double Int
2815 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2816 "movd\t{$src, $dst|$dst, $src}",
2818 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2819 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2820 "movd\t{$src, $dst|$dst, $src}",
2822 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2824 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2825 "movd\t{$src, $dst|$dst, $src}",
2827 (v4i32 (scalar_to_vector GR32:$src)))]>;
2828 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2829 "movd\t{$src, $dst|$dst, $src}",
2831 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2832 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2833 "mov{d|q}\t{$src, $dst|$dst, $src}",
2835 (v2i64 (scalar_to_vector GR64:$src)))]>;
2836 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2837 "mov{d|q}\t{$src, $dst|$dst, $src}",
2838 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2841 // Move Int Doubleword to Single Scalar
2842 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2843 "movd\t{$src, $dst|$dst, $src}",
2844 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2846 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2847 "movd\t{$src, $dst|$dst, $src}",
2848 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2850 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2851 "movd\t{$src, $dst|$dst, $src}",
2852 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2854 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2855 "movd\t{$src, $dst|$dst, $src}",
2856 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2858 // Move Packed Doubleword Int to Packed Double Int
2859 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2860 "movd\t{$src, $dst|$dst, $src}",
2861 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2863 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2864 (ins i32mem:$dst, VR128:$src),
2865 "movd\t{$src, $dst|$dst, $src}",
2866 [(store (i32 (vector_extract (v4i32 VR128:$src),
2867 (iPTR 0))), addr:$dst)]>, VEX;
2868 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2869 "movd\t{$src, $dst|$dst, $src}",
2870 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2872 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2873 "movd\t{$src, $dst|$dst, $src}",
2874 [(store (i32 (vector_extract (v4i32 VR128:$src),
2875 (iPTR 0))), addr:$dst)]>;
2877 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2878 "mov{d|q}\t{$src, $dst|$dst, $src}",
2879 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2881 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2882 "movq\t{$src, $dst|$dst, $src}",
2883 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2885 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2886 "mov{d|q}\t{$src, $dst|$dst, $src}",
2887 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2888 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2889 "movq\t{$src, $dst|$dst, $src}",
2890 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2892 // Move Scalar Single to Double Int
2893 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2894 "movd\t{$src, $dst|$dst, $src}",
2895 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2896 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2897 "movd\t{$src, $dst|$dst, $src}",
2898 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2899 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2900 "movd\t{$src, $dst|$dst, $src}",
2901 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2902 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2903 "movd\t{$src, $dst|$dst, $src}",
2904 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2906 // movd / movq to XMM register zero-extends
2907 let AddedComplexity = 15 in {
2908 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2909 "movd\t{$src, $dst|$dst, $src}",
2910 [(set VR128:$dst, (v4i32 (X86vzmovl
2911 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2913 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2914 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2915 [(set VR128:$dst, (v2i64 (X86vzmovl
2916 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2919 let AddedComplexity = 15 in {
2920 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2921 "movd\t{$src, $dst|$dst, $src}",
2922 [(set VR128:$dst, (v4i32 (X86vzmovl
2923 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2924 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2925 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2926 [(set VR128:$dst, (v2i64 (X86vzmovl
2927 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2930 let AddedComplexity = 20 in {
2931 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2932 "movd\t{$src, $dst|$dst, $src}",
2934 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2935 (loadi32 addr:$src))))))]>,
2937 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2938 "movd\t{$src, $dst|$dst, $src}",
2940 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2941 (loadi32 addr:$src))))))]>;
2943 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2944 (MOVZDI2PDIrm addr:$src)>;
2945 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2946 (MOVZDI2PDIrm addr:$src)>;
2947 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2948 (MOVZDI2PDIrm addr:$src)>;
2951 //===---------------------------------------------------------------------===//
2952 // SSE2 - Move Quadword
2953 //===---------------------------------------------------------------------===//
2955 // Move Quadword Int to Packed Quadword Int
2956 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2957 "vmovq\t{$src, $dst|$dst, $src}",
2959 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2960 VEX, Requires<[HasAVX]>;
2961 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2962 "movq\t{$src, $dst|$dst, $src}",
2964 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2965 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
2967 // Move Packed Quadword Int to Quadword Int
2968 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2969 "movq\t{$src, $dst|$dst, $src}",
2970 [(store (i64 (vector_extract (v2i64 VR128:$src),
2971 (iPTR 0))), addr:$dst)]>, VEX;
2972 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2973 "movq\t{$src, $dst|$dst, $src}",
2974 [(store (i64 (vector_extract (v2i64 VR128:$src),
2975 (iPTR 0))), addr:$dst)]>;
2977 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
2978 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
2980 // Store / copy lower 64-bits of a XMM register.
2981 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2982 "movq\t{$src, $dst|$dst, $src}",
2983 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
2984 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2985 "movq\t{$src, $dst|$dst, $src}",
2986 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
2988 let AddedComplexity = 20 in
2989 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2990 "vmovq\t{$src, $dst|$dst, $src}",
2992 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
2993 (loadi64 addr:$src))))))]>,
2994 XS, VEX, Requires<[HasAVX]>;
2996 let AddedComplexity = 20 in {
2997 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2998 "movq\t{$src, $dst|$dst, $src}",
3000 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3001 (loadi64 addr:$src))))))]>,
3002 XS, Requires<[HasSSE2]>;
3004 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3005 (MOVZQI2PQIrm addr:$src)>;
3006 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3007 (MOVZQI2PQIrm addr:$src)>;
3008 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3011 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3012 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3013 let AddedComplexity = 15 in
3014 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3015 "vmovq\t{$src, $dst|$dst, $src}",
3016 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3017 XS, VEX, Requires<[HasAVX]>;
3018 let AddedComplexity = 15 in
3019 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3020 "movq\t{$src, $dst|$dst, $src}",
3021 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3022 XS, Requires<[HasSSE2]>;
3024 let AddedComplexity = 20 in
3025 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3026 "vmovq\t{$src, $dst|$dst, $src}",
3027 [(set VR128:$dst, (v2i64 (X86vzmovl
3028 (loadv2i64 addr:$src))))]>,
3029 XS, VEX, Requires<[HasAVX]>;
3030 let AddedComplexity = 20 in {
3031 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3032 "movq\t{$src, $dst|$dst, $src}",
3033 [(set VR128:$dst, (v2i64 (X86vzmovl
3034 (loadv2i64 addr:$src))))]>,
3035 XS, Requires<[HasSSE2]>;
3037 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3038 (MOVZPQILo2PQIrm addr:$src)>;
3041 // Instructions to match in the assembler
3042 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3043 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3044 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3045 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3046 // Recognize "movd" with GR64 destination, but encode as a "movq"
3047 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3048 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3050 // Instructions for the disassembler
3051 // xr = XMM register
3054 let Predicates = [HasAVX] in
3055 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3056 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3057 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3058 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3060 //===---------------------------------------------------------------------===//
3061 // SSE2 - Misc Instructions
3062 //===---------------------------------------------------------------------===//
3065 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3066 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3067 TB, Requires<[HasSSE2]>;
3069 // Load, store, and memory fence
3070 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3071 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3072 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3073 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3074 def : Pat<(X86LFence), (LFENCE)>;
3075 def : Pat<(X86MFence), (MFENCE)>;
3078 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3079 // was introduced with SSE2, it's backward compatible.
3080 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3082 // Alias instructions that map zero vector to pxor / xorp* for sse.
3083 // We set canFoldAsLoad because this can be converted to a constant-pool
3084 // load of an all-ones value if folding it would be beneficial.
3085 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3086 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3087 // FIXME: Change encoding to pseudo.
3088 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3089 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3091 //===---------------------------------------------------------------------===//
3092 // SSE3 - Conversion Instructions
3093 //===---------------------------------------------------------------------===//
3095 // Convert Packed Double FP to Packed DW Integers
3096 let Predicates = [HasAVX] in {
3097 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3098 // register, but the same isn't true when using memory operands instead.
3099 // Provide other assembly rr and rm forms to address this explicitly.
3100 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3101 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3102 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3103 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3106 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3107 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3108 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3109 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3112 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3113 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3114 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3115 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3118 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3119 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3120 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3121 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3123 // Convert Packed DW Integers to Packed Double FP
3124 let Predicates = [HasAVX] in {
3125 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3126 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3127 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3128 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3129 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3130 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3131 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3132 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3135 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3136 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3137 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3138 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3140 // AVX 256-bit register conversion intrinsics
3141 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3142 (VCVTDQ2PDYrr VR128:$src)>;
3143 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3144 (VCVTDQ2PDYrm addr:$src)>;
3146 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3147 (VCVTPD2DQYrr VR256:$src)>;
3148 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3149 (VCVTPD2DQYrm addr:$src)>;
3151 //===---------------------------------------------------------------------===//
3152 // SSE3 - Move Instructions
3153 //===---------------------------------------------------------------------===//
3155 // Replicate Single FP
3156 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3157 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3158 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3159 [(set VR128:$dst, (v4f32 (rep_frag
3160 VR128:$src, (undef))))]>;
3161 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3162 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3163 [(set VR128:$dst, (rep_frag
3164 (memopv4f32 addr:$src), (undef)))]>;
3167 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3169 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3170 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3171 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3172 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3175 let Predicates = [HasAVX] in {
3176 // FIXME: Merge above classes when we have patterns for the ymm version
3177 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3178 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3179 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3180 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3182 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3183 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3185 // Replicate Double FP
3186 multiclass sse3_replicate_dfp<string OpcodeStr> {
3187 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3188 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3189 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3190 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3191 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3193 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3197 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3198 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3199 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3201 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3202 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3206 let Predicates = [HasAVX] in {
3207 // FIXME: Merge above classes when we have patterns for the ymm version
3208 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3209 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3211 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3213 // Move Unaligned Integer
3214 let Predicates = [HasAVX] in {
3215 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3216 "vlddqu\t{$src, $dst|$dst, $src}",
3217 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3218 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3219 "vlddqu\t{$src, $dst|$dst, $src}",
3220 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3222 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3223 "lddqu\t{$src, $dst|$dst, $src}",
3224 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3226 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3228 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3230 // Several Move patterns
3231 let AddedComplexity = 5 in {
3232 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3233 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3234 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3235 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3236 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3237 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3238 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3239 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3242 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3243 let AddedComplexity = 15 in
3244 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3245 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3246 let AddedComplexity = 20 in
3247 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3248 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3250 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3251 let AddedComplexity = 15 in
3252 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3253 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3254 let AddedComplexity = 20 in
3255 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3256 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3258 //===---------------------------------------------------------------------===//
3259 // SSE3 - Arithmetic
3260 //===---------------------------------------------------------------------===//
3262 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3263 X86MemOperand x86memop, bit Is2Addr = 1> {
3264 def rr : I<0xD0, MRMSrcReg,
3265 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3267 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3268 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3269 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3270 def rm : I<0xD0, MRMSrcMem,
3271 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3273 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3274 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3275 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3278 let Predicates = [HasAVX],
3279 ExeDomain = SSEPackedDouble in {
3280 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3281 f128mem, 0>, TB, XD, VEX_4V;
3282 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3283 f128mem, 0>, TB, OpSize, VEX_4V;
3284 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3285 f256mem, 0>, TB, XD, VEX_4V;
3286 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3287 f256mem, 0>, TB, OpSize, VEX_4V;
3289 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3290 ExeDomain = SSEPackedDouble in {
3291 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3293 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3294 f128mem>, TB, OpSize;
3297 //===---------------------------------------------------------------------===//
3298 // SSE3 Instructions
3299 //===---------------------------------------------------------------------===//
3302 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3303 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3304 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3306 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3307 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3308 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3310 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3312 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3314 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3316 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3317 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3318 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3320 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3322 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3324 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3326 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3327 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3328 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3331 let Predicates = [HasAVX] in {
3332 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3333 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3334 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3335 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3336 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3337 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3338 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3339 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3340 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3341 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3342 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3343 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3344 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3345 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3346 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3347 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3350 let Constraints = "$src1 = $dst" in {
3351 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3352 int_x86_sse3_hadd_ps>;
3353 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3354 int_x86_sse3_hadd_pd>;
3355 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3356 int_x86_sse3_hsub_ps>;
3357 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3358 int_x86_sse3_hsub_pd>;
3361 //===---------------------------------------------------------------------===//
3362 // SSSE3 - Packed Absolute Instructions
3363 //===---------------------------------------------------------------------===//
3366 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3367 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3368 PatFrag mem_frag128, Intrinsic IntId128> {
3369 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3371 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3372 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3375 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3377 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3380 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3383 let Predicates = [HasAVX] in {
3384 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3385 int_x86_ssse3_pabs_b_128>, VEX;
3386 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3387 int_x86_ssse3_pabs_w_128>, VEX;
3388 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3389 int_x86_ssse3_pabs_d_128>, VEX;
3392 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3393 int_x86_ssse3_pabs_b_128>;
3394 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3395 int_x86_ssse3_pabs_w_128>;
3396 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3397 int_x86_ssse3_pabs_d_128>;
3399 //===---------------------------------------------------------------------===//
3400 // SSSE3 - Packed Binary Operator Instructions
3401 //===---------------------------------------------------------------------===//
3403 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3404 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3405 PatFrag mem_frag128, Intrinsic IntId128,
3407 let isCommutable = 1 in
3408 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3409 (ins VR128:$src1, VR128:$src2),
3411 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3412 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3413 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3415 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3416 (ins VR128:$src1, i128mem:$src2),
3418 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3419 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3421 (IntId128 VR128:$src1,
3422 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3425 let Predicates = [HasAVX] in {
3426 let isCommutable = 0 in {
3427 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3428 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3429 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3430 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3431 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3432 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3433 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3434 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3435 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3436 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3437 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3438 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3439 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3440 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3441 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3442 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3443 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3444 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3445 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3446 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3447 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3448 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3450 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3451 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3454 // None of these have i8 immediate fields.
3455 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3456 let isCommutable = 0 in {
3457 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3458 int_x86_ssse3_phadd_w_128>;
3459 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3460 int_x86_ssse3_phadd_d_128>;
3461 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3462 int_x86_ssse3_phadd_sw_128>;
3463 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3464 int_x86_ssse3_phsub_w_128>;
3465 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3466 int_x86_ssse3_phsub_d_128>;
3467 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3468 int_x86_ssse3_phsub_sw_128>;
3469 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3470 int_x86_ssse3_pmadd_ub_sw_128>;
3471 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3472 int_x86_ssse3_pshuf_b_128>;
3473 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3474 int_x86_ssse3_psign_b_128>;
3475 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3476 int_x86_ssse3_psign_w_128>;
3477 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3478 int_x86_ssse3_psign_d_128>;
3480 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3481 int_x86_ssse3_pmul_hr_sw_128>;
3484 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3485 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3486 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3487 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3489 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3490 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3491 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3492 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3493 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3494 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3496 //===---------------------------------------------------------------------===//
3497 // SSSE3 - Packed Align Instruction Patterns
3498 //===---------------------------------------------------------------------===//
3500 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3501 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3502 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3504 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3506 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3508 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3509 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3511 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3513 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3517 let Predicates = [HasAVX] in
3518 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3519 let Constraints = "$src1 = $dst" in
3520 defm PALIGN : ssse3_palign<"palignr">;
3522 let AddedComplexity = 5 in {
3523 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3524 (PALIGNR128rr VR128:$src2, VR128:$src1,
3525 (SHUFFLE_get_palign_imm VR128:$src3))>,
3526 Requires<[HasSSSE3]>;
3527 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3528 (PALIGNR128rr VR128:$src2, VR128:$src1,
3529 (SHUFFLE_get_palign_imm VR128:$src3))>,
3530 Requires<[HasSSSE3]>;
3531 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3532 (PALIGNR128rr VR128:$src2, VR128:$src1,
3533 (SHUFFLE_get_palign_imm VR128:$src3))>,
3534 Requires<[HasSSSE3]>;
3535 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3536 (PALIGNR128rr VR128:$src2, VR128:$src1,
3537 (SHUFFLE_get_palign_imm VR128:$src3))>,
3538 Requires<[HasSSSE3]>;
3541 //===---------------------------------------------------------------------===//
3542 // SSSE3 Misc Instructions
3543 //===---------------------------------------------------------------------===//
3545 // Thread synchronization
3546 let usesCustomInserter = 1 in {
3547 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3548 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3549 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3550 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3553 let Uses = [EAX, ECX, EDX] in
3554 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3555 Requires<[HasSSE3]>;
3556 let Uses = [ECX, EAX] in
3557 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3558 Requires<[HasSSE3]>;
3560 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3561 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3563 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3564 Requires<[In32BitMode]>;
3565 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3566 Requires<[In64BitMode]>;
3568 //===---------------------------------------------------------------------===//
3569 // Non-Instruction Patterns
3570 //===---------------------------------------------------------------------===//
3572 // extload f32 -> f64. This matches load+fextend because we have a hack in
3573 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3575 // Since these loads aren't folded into the fextend, we have to match it
3577 let Predicates = [HasSSE2] in
3578 def : Pat<(fextend (loadf32 addr:$src)),
3579 (CVTSS2SDrm addr:$src)>;
3582 let Predicates = [HasXMMInt] in {
3583 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3584 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3585 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3586 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3587 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3588 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3589 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3590 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3591 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3592 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3593 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3594 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3595 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3596 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3597 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3598 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3599 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3600 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3601 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3602 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3603 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3604 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3605 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3606 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3607 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3608 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3609 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3610 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3611 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3612 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3615 let Predicates = [HasAVX] in {
3616 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3619 // Move scalar to XMM zero-extended
3620 // movd to XMM register zero-extends
3621 let AddedComplexity = 15 in {
3622 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3623 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3624 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3625 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3626 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3627 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3628 (MOVSSrr (v4f32 (V_SET0PS)),
3629 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3630 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3631 (MOVSSrr (v4i32 (V_SET0PI)),
3632 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3635 // Splat v2f64 / v2i64
3636 let AddedComplexity = 10 in {
3637 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3638 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3639 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3640 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3641 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3642 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3643 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3644 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3647 // Special unary SHUFPSrri case.
3648 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3649 (SHUFPSrri VR128:$src1, VR128:$src1,
3650 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3651 let AddedComplexity = 5 in
3652 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3653 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3654 Requires<[HasSSE2]>;
3655 // Special unary SHUFPDrri case.
3656 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3657 (SHUFPDrri VR128:$src1, VR128:$src1,
3658 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3659 Requires<[HasSSE2]>;
3660 // Special unary SHUFPDrri case.
3661 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3662 (SHUFPDrri VR128:$src1, VR128:$src1,
3663 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3664 Requires<[HasSSE2]>;
3665 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3666 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3667 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3668 Requires<[HasSSE2]>;
3670 // Special binary v4i32 shuffle cases with SHUFPS.
3671 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3672 (SHUFPSrri VR128:$src1, VR128:$src2,
3673 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3674 Requires<[HasSSE2]>;
3675 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3676 (SHUFPSrmi VR128:$src1, addr:$src2,
3677 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3678 Requires<[HasSSE2]>;
3679 // Special binary v2i64 shuffle cases using SHUFPDrri.
3680 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3681 (SHUFPDrri VR128:$src1, VR128:$src2,
3682 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3683 Requires<[HasSSE2]>;
3685 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3686 let AddedComplexity = 15 in {
3687 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3688 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3689 Requires<[OptForSpeed, HasSSE2]>;
3690 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3691 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3692 Requires<[OptForSpeed, HasSSE2]>;
3694 let AddedComplexity = 10 in {
3695 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3696 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3697 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3698 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3699 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3700 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3701 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3702 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3705 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3706 let AddedComplexity = 15 in {
3707 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3708 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3709 Requires<[OptForSpeed, HasSSE2]>;
3710 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3711 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3712 Requires<[OptForSpeed, HasSSE2]>;
3714 let AddedComplexity = 10 in {
3715 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3716 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3717 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3718 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3719 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3720 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3721 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3722 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3725 let AddedComplexity = 20 in {
3726 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3727 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3728 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3730 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3731 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3732 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3734 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3735 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3736 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3737 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3738 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3741 let AddedComplexity = 20 in {
3742 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3743 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3744 (MOVLPSrm VR128:$src1, addr:$src2)>;
3745 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3746 (MOVLPDrm VR128:$src1, addr:$src2)>;
3747 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3748 (MOVLPSrm VR128:$src1, addr:$src2)>;
3749 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3750 (MOVLPDrm VR128:$src1, addr:$src2)>;
3753 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3754 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3755 (MOVLPSmr addr:$src1, VR128:$src2)>;
3756 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3757 (MOVLPDmr addr:$src1, VR128:$src2)>;
3758 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3760 (MOVLPSmr addr:$src1, VR128:$src2)>;
3761 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3762 (MOVLPDmr addr:$src1, VR128:$src2)>;
3764 let AddedComplexity = 15 in {
3765 // Setting the lowest element in the vector.
3766 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3767 (MOVSSrr (v4i32 VR128:$src1),
3768 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3769 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3770 (MOVSDrr (v2i64 VR128:$src1),
3771 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3773 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3774 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3775 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3776 Requires<[HasSSE2]>;
3777 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3778 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3779 Requires<[HasSSE2]>;
3782 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3783 // fall back to this for SSE1)
3784 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3785 (SHUFPSrri VR128:$src2, VR128:$src1,
3786 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3788 // Set lowest element and zero upper elements.
3789 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3790 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3792 // vector -> vector casts
3793 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3794 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3795 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3796 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3798 // Use movaps / movups for SSE integer load / store (one byte shorter).
3799 let Predicates = [HasSSE1] in {
3800 def : Pat<(alignedloadv4i32 addr:$src),
3801 (MOVAPSrm addr:$src)>;
3802 def : Pat<(loadv4i32 addr:$src),
3803 (MOVUPSrm addr:$src)>;
3804 def : Pat<(alignedloadv2i64 addr:$src),
3805 (MOVAPSrm addr:$src)>;
3806 def : Pat<(loadv2i64 addr:$src),
3807 (MOVUPSrm addr:$src)>;
3809 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3810 (MOVAPSmr addr:$dst, VR128:$src)>;
3811 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3812 (MOVAPSmr addr:$dst, VR128:$src)>;
3813 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3814 (MOVAPSmr addr:$dst, VR128:$src)>;
3815 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3816 (MOVAPSmr addr:$dst, VR128:$src)>;
3817 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3818 (MOVUPSmr addr:$dst, VR128:$src)>;
3819 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3820 (MOVUPSmr addr:$dst, VR128:$src)>;
3821 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3822 (MOVUPSmr addr:$dst, VR128:$src)>;
3823 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3824 (MOVUPSmr addr:$dst, VR128:$src)>;
3827 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3828 let Predicates = [HasAVX] in {
3829 def : Pat<(alignedloadv4i32 addr:$src),
3830 (VMOVAPSrm addr:$src)>;
3831 def : Pat<(loadv4i32 addr:$src),
3832 (VMOVUPSrm addr:$src)>;
3833 def : Pat<(alignedloadv2i64 addr:$src),
3834 (VMOVAPSrm addr:$src)>;
3835 def : Pat<(loadv2i64 addr:$src),
3836 (VMOVUPSrm addr:$src)>;
3838 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3839 (VMOVAPSmr addr:$dst, VR128:$src)>;
3840 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3841 (VMOVAPSmr addr:$dst, VR128:$src)>;
3842 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3843 (VMOVAPSmr addr:$dst, VR128:$src)>;
3844 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3845 (VMOVAPSmr addr:$dst, VR128:$src)>;
3846 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3847 (VMOVUPSmr addr:$dst, VR128:$src)>;
3848 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3849 (VMOVUPSmr addr:$dst, VR128:$src)>;
3850 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3851 (VMOVUPSmr addr:$dst, VR128:$src)>;
3852 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3853 (VMOVUPSmr addr:$dst, VR128:$src)>;
3856 //===----------------------------------------------------------------------===//
3857 // SSE4.1 - Packed Move with Sign/Zero Extend
3858 //===----------------------------------------------------------------------===//
3860 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3861 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3862 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3863 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3865 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3866 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3868 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3872 let Predicates = [HasAVX] in {
3873 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3875 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3877 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3879 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3881 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3883 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3887 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3888 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3889 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3890 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3891 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3892 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3894 // Common patterns involving scalar load.
3895 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3896 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3897 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3898 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3900 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3901 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3902 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3903 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3905 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3906 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3907 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3908 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3910 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3911 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3912 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3913 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3915 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3916 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3917 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3918 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3920 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3921 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3922 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3923 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3926 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3927 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3928 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3929 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3931 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3932 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3934 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
3938 let Predicates = [HasAVX] in {
3939 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
3941 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
3943 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
3945 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
3949 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
3950 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
3951 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
3952 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
3954 // Common patterns involving scalar load
3955 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
3956 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
3957 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
3958 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
3960 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
3961 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
3962 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
3963 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
3966 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3967 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3968 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3969 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3971 // Expecting a i16 load any extended to i32 value.
3972 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
3973 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3974 [(set VR128:$dst, (IntId (bitconvert
3975 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
3979 let Predicates = [HasAVX] in {
3980 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
3982 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
3985 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
3986 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
3988 // Common patterns involving scalar load
3989 def : Pat<(int_x86_sse41_pmovsxbq
3990 (bitconvert (v4i32 (X86vzmovl
3991 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
3992 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
3994 def : Pat<(int_x86_sse41_pmovzxbq
3995 (bitconvert (v4i32 (X86vzmovl
3996 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
3997 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
3999 //===----------------------------------------------------------------------===//
4000 // SSE4.1 - Extract Instructions
4001 //===----------------------------------------------------------------------===//
4003 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4004 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4005 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4006 (ins VR128:$src1, i32i8imm:$src2),
4007 !strconcat(OpcodeStr,
4008 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4009 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4011 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4012 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4013 !strconcat(OpcodeStr,
4014 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4017 // There's an AssertZext in the way of writing the store pattern
4018 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4021 let Predicates = [HasAVX] in {
4022 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4023 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4024 (ins VR128:$src1, i32i8imm:$src2),
4025 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4028 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4031 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4032 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4033 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4034 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4035 !strconcat(OpcodeStr,
4036 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4039 // There's an AssertZext in the way of writing the store pattern
4040 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4043 let Predicates = [HasAVX] in
4044 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4046 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4049 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4050 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4051 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4052 (ins VR128:$src1, i32i8imm:$src2),
4053 !strconcat(OpcodeStr,
4054 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4056 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4057 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4058 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4059 !strconcat(OpcodeStr,
4060 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4061 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4062 addr:$dst)]>, OpSize;
4065 let Predicates = [HasAVX] in
4066 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4068 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4070 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4071 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4072 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4073 (ins VR128:$src1, i32i8imm:$src2),
4074 !strconcat(OpcodeStr,
4075 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4077 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4078 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4079 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4080 !strconcat(OpcodeStr,
4081 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4082 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4083 addr:$dst)]>, OpSize, REX_W;
4086 let Predicates = [HasAVX] in
4087 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4089 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4091 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4093 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4094 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4095 (ins VR128:$src1, i32i8imm:$src2),
4096 !strconcat(OpcodeStr,
4097 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4099 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4101 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4102 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4103 !strconcat(OpcodeStr,
4104 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4105 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4106 addr:$dst)]>, OpSize;
4109 let Predicates = [HasAVX] in {
4110 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4111 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4112 (ins VR128:$src1, i32i8imm:$src2),
4113 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4116 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4118 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4119 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4122 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4123 Requires<[HasSSE41]>;
4125 //===----------------------------------------------------------------------===//
4126 // SSE4.1 - Insert Instructions
4127 //===----------------------------------------------------------------------===//
4129 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4130 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4131 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4133 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4135 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4137 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4138 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4139 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4141 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4143 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4145 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4146 imm:$src3))]>, OpSize;
4149 let Predicates = [HasAVX] in
4150 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4151 let Constraints = "$src1 = $dst" in
4152 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4154 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4155 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4156 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4158 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4160 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4162 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4164 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4165 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4167 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4169 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4171 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4172 imm:$src3)))]>, OpSize;
4175 let Predicates = [HasAVX] in
4176 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4177 let Constraints = "$src1 = $dst" in
4178 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4180 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4181 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4182 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4184 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4186 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4188 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4190 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4191 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4193 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4195 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4197 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4198 imm:$src3)))]>, OpSize;
4201 let Predicates = [HasAVX] in
4202 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4203 let Constraints = "$src1 = $dst" in
4204 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4206 // insertps has a few different modes, there's the first two here below which
4207 // are optimized inserts that won't zero arbitrary elements in the destination
4208 // vector. The next one matches the intrinsic and could zero arbitrary elements
4209 // in the target vector.
4210 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4211 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4212 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4214 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4216 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4218 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4220 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4221 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4223 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4225 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4227 (X86insrtps VR128:$src1,
4228 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4229 imm:$src3))]>, OpSize;
4232 let Constraints = "$src1 = $dst" in
4233 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4234 let Predicates = [HasAVX] in
4235 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4237 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4238 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4240 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4241 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4242 Requires<[HasSSE41]>;
4244 //===----------------------------------------------------------------------===//
4245 // SSE4.1 - Round Instructions
4246 //===----------------------------------------------------------------------===//
4248 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4249 X86MemOperand x86memop, RegisterClass RC,
4250 PatFrag mem_frag32, PatFrag mem_frag64,
4251 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4252 // Intrinsic operation, reg.
4253 // Vector intrinsic operation, reg
4254 def PSr : SS4AIi8<opcps, MRMSrcReg,
4255 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4256 !strconcat(OpcodeStr,
4257 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4258 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4261 // Vector intrinsic operation, mem
4262 def PSm : Ii8<opcps, MRMSrcMem,
4263 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4264 !strconcat(OpcodeStr,
4265 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4267 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4269 Requires<[HasSSE41]>;
4271 // Vector intrinsic operation, reg
4272 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4273 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4274 !strconcat(OpcodeStr,
4275 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4276 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4279 // Vector intrinsic operation, mem
4280 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4281 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4282 !strconcat(OpcodeStr,
4283 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4285 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4289 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4290 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4291 // Intrinsic operation, reg.
4292 // Vector intrinsic operation, reg
4293 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4294 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4295 !strconcat(OpcodeStr,
4296 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4299 // Vector intrinsic operation, mem
4300 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4301 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4302 !strconcat(OpcodeStr,
4303 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4304 []>, TA, OpSize, Requires<[HasSSE41]>;
4306 // Vector intrinsic operation, reg
4307 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4308 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4309 !strconcat(OpcodeStr,
4310 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4313 // Vector intrinsic operation, mem
4314 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4315 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4316 !strconcat(OpcodeStr,
4317 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4321 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4324 Intrinsic F64Int, bit Is2Addr = 1> {
4325 // Intrinsic operation, reg.
4326 def SSr : SS4AIi8<opcss, MRMSrcReg,
4327 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4329 !strconcat(OpcodeStr,
4330 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4331 !strconcat(OpcodeStr,
4332 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4333 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4336 // Intrinsic operation, mem.
4337 def SSm : SS4AIi8<opcss, MRMSrcMem,
4338 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4340 !strconcat(OpcodeStr,
4341 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4342 !strconcat(OpcodeStr,
4343 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4345 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4348 // Intrinsic operation, reg.
4349 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4350 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4352 !strconcat(OpcodeStr,
4353 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4354 !strconcat(OpcodeStr,
4355 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4356 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4359 // Intrinsic operation, mem.
4360 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4361 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4363 !strconcat(OpcodeStr,
4364 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4365 !strconcat(OpcodeStr,
4366 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4368 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4372 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4374 // Intrinsic operation, reg.
4375 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4376 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4377 !strconcat(OpcodeStr,
4378 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4381 // Intrinsic operation, mem.
4382 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4383 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4384 !strconcat(OpcodeStr,
4385 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4388 // Intrinsic operation, reg.
4389 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4390 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4391 !strconcat(OpcodeStr,
4392 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4395 // Intrinsic operation, mem.
4396 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4397 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4398 !strconcat(OpcodeStr,
4399 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4403 // FP round - roundss, roundps, roundsd, roundpd
4404 let Predicates = [HasAVX] in {
4406 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4407 memopv4f32, memopv2f64,
4408 int_x86_sse41_round_ps,
4409 int_x86_sse41_round_pd>, VEX;
4410 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4411 memopv8f32, memopv4f64,
4412 int_x86_avx_round_ps_256,
4413 int_x86_avx_round_pd_256>, VEX;
4414 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4415 int_x86_sse41_round_ss,
4416 int_x86_sse41_round_sd, 0>, VEX_4V;
4418 // Instructions for the assembler
4419 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4421 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4423 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4426 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4427 memopv4f32, memopv2f64,
4428 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4429 let Constraints = "$src1 = $dst" in
4430 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4431 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4433 //===----------------------------------------------------------------------===//
4434 // SSE4.1 - Packed Bit Test
4435 //===----------------------------------------------------------------------===//
4437 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4438 // the intel intrinsic that corresponds to this.
4439 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4440 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4441 "vptest\t{$src2, $src1|$src1, $src2}",
4442 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4444 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4445 "vptest\t{$src2, $src1|$src1, $src2}",
4446 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4449 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4450 "vptest\t{$src2, $src1|$src1, $src2}",
4451 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4453 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4454 "vptest\t{$src2, $src1|$src1, $src2}",
4455 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4459 let Defs = [EFLAGS] in {
4460 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4461 "ptest \t{$src2, $src1|$src1, $src2}",
4462 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4464 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4465 "ptest \t{$src2, $src1|$src1, $src2}",
4466 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4470 // The bit test instructions below are AVX only
4471 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4472 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4473 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4474 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4475 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4476 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4477 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4478 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4482 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4483 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4484 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4485 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4486 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4489 //===----------------------------------------------------------------------===//
4490 // SSE4.1 - Misc Instructions
4491 //===----------------------------------------------------------------------===//
4493 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4494 "popcnt{w}\t{$src, $dst|$dst, $src}",
4495 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4496 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4497 "popcnt{w}\t{$src, $dst|$dst, $src}",
4498 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4500 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4501 "popcnt{l}\t{$src, $dst|$dst, $src}",
4502 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4503 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4504 "popcnt{l}\t{$src, $dst|$dst, $src}",
4505 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4507 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4508 "popcnt{q}\t{$src, $dst|$dst, $src}",
4509 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4510 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4511 "popcnt{q}\t{$src, $dst|$dst, $src}",
4512 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4516 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4517 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4518 Intrinsic IntId128> {
4519 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4521 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4522 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4523 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4525 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4528 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4531 let Predicates = [HasAVX] in
4532 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4533 int_x86_sse41_phminposuw>, VEX;
4534 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4535 int_x86_sse41_phminposuw>;
4537 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4538 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4539 Intrinsic IntId128, bit Is2Addr = 1> {
4540 let isCommutable = 1 in
4541 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4542 (ins VR128:$src1, VR128:$src2),
4544 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4545 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4546 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4547 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4548 (ins VR128:$src1, i128mem:$src2),
4550 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4551 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4553 (IntId128 VR128:$src1,
4554 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4557 let Predicates = [HasAVX] in {
4558 let isCommutable = 0 in
4559 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4561 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4563 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4565 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4567 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4569 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4571 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4573 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4575 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4577 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4579 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4583 let Constraints = "$src1 = $dst" in {
4584 let isCommutable = 0 in
4585 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4586 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4587 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4588 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4589 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4590 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4591 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4592 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4593 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4594 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4595 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4598 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4599 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4600 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4601 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4603 /// SS48I_binop_rm - Simple SSE41 binary operator.
4604 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4605 ValueType OpVT, bit Is2Addr = 1> {
4606 let isCommutable = 1 in
4607 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4608 (ins VR128:$src1, VR128:$src2),
4610 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4611 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4612 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4614 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4615 (ins VR128:$src1, i128mem:$src2),
4617 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4618 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4619 [(set VR128:$dst, (OpNode VR128:$src1,
4620 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4624 let Predicates = [HasAVX] in
4625 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4626 let Constraints = "$src1 = $dst" in
4627 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4629 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4630 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4631 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4632 X86MemOperand x86memop, bit Is2Addr = 1> {
4633 let isCommutable = 1 in
4634 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4635 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4637 !strconcat(OpcodeStr,
4638 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4639 !strconcat(OpcodeStr,
4640 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4641 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4643 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4644 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4646 !strconcat(OpcodeStr,
4647 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4648 !strconcat(OpcodeStr,
4649 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4652 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4656 let Predicates = [HasAVX] in {
4657 let isCommutable = 0 in {
4658 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4659 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4660 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4661 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4662 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4663 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4664 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4665 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4666 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4667 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4668 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4669 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4671 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4672 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4673 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4674 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4675 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4676 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4679 let Constraints = "$src1 = $dst" in {
4680 let isCommutable = 0 in {
4681 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4682 VR128, memopv16i8, i128mem>;
4683 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4684 VR128, memopv16i8, i128mem>;
4685 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4686 VR128, memopv16i8, i128mem>;
4687 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4688 VR128, memopv16i8, i128mem>;
4690 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4691 VR128, memopv16i8, i128mem>;
4692 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4693 VR128, memopv16i8, i128mem>;
4696 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4697 let Predicates = [HasAVX] in {
4698 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4699 RegisterClass RC, X86MemOperand x86memop,
4700 PatFrag mem_frag, Intrinsic IntId> {
4701 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4702 (ins RC:$src1, RC:$src2, RC:$src3),
4703 !strconcat(OpcodeStr,
4704 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4705 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4706 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4708 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4709 (ins RC:$src1, x86memop:$src2, RC:$src3),
4710 !strconcat(OpcodeStr,
4711 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4713 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4715 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4719 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4720 memopv16i8, int_x86_sse41_blendvpd>;
4721 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4722 memopv16i8, int_x86_sse41_blendvps>;
4723 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4724 memopv16i8, int_x86_sse41_pblendvb>;
4725 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4726 memopv32i8, int_x86_avx_blendv_pd_256>;
4727 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4728 memopv32i8, int_x86_avx_blendv_ps_256>;
4730 /// SS41I_ternary_int - SSE 4.1 ternary operator
4731 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4732 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4733 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4734 (ins VR128:$src1, VR128:$src2),
4735 !strconcat(OpcodeStr,
4736 "\t{$src2, $dst|$dst, $src2}"),
4737 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4740 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4741 (ins VR128:$src1, i128mem:$src2),
4742 !strconcat(OpcodeStr,
4743 "\t{$src2, $dst|$dst, $src2}"),
4746 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4750 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4751 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4752 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4754 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4755 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4757 let Predicates = [HasAVX] in
4758 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4759 "vmovntdqa\t{$src, $dst|$dst, $src}",
4760 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4762 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4763 "movntdqa\t{$src, $dst|$dst, $src}",
4764 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4767 //===----------------------------------------------------------------------===//
4768 // SSE4.2 - Compare Instructions
4769 //===----------------------------------------------------------------------===//
4771 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4772 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4773 Intrinsic IntId128, bit Is2Addr = 1> {
4774 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4775 (ins VR128:$src1, VR128:$src2),
4777 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4778 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4779 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4781 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4782 (ins VR128:$src1, i128mem:$src2),
4784 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4785 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4787 (IntId128 VR128:$src1,
4788 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4791 let Predicates = [HasAVX] in
4792 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4794 let Constraints = "$src1 = $dst" in
4795 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4797 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4798 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4799 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4800 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4802 //===----------------------------------------------------------------------===//
4803 // SSE4.2 - String/text Processing Instructions
4804 //===----------------------------------------------------------------------===//
4806 // Packed Compare Implicit Length Strings, Return Mask
4807 multiclass pseudo_pcmpistrm<string asm> {
4808 def REG : PseudoI<(outs VR128:$dst),
4809 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4810 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4812 def MEM : PseudoI<(outs VR128:$dst),
4813 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4814 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4815 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4818 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4819 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4820 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4823 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4824 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4825 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4826 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4827 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4828 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4829 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4832 let Defs = [XMM0, EFLAGS] in {
4833 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4834 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4835 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4836 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4837 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4838 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4841 // Packed Compare Explicit Length Strings, Return Mask
4842 multiclass pseudo_pcmpestrm<string asm> {
4843 def REG : PseudoI<(outs VR128:$dst),
4844 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4845 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4846 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4847 def MEM : PseudoI<(outs VR128:$dst),
4848 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4849 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4850 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4853 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4854 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4855 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4858 let Predicates = [HasAVX],
4859 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4860 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4861 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4862 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4863 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4864 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4865 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4868 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4869 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4870 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4871 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4872 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4873 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4874 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4877 // Packed Compare Implicit Length Strings, Return Index
4878 let Defs = [ECX, EFLAGS] in {
4879 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4880 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4881 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4882 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4883 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4884 (implicit EFLAGS)]>, OpSize;
4885 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4886 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4887 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4888 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4889 (implicit EFLAGS)]>, OpSize;
4893 let Predicates = [HasAVX] in {
4894 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4896 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4898 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4900 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4902 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4904 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4908 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4909 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4910 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4911 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4912 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4913 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4915 // Packed Compare Explicit Length Strings, Return Index
4916 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4917 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4918 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4919 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4920 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4921 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4922 (implicit EFLAGS)]>, OpSize;
4923 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4924 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4925 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4927 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4928 (implicit EFLAGS)]>, OpSize;
4932 let Predicates = [HasAVX] in {
4933 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4935 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
4937 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
4939 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
4941 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
4943 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
4947 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
4948 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
4949 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
4950 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
4951 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
4952 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
4954 //===----------------------------------------------------------------------===//
4955 // SSE4.2 - CRC Instructions
4956 //===----------------------------------------------------------------------===//
4958 // No CRC instructions have AVX equivalents
4960 // crc intrinsic instruction
4961 // This set of instructions are only rm, the only difference is the size
4963 let Constraints = "$src1 = $dst" in {
4964 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
4965 (ins GR32:$src1, i8mem:$src2),
4966 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4968 (int_x86_sse42_crc32_8 GR32:$src1,
4969 (load addr:$src2)))]>;
4970 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
4971 (ins GR32:$src1, GR8:$src2),
4972 "crc32{b} \t{$src2, $src1|$src1, $src2}",
4974 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
4975 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4976 (ins GR32:$src1, i16mem:$src2),
4977 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4979 (int_x86_sse42_crc32_16 GR32:$src1,
4980 (load addr:$src2)))]>,
4982 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4983 (ins GR32:$src1, GR16:$src2),
4984 "crc32{w} \t{$src2, $src1|$src1, $src2}",
4986 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
4988 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
4989 (ins GR32:$src1, i32mem:$src2),
4990 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4992 (int_x86_sse42_crc32_32 GR32:$src1,
4993 (load addr:$src2)))]>;
4994 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
4995 (ins GR32:$src1, GR32:$src2),
4996 "crc32{l} \t{$src2, $src1|$src1, $src2}",
4998 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
4999 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5000 (ins GR64:$src1, i8mem:$src2),
5001 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5003 (int_x86_sse42_crc64_8 GR64:$src1,
5004 (load addr:$src2)))]>,
5006 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5007 (ins GR64:$src1, GR8:$src2),
5008 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5010 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5012 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5013 (ins GR64:$src1, i64mem:$src2),
5014 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5016 (int_x86_sse42_crc64_64 GR64:$src1,
5017 (load addr:$src2)))]>,
5019 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5020 (ins GR64:$src1, GR64:$src2),
5021 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5023 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5027 //===----------------------------------------------------------------------===//
5028 // AES-NI Instructions
5029 //===----------------------------------------------------------------------===//
5031 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5032 Intrinsic IntId128, bit Is2Addr = 1> {
5033 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5034 (ins VR128:$src1, VR128:$src2),
5036 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5037 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5038 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5040 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5041 (ins VR128:$src1, i128mem:$src2),
5043 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5044 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5046 (IntId128 VR128:$src1,
5047 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5050 // Perform One Round of an AES Encryption/Decryption Flow
5051 let Predicates = [HasAVX, HasAES] in {
5052 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5053 int_x86_aesni_aesenc, 0>, VEX_4V;
5054 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5055 int_x86_aesni_aesenclast, 0>, VEX_4V;
5056 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5057 int_x86_aesni_aesdec, 0>, VEX_4V;
5058 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5059 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5062 let Constraints = "$src1 = $dst" in {
5063 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5064 int_x86_aesni_aesenc>;
5065 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5066 int_x86_aesni_aesenclast>;
5067 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5068 int_x86_aesni_aesdec>;
5069 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5070 int_x86_aesni_aesdeclast>;
5073 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5074 (AESENCrr VR128:$src1, VR128:$src2)>;
5075 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5076 (AESENCrm VR128:$src1, addr:$src2)>;
5077 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5078 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5079 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5080 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5081 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5082 (AESDECrr VR128:$src1, VR128:$src2)>;
5083 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5084 (AESDECrm VR128:$src1, addr:$src2)>;
5085 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5086 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5087 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5088 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5090 // Perform the AES InvMixColumn Transformation
5091 let Predicates = [HasAVX, HasAES] in {
5092 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5094 "vaesimc\t{$src1, $dst|$dst, $src1}",
5096 (int_x86_aesni_aesimc VR128:$src1))]>,
5098 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5099 (ins i128mem:$src1),
5100 "vaesimc\t{$src1, $dst|$dst, $src1}",
5102 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5105 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5107 "aesimc\t{$src1, $dst|$dst, $src1}",
5109 (int_x86_aesni_aesimc VR128:$src1))]>,
5111 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5112 (ins i128mem:$src1),
5113 "aesimc\t{$src1, $dst|$dst, $src1}",
5115 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5118 // AES Round Key Generation Assist
5119 let Predicates = [HasAVX, HasAES] in {
5120 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5121 (ins VR128:$src1, i8imm:$src2),
5122 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5124 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5126 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5127 (ins i128mem:$src1, i8imm:$src2),
5128 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5130 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5134 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5135 (ins VR128:$src1, i8imm:$src2),
5136 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5138 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5140 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5141 (ins i128mem:$src1, i8imm:$src2),
5142 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5144 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5148 //===----------------------------------------------------------------------===//
5149 // CLMUL Instructions
5150 //===----------------------------------------------------------------------===//
5152 // Only the AVX version of CLMUL instructions are described here.
5154 // Carry-less Multiplication instructions
5155 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5156 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5157 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5160 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5161 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5162 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5166 multiclass avx_vpclmul<string asm> {
5167 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5168 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5171 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5172 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5175 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5176 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5177 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5178 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5180 //===----------------------------------------------------------------------===//
5182 //===----------------------------------------------------------------------===//
5185 // Load from memory and broadcast to all elements of the destination operand
5186 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5187 X86MemOperand x86memop, Intrinsic Int> :
5188 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5189 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5190 [(set RC:$dst, (Int addr:$src))]>, VEX;
5192 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5193 int_x86_avx_vbroadcastss>;
5194 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5195 int_x86_avx_vbroadcastss_256>;
5196 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5197 int_x86_avx_vbroadcast_sd_256>;
5198 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5199 int_x86_avx_vbroadcastf128_pd_256>;
5201 // Insert packed floating-point values
5202 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5203 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5204 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5206 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5207 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5208 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5211 // Extract packed floating-point values
5212 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5213 (ins VR256:$src1, i8imm:$src2),
5214 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5216 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5217 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5218 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5221 // Conditional SIMD Packed Loads and Stores
5222 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5223 Intrinsic IntLd, Intrinsic IntLd256,
5224 Intrinsic IntSt, Intrinsic IntSt256,
5225 PatFrag pf128, PatFrag pf256> {
5226 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5227 (ins VR128:$src1, f128mem:$src2),
5228 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5229 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5231 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5232 (ins VR256:$src1, f256mem:$src2),
5233 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5234 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5236 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5237 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5239 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5240 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5241 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5242 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5243 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5246 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5247 int_x86_avx_maskload_ps,
5248 int_x86_avx_maskload_ps_256,
5249 int_x86_avx_maskstore_ps,
5250 int_x86_avx_maskstore_ps_256,
5251 memopv4f32, memopv8f32>;
5252 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5253 int_x86_avx_maskload_pd,
5254 int_x86_avx_maskload_pd_256,
5255 int_x86_avx_maskstore_pd,
5256 int_x86_avx_maskstore_pd_256,
5257 memopv2f64, memopv4f64>;
5259 // Permute Floating-Point Values
5260 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5261 RegisterClass RC, X86MemOperand x86memop_f,
5262 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5263 Intrinsic IntVar, Intrinsic IntImm> {
5264 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5265 (ins RC:$src1, RC:$src2),
5266 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5267 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5268 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5269 (ins RC:$src1, x86memop_i:$src2),
5270 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5271 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5273 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5274 (ins RC:$src1, i8imm:$src2),
5275 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5276 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5277 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5278 (ins x86memop_f:$src1, i8imm:$src2),
5279 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5280 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5283 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5284 memopv4f32, memopv4i32,
5285 int_x86_avx_vpermilvar_ps,
5286 int_x86_avx_vpermil_ps>;
5287 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5288 memopv8f32, memopv8i32,
5289 int_x86_avx_vpermilvar_ps_256,
5290 int_x86_avx_vpermil_ps_256>;
5291 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5292 memopv2f64, memopv2i64,
5293 int_x86_avx_vpermilvar_pd,
5294 int_x86_avx_vpermil_pd>;
5295 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5296 memopv4f64, memopv4i64,
5297 int_x86_avx_vpermilvar_pd_256,
5298 int_x86_avx_vpermil_pd_256>;
5300 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5301 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5302 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5304 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5305 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5306 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5309 // Zero All YMM registers
5310 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5311 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5313 // Zero Upper bits of YMM registers
5314 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5315 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5317 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5318 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5319 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5320 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5321 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5322 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5324 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5326 (VINSERTF128rr VR256:$src1, VR128:$src2,
5327 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5328 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5330 (VINSERTF128rr VR256:$src1, VR128:$src2,
5331 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5332 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5334 (VINSERTF128rr VR256:$src1, VR128:$src2,
5335 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5336 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5338 (VINSERTF128rr VR256:$src1, VR128:$src2,
5339 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5341 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5342 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5343 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5344 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5345 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5346 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5348 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5349 (v4f32 (VEXTRACTF128rr
5350 (v8f32 VR256:$src1),
5351 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5352 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5353 (v2f64 (VEXTRACTF128rr
5354 (v4f64 VR256:$src1),
5355 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5356 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5357 (v4i32 (VEXTRACTF128rr
5358 (v8i32 VR256:$src1),
5359 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5360 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5361 (v2i64 (VEXTRACTF128rr
5362 (v4i64 VR256:$src1),
5363 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5365 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5366 (VBROADCASTF128 addr:$src)>;
5368 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5369 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5370 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5371 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5372 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5373 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5375 def : Pat<(int_x86_avx_vperm2f128_ps_256
5376 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5377 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5378 def : Pat<(int_x86_avx_vperm2f128_pd_256
5379 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5380 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5381 def : Pat<(int_x86_avx_vperm2f128_si_256
5382 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5383 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5385 //===----------------------------------------------------------------------===//
5386 // SSE Shuffle pattern fragments
5387 //===----------------------------------------------------------------------===//
5389 // This is part of a "work in progress" refactoring. The idea is that all
5390 // vector shuffles are going to be translated into target specific nodes and
5391 // directly matched by the patterns below (which can be changed along the way)
5392 // The AVX version of some but not all of them are described here, and more
5393 // should come in a near future.
5395 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5396 // SSE2 loads, which are always promoted to v2i64. The last one should match
5397 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5398 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5399 // we investigate further.
5400 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5402 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5403 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5405 (PSHUFDmi addr:$src1, imm:$imm)>;
5406 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5408 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5410 // Shuffle with PSHUFD instruction.
5411 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5412 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5413 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5414 (PSHUFDri VR128:$src1, imm:$imm)>;
5416 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5417 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5418 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5419 (PSHUFDri VR128:$src1, imm:$imm)>;
5421 // Shuffle with SHUFPD instruction.
5422 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5423 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5424 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5425 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5426 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5427 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5429 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5430 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5431 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5432 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5434 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5435 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5436 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5437 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5439 // Shuffle with SHUFPS instruction.
5440 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5441 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5442 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5443 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5444 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5445 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5447 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5448 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5449 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5450 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5452 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5453 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5454 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5455 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5456 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5457 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5459 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5460 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5461 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5462 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5464 // Shuffle with MOVHLPS instruction
5465 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5466 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5467 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5468 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5470 // Shuffle with MOVDDUP instruction
5471 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5472 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5473 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5474 (MOVDDUPrm addr:$src)>;
5476 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5477 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5478 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5479 (MOVDDUPrm addr:$src)>;
5481 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5482 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5483 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5484 (MOVDDUPrm addr:$src)>;
5486 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5487 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5488 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5489 (MOVDDUPrm addr:$src)>;
5491 def : Pat<(X86Movddup (bc_v2f64
5492 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5493 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5494 def : Pat<(X86Movddup (bc_v2f64
5495 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5496 (MOVDDUPrm addr:$src)>;
5499 // Shuffle with UNPCKLPS
5500 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5501 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5502 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5503 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5504 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5505 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5507 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5508 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5509 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5510 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5511 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5512 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5514 // Shuffle with UNPCKHPS
5515 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5516 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5517 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5518 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5520 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5521 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5522 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5523 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5525 // Shuffle with UNPCKLPD
5526 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5527 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5528 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5529 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5530 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5531 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5533 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5534 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5535 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5536 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5537 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5538 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5540 // Shuffle with UNPCKHPD
5541 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5542 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5543 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5544 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5546 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5547 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5548 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5549 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5551 // Shuffle with PUNPCKLBW
5552 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5553 (bc_v16i8 (memopv2i64 addr:$src2)))),
5554 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5555 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5556 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5558 // Shuffle with PUNPCKLWD
5559 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5560 (bc_v8i16 (memopv2i64 addr:$src2)))),
5561 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5562 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5563 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5565 // Shuffle with PUNPCKLDQ
5566 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5567 (bc_v4i32 (memopv2i64 addr:$src2)))),
5568 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5569 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5570 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5572 // Shuffle with PUNPCKLQDQ
5573 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5574 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5575 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5576 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5578 // Shuffle with PUNPCKHBW
5579 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5580 (bc_v16i8 (memopv2i64 addr:$src2)))),
5581 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5582 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5583 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5585 // Shuffle with PUNPCKHWD
5586 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5587 (bc_v8i16 (memopv2i64 addr:$src2)))),
5588 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5589 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5590 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5592 // Shuffle with PUNPCKHDQ
5593 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5594 (bc_v4i32 (memopv2i64 addr:$src2)))),
5595 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5596 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5597 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5599 // Shuffle with PUNPCKHQDQ
5600 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5601 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5602 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5603 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5605 // Shuffle with MOVLHPS
5606 def : Pat<(X86Movlhps VR128:$src1,
5607 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5608 (MOVHPSrm VR128:$src1, addr:$src2)>;
5609 def : Pat<(X86Movlhps VR128:$src1,
5610 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5611 (MOVHPSrm VR128:$src1, addr:$src2)>;
5612 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5613 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5614 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5615 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5616 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5617 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5619 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5620 // is during lowering, where it's not possible to recognize the load fold cause
5621 // it has two uses through a bitcast. One use disappears at isel time and the
5622 // fold opportunity reappears.
5623 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5624 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5626 // Shuffle with MOVLHPD
5627 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5628 (scalar_to_vector (loadf64 addr:$src2)))),
5629 (MOVHPDrm VR128:$src1, addr:$src2)>;
5631 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5632 // is during lowering, where it's not possible to recognize the load fold cause
5633 // it has two uses through a bitcast. One use disappears at isel time and the
5634 // fold opportunity reappears.
5635 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5636 (scalar_to_vector (loadf64 addr:$src2)))),
5637 (MOVHPDrm VR128:$src1, addr:$src2)>;
5639 // Shuffle with MOVSS
5640 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5641 (MOVSSrr VR128:$src1, FR32:$src2)>;
5642 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5643 (MOVSSrr (v4i32 VR128:$src1),
5644 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5645 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5646 (MOVSSrr (v4f32 VR128:$src1),
5647 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5648 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5649 // is during lowering, where it's not possible to recognize the load fold cause
5650 // it has two uses through a bitcast. One use disappears at isel time and the
5651 // fold opportunity reappears.
5652 def : Pat<(X86Movss VR128:$src1,
5653 (bc_v4i32 (v2i64 (load addr:$src2)))),
5654 (MOVLPSrm VR128:$src1, addr:$src2)>;
5656 // Shuffle with MOVSD
5657 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5658 (MOVSDrr VR128:$src1, FR64:$src2)>;
5659 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5660 (MOVSDrr (v2i64 VR128:$src1),
5661 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5662 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5663 (MOVSDrr (v2f64 VR128:$src1),
5664 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5665 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5666 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5667 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5668 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5670 // Shuffle with MOVSHDUP
5671 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5672 (MOVSHDUPrr VR128:$src)>;
5673 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5674 (MOVSHDUPrm addr:$src)>;
5676 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5677 (MOVSHDUPrr VR128:$src)>;
5678 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5679 (MOVSHDUPrm addr:$src)>;
5681 // Shuffle with MOVSLDUP
5682 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5683 (MOVSLDUPrr VR128:$src)>;
5684 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5685 (MOVSLDUPrm addr:$src)>;
5687 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5688 (MOVSLDUPrr VR128:$src)>;
5689 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5690 (MOVSLDUPrm addr:$src)>;
5692 // Shuffle with PSHUFHW
5693 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5694 (PSHUFHWri VR128:$src, imm:$imm)>;
5695 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5696 (PSHUFHWmi addr:$src, imm:$imm)>;
5698 // Shuffle with PSHUFLW
5699 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5700 (PSHUFLWri VR128:$src, imm:$imm)>;
5701 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5702 (PSHUFLWmi addr:$src, imm:$imm)>;
5704 // Shuffle with PALIGN
5705 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5706 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5707 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5708 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5709 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5710 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5711 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5712 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5714 // Shuffle with MOVLPS
5715 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5716 (MOVLPSrm VR128:$src1, addr:$src2)>;
5717 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5718 (MOVLPSrm VR128:$src1, addr:$src2)>;
5719 def : Pat<(X86Movlps VR128:$src1,
5720 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5721 (MOVLPSrm VR128:$src1, addr:$src2)>;
5722 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5723 // is during lowering, where it's not possible to recognize the load fold cause
5724 // it has two uses through a bitcast. One use disappears at isel time and the
5725 // fold opportunity reappears.
5726 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5727 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5729 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5730 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5732 // Shuffle with MOVLPD
5733 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5734 (MOVLPDrm VR128:$src1, addr:$src2)>;
5735 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5736 (MOVLPDrm VR128:$src1, addr:$src2)>;
5737 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5738 (scalar_to_vector (loadf64 addr:$src2)))),
5739 (MOVLPDrm VR128:$src1, addr:$src2)>;
5741 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5742 def : Pat<(store (f64 (vector_extract
5743 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5744 (MOVHPSmr addr:$dst, VR128:$src)>;
5745 def : Pat<(store (f64 (vector_extract
5746 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5747 (MOVHPDmr addr:$dst, VR128:$src)>;
5749 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5750 (MOVLPSmr addr:$src1, VR128:$src2)>;
5751 def : Pat<(store (v4i32 (X86Movlps
5752 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5753 (MOVLPSmr addr:$src1, VR128:$src2)>;
5755 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5756 (MOVLPDmr addr:$src1, VR128:$src2)>;
5757 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5758 (MOVLPDmr addr:$src1, VR128:$src2)>;