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 = $dst" in {
1060 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1061 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1062 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1063 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1064 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1065 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1066 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1067 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1068 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1069 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1070 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1071 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1072 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1073 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1074 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1075 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1077 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1078 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1079 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1080 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1081 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1082 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1083 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1084 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1085 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1086 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1087 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1088 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1089 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1092 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1093 Intrinsic Int, string asm> {
1094 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1095 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1096 [(set VR128:$dst, (Int VR128:$src1,
1097 VR128:$src, imm:$cc))]>;
1098 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1099 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1100 [(set VR128:$dst, (Int VR128:$src1,
1101 (load addr:$src), imm:$cc))]>;
1104 // Aliases to match intrinsics which expect XMM operand(s).
1105 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1106 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1108 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1109 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1111 let Constraints = "$src1 = $dst" in {
1112 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1113 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1114 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1115 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1119 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1120 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1121 ValueType vt, X86MemOperand x86memop,
1122 PatFrag ld_frag, string OpcodeStr, Domain d> {
1123 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1124 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1125 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1126 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1127 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1128 [(set EFLAGS, (OpNode (vt RC:$src1),
1129 (ld_frag addr:$src2)))], d>;
1132 let Defs = [EFLAGS] in {
1133 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1134 "ucomiss", SSEPackedSingle>, VEX;
1135 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1136 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1137 let Pattern = []<dag> in {
1138 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1139 "comiss", SSEPackedSingle>, VEX;
1140 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1141 "comisd", SSEPackedDouble>, OpSize, VEX;
1144 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1145 load, "ucomiss", SSEPackedSingle>, VEX;
1146 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1147 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1149 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1150 load, "comiss", SSEPackedSingle>, VEX;
1151 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1152 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1153 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1154 "ucomiss", SSEPackedSingle>, TB;
1155 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1156 "ucomisd", SSEPackedDouble>, TB, OpSize;
1158 let Pattern = []<dag> in {
1159 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1160 "comiss", SSEPackedSingle>, TB;
1161 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1162 "comisd", SSEPackedDouble>, TB, OpSize;
1165 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1166 load, "ucomiss", SSEPackedSingle>, TB;
1167 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1168 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1170 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1171 "comiss", SSEPackedSingle>, TB;
1172 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1173 "comisd", SSEPackedDouble>, TB, OpSize;
1174 } // Defs = [EFLAGS]
1176 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1177 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1178 Intrinsic Int, string asm, string asm_alt,
1180 let isAsmParserOnly = 1 in {
1181 def rri : PIi8<0xC2, MRMSrcReg,
1182 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1183 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1184 def rmi : PIi8<0xC2, MRMSrcMem,
1185 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1186 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1189 // Accept explicit immediate argument form instead of comparison code.
1190 def rri_alt : PIi8<0xC2, MRMSrcReg,
1191 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1193 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1194 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1198 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1199 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1200 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1201 SSEPackedSingle>, VEX_4V;
1202 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1203 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1204 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1205 SSEPackedDouble>, OpSize, VEX_4V;
1206 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1207 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1208 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1209 SSEPackedSingle>, VEX_4V;
1210 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1211 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1212 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1213 SSEPackedDouble>, OpSize, VEX_4V;
1214 let Constraints = "$src1 = $dst" in {
1215 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1216 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1217 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1218 SSEPackedSingle>, TB;
1219 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1220 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1221 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1222 SSEPackedDouble>, TB, OpSize;
1225 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1226 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1227 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1228 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1229 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1230 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1231 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1232 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1234 //===----------------------------------------------------------------------===//
1235 // SSE 1 & 2 - Shuffle Instructions
1236 //===----------------------------------------------------------------------===//
1238 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1239 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1240 ValueType vt, string asm, PatFrag mem_frag,
1241 Domain d, bit IsConvertibleToThreeAddress = 0> {
1242 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1243 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1244 [(set RC:$dst, (vt (shufp:$src3
1245 RC:$src1, (mem_frag addr:$src2))))], d>;
1246 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1247 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1248 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1250 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1253 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1254 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1255 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1256 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1257 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1258 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1259 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1260 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1261 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1262 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1263 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1264 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1266 let Constraints = "$src1 = $dst" in {
1267 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1268 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1269 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1271 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1272 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1273 memopv2f64, SSEPackedDouble>, TB, OpSize;
1276 //===----------------------------------------------------------------------===//
1277 // SSE 1 & 2 - Unpack Instructions
1278 //===----------------------------------------------------------------------===//
1280 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1281 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1282 PatFrag mem_frag, RegisterClass RC,
1283 X86MemOperand x86memop, string asm,
1285 def rr : PI<opc, MRMSrcReg,
1286 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1288 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1289 def rm : PI<opc, MRMSrcMem,
1290 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1292 (vt (OpNode RC:$src1,
1293 (mem_frag addr:$src2))))], d>;
1296 let AddedComplexity = 10 in {
1297 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1298 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1299 SSEPackedSingle>, VEX_4V;
1300 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1301 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1302 SSEPackedDouble>, OpSize, VEX_4V;
1303 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1304 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1305 SSEPackedSingle>, VEX_4V;
1306 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1307 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1308 SSEPackedDouble>, OpSize, VEX_4V;
1310 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1311 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1312 SSEPackedSingle>, VEX_4V;
1313 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1314 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1315 SSEPackedDouble>, OpSize, VEX_4V;
1316 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1317 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1318 SSEPackedSingle>, VEX_4V;
1319 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1320 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1321 SSEPackedDouble>, OpSize, VEX_4V;
1323 let Constraints = "$src1 = $dst" in {
1324 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1325 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1326 SSEPackedSingle>, TB;
1327 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1328 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1329 SSEPackedDouble>, TB, OpSize;
1330 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1331 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1332 SSEPackedSingle>, TB;
1333 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1334 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1335 SSEPackedDouble>, TB, OpSize;
1336 } // Constraints = "$src1 = $dst"
1337 } // AddedComplexity
1339 //===----------------------------------------------------------------------===//
1340 // SSE 1 & 2 - Extract Floating-Point Sign mask
1341 //===----------------------------------------------------------------------===//
1343 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1344 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1346 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1347 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1348 [(set GR32:$dst, (Int RC:$src))], d>;
1349 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1350 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1354 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1355 "movmskps", SSEPackedSingle>, VEX;
1356 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1357 "movmskpd", SSEPackedDouble>, OpSize,
1359 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1360 "movmskps", SSEPackedSingle>, VEX;
1361 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1362 "movmskpd", SSEPackedDouble>, OpSize,
1364 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1365 SSEPackedSingle>, TB;
1366 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1367 SSEPackedDouble>, TB, OpSize;
1370 def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
1371 "movmskpd\t{$src, $dst|$dst, $src}",
1372 [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1373 def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1374 "movmskpd\t{$src, $dst|$dst, $src}",
1375 [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1376 def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
1377 "movmskps\t{$src, $dst|$dst, $src}",
1378 [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1379 def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1380 "movmskps\t{$src, $dst|$dst, $src}",
1381 [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1384 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1385 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1386 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1387 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1389 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1390 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1391 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1392 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1395 //===----------------------------------------------------------------------===//
1396 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1397 //===----------------------------------------------------------------------===//
1399 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1400 // names that start with 'Fs'.
1402 // Alias instructions that map fld0 to pxor for sse.
1403 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1404 canFoldAsLoad = 1 in {
1405 // FIXME: Set encoding to pseudo!
1406 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1407 [(set FR32:$dst, fp32imm0)]>,
1408 Requires<[HasSSE1]>, TB, OpSize;
1409 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1410 [(set FR64:$dst, fpimm0)]>,
1411 Requires<[HasSSE2]>, TB, OpSize;
1412 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1413 [(set FR32:$dst, fp32imm0)]>,
1414 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1415 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1416 [(set FR64:$dst, fpimm0)]>,
1417 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1420 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1421 // bits are disregarded.
1422 let neverHasSideEffects = 1 in {
1423 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1424 "movaps\t{$src, $dst|$dst, $src}", []>;
1425 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1426 "movapd\t{$src, $dst|$dst, $src}", []>;
1429 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1430 // bits are disregarded.
1431 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1432 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1433 "movaps\t{$src, $dst|$dst, $src}",
1434 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1435 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1436 "movapd\t{$src, $dst|$dst, $src}",
1437 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1440 //===----------------------------------------------------------------------===//
1441 // SSE 1 & 2 - Logical Instructions
1442 //===----------------------------------------------------------------------===//
1444 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1446 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1448 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1449 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1451 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1452 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1454 let Constraints = "$src1 = $dst" in {
1455 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1456 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1458 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1459 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1463 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1464 let mayLoad = 0 in {
1465 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1466 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1467 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1470 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1471 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1473 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1475 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1476 SDNode OpNode, int HasPat = 0,
1477 list<list<dag>> Pattern = []> {
1478 let Pattern = []<dag> in {
1479 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1480 !strconcat(OpcodeStr, "ps"), f128mem,
1481 !if(HasPat, Pattern[0], // rr
1482 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1484 !if(HasPat, Pattern[2], // rm
1485 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1486 (memopv2i64 addr:$src2)))]), 0>,
1489 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1490 !strconcat(OpcodeStr, "pd"), f128mem,
1491 !if(HasPat, Pattern[1], // rr
1492 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1495 !if(HasPat, Pattern[3], // rm
1496 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1497 (memopv2i64 addr:$src2)))]), 0>,
1500 let Constraints = "$src1 = $dst" in {
1501 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1502 !strconcat(OpcodeStr, "ps"), f128mem,
1503 !if(HasPat, Pattern[0], // rr
1504 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1506 !if(HasPat, Pattern[2], // rm
1507 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1508 (memopv2i64 addr:$src2)))])>, TB;
1510 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1511 !strconcat(OpcodeStr, "pd"), f128mem,
1512 !if(HasPat, Pattern[1], // rr
1513 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1516 !if(HasPat, Pattern[3], // rm
1517 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1518 (memopv2i64 addr:$src2)))])>,
1523 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1525 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1526 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1527 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1529 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1530 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1533 // AVX 256-bit packed logical ops forms
1534 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1535 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1536 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1537 let isCommutable = 0 in
1538 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1540 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1541 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1542 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1543 let isCommutable = 0 in
1544 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1546 [(set VR128:$dst, (X86pandn VR128:$src1, VR128:$src2))],
1550 [(set VR128:$dst, (X86pandn VR128:$src1, (memopv2i64 addr:$src2)))],
1554 //===----------------------------------------------------------------------===//
1555 // SSE 1 & 2 - Arithmetic Instructions
1556 //===----------------------------------------------------------------------===//
1558 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1561 /// In addition, we also have a special variant of the scalar form here to
1562 /// represent the associated intrinsic operation. This form is unlike the
1563 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1564 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1566 /// These three forms can each be reg+reg or reg+mem.
1569 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1571 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1573 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1574 OpNode, FR32, f32mem, Is2Addr>, XS;
1575 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1576 OpNode, FR64, f64mem, Is2Addr>, XD;
1579 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1581 let mayLoad = 0 in {
1582 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1583 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1584 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1585 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1589 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1591 let mayLoad = 0 in {
1592 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1593 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1594 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1595 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1599 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1601 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1602 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1603 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1604 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1607 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1609 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1610 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1611 SSEPackedSingle, Is2Addr>, TB;
1613 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1614 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1615 SSEPackedDouble, Is2Addr>, TB, OpSize;
1618 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1619 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1620 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1621 SSEPackedSingle, 0>, TB;
1623 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1624 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1625 SSEPackedDouble, 0>, TB, OpSize;
1628 // Binary Arithmetic instructions
1629 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1630 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1631 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1632 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1633 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1634 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1635 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1636 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1638 let isCommutable = 0 in {
1639 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1640 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1641 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1642 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1643 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1644 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1645 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1646 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1647 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1648 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1649 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1650 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1651 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1652 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1653 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1654 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1655 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1656 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1657 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1658 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1661 let Constraints = "$src1 = $dst" in {
1662 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1663 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1664 basic_sse12_fp_binop_s_int<0x58, "add">;
1665 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1666 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1667 basic_sse12_fp_binop_s_int<0x59, "mul">;
1669 let isCommutable = 0 in {
1670 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1671 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1672 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1673 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1674 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1675 basic_sse12_fp_binop_s_int<0x5E, "div">;
1676 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1677 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1678 basic_sse12_fp_binop_s_int<0x5F, "max">,
1679 basic_sse12_fp_binop_p_int<0x5F, "max">;
1680 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1681 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1682 basic_sse12_fp_binop_s_int<0x5D, "min">,
1683 basic_sse12_fp_binop_p_int<0x5D, "min">;
1688 /// In addition, we also have a special variant of the scalar form here to
1689 /// represent the associated intrinsic operation. This form is unlike the
1690 /// plain scalar form, in that it takes an entire vector (instead of a
1691 /// scalar) and leaves the top elements undefined.
1693 /// And, we have a special variant form for a full-vector intrinsic form.
1695 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1696 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1697 SDNode OpNode, Intrinsic F32Int> {
1698 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1699 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1700 [(set FR32:$dst, (OpNode FR32:$src))]>;
1701 // For scalar unary operations, fold a load into the operation
1702 // only in OptForSize mode. It eliminates an instruction, but it also
1703 // eliminates a whole-register clobber (the load), so it introduces a
1704 // partial register update condition.
1705 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1706 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1707 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1708 Requires<[HasSSE1, OptForSize]>;
1709 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1710 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1711 [(set VR128:$dst, (F32Int VR128:$src))]>;
1712 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1713 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1714 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1717 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1718 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1719 SDNode OpNode, Intrinsic F32Int> {
1720 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1721 !strconcat(OpcodeStr,
1722 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1723 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1724 !strconcat(OpcodeStr,
1725 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1726 []>, XS, Requires<[HasAVX, OptForSize]>;
1727 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1728 !strconcat(OpcodeStr,
1729 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1730 [(set VR128:$dst, (F32Int VR128:$src))]>;
1731 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1732 !strconcat(OpcodeStr,
1733 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1734 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1737 /// sse1_fp_unop_p - SSE1 unops in packed form.
1738 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1739 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1740 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1741 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1742 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1743 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1744 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1747 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1748 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1749 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1750 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1751 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1752 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1753 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1754 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1757 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1758 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1759 Intrinsic V4F32Int> {
1760 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1761 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1762 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1763 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1764 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1765 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1768 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1769 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1770 Intrinsic V4F32Int> {
1771 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1772 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1773 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1774 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1775 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1776 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1779 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1780 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1781 SDNode OpNode, Intrinsic F64Int> {
1782 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1783 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1784 [(set FR64:$dst, (OpNode FR64:$src))]>;
1785 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1786 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1787 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1788 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1789 Requires<[HasSSE2, OptForSize]>;
1790 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1791 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1792 [(set VR128:$dst, (F64Int VR128:$src))]>;
1793 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1794 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1795 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1798 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1799 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1800 SDNode OpNode, Intrinsic F64Int> {
1801 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1802 !strconcat(OpcodeStr,
1803 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1804 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1805 (ins FR64:$src1, f64mem:$src2),
1806 !strconcat(OpcodeStr,
1807 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1808 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1809 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1810 [(set VR128:$dst, (F64Int VR128:$src))]>;
1811 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1812 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1813 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1816 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1817 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1819 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1820 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1821 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1822 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1823 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1824 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1827 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1828 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1829 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1830 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1831 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1832 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1833 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1834 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1837 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1838 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1839 Intrinsic V2F64Int> {
1840 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1841 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1842 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1843 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1844 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1845 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1848 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1849 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1850 Intrinsic V2F64Int> {
1851 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1852 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1853 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1854 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1855 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1856 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1859 let Predicates = [HasAVX] in {
1861 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1862 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1865 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1866 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1867 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1868 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1869 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1870 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1871 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1872 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1875 // Reciprocal approximations. Note that these typically require refinement
1876 // in order to obtain suitable precision.
1877 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1878 int_x86_sse_rsqrt_ss>, VEX_4V;
1879 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1880 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1881 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1882 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1884 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1886 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1887 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1888 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1889 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1893 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1894 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1895 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1896 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1897 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1898 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1900 // Reciprocal approximations. Note that these typically require refinement
1901 // in order to obtain suitable precision.
1902 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1903 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1904 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1905 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1906 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1907 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1909 // There is no f64 version of the reciprocal approximation instructions.
1911 //===----------------------------------------------------------------------===//
1912 // SSE 1 & 2 - Non-temporal stores
1913 //===----------------------------------------------------------------------===//
1915 let AddedComplexity = 400 in { // Prefer non-temporal versions
1916 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1917 (ins f128mem:$dst, VR128:$src),
1918 "movntps\t{$src, $dst|$dst, $src}",
1919 [(alignednontemporalstore (v4f32 VR128:$src),
1921 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1922 (ins f128mem:$dst, VR128:$src),
1923 "movntpd\t{$src, $dst|$dst, $src}",
1924 [(alignednontemporalstore (v2f64 VR128:$src),
1926 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1927 (ins f128mem:$dst, VR128:$src),
1928 "movntdq\t{$src, $dst|$dst, $src}",
1929 [(alignednontemporalstore (v2f64 VR128:$src),
1932 let ExeDomain = SSEPackedInt in
1933 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1934 (ins f128mem:$dst, VR128:$src),
1935 "movntdq\t{$src, $dst|$dst, $src}",
1936 [(alignednontemporalstore (v4f32 VR128:$src),
1939 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1940 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
1942 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1943 (ins f256mem:$dst, VR256:$src),
1944 "movntps\t{$src, $dst|$dst, $src}",
1945 [(alignednontemporalstore (v8f32 VR256:$src),
1947 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1948 (ins f256mem:$dst, VR256:$src),
1949 "movntpd\t{$src, $dst|$dst, $src}",
1950 [(alignednontemporalstore (v4f64 VR256:$src),
1952 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1953 (ins f256mem:$dst, VR256:$src),
1954 "movntdq\t{$src, $dst|$dst, $src}",
1955 [(alignednontemporalstore (v4f64 VR256:$src),
1957 let ExeDomain = SSEPackedInt in
1958 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1959 (ins f256mem:$dst, VR256:$src),
1960 "movntdq\t{$src, $dst|$dst, $src}",
1961 [(alignednontemporalstore (v8f32 VR256:$src),
1965 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1966 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1967 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
1968 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
1969 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
1970 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
1972 let AddedComplexity = 400 in { // Prefer non-temporal versions
1973 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1974 "movntps\t{$src, $dst|$dst, $src}",
1975 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1976 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1977 "movntpd\t{$src, $dst|$dst, $src}",
1978 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1980 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1981 "movntdq\t{$src, $dst|$dst, $src}",
1982 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1984 let ExeDomain = SSEPackedInt in
1985 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1986 "movntdq\t{$src, $dst|$dst, $src}",
1987 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1989 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1990 (MOVNTDQmr addr:$dst, VR128:$src)>;
1992 // There is no AVX form for instructions below this point
1993 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1994 "movnti{l}\t{$src, $dst|$dst, $src}",
1995 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
1996 TB, Requires<[HasSSE2]>;
1997 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
1998 "movnti{q}\t{$src, $dst|$dst, $src}",
1999 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2000 TB, Requires<[HasSSE2]>;
2003 //===----------------------------------------------------------------------===//
2004 // SSE 1 & 2 - Misc Instructions (No AVX form)
2005 //===----------------------------------------------------------------------===//
2007 // Prefetch intrinsic.
2008 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2009 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2010 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2011 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2012 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2013 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2014 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2015 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2017 // Load, store, and memory fence
2018 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2019 TB, Requires<[HasSSE1]>;
2020 def : Pat<(X86SFence), (SFENCE)>;
2022 // Alias instructions that map zero vector to pxor / xorp* for sse.
2023 // We set canFoldAsLoad because this can be converted to a constant-pool
2024 // load of an all-zeros value if folding it would be beneficial.
2025 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2026 // JIT implementation, it does not expand the instructions below like
2027 // X86MCInstLower does.
2028 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2029 isCodeGenOnly = 1 in {
2030 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2031 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2032 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2033 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2034 let ExeDomain = SSEPackedInt in
2035 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2036 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2039 // The same as done above but for AVX. The 128-bit versions are the
2040 // same, but re-encoded. The 256-bit does not support PI version.
2041 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2042 // JIT implementatioan, it does not expand the instructions below like
2043 // X86MCInstLower does.
2044 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2045 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2046 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2047 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2048 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2049 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2050 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2051 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2052 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2053 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2054 let ExeDomain = SSEPackedInt in
2055 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2056 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2059 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2060 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2061 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2063 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2064 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2066 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
2067 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
2068 // represent this instead of always zeroing SRC1. One possible solution is
2069 // to represent the instruction w/ something similar as the "$src1 = $dst"
2070 // constraint but without the tied operands.
2071 def : Pat<(extloadf32 addr:$src),
2072 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)), addr:$src)>,
2073 Requires<[HasAVX, OptForSpeed]>;
2075 //===----------------------------------------------------------------------===//
2076 // SSE 1 & 2 - Load/Store XCSR register
2077 //===----------------------------------------------------------------------===//
2079 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2080 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2081 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2082 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2084 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2085 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2086 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2087 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2089 //===---------------------------------------------------------------------===//
2090 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2091 //===---------------------------------------------------------------------===//
2093 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2095 let neverHasSideEffects = 1 in {
2096 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2097 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2098 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2099 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2101 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2102 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2103 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2104 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2106 let canFoldAsLoad = 1, mayLoad = 1 in {
2107 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2108 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2109 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2110 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2111 let Predicates = [HasAVX] in {
2112 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2113 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2114 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2115 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2119 let mayStore = 1 in {
2120 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2121 (ins i128mem:$dst, VR128:$src),
2122 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2123 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2124 (ins i256mem:$dst, VR256:$src),
2125 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2126 let Predicates = [HasAVX] in {
2127 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2128 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2129 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2130 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2134 let neverHasSideEffects = 1 in
2135 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2136 "movdqa\t{$src, $dst|$dst, $src}", []>;
2138 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2139 "movdqu\t{$src, $dst|$dst, $src}",
2140 []>, XS, Requires<[HasSSE2]>;
2142 let canFoldAsLoad = 1, mayLoad = 1 in {
2143 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2144 "movdqa\t{$src, $dst|$dst, $src}",
2145 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2146 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2147 "movdqu\t{$src, $dst|$dst, $src}",
2148 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2149 XS, Requires<[HasSSE2]>;
2152 let mayStore = 1 in {
2153 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2154 "movdqa\t{$src, $dst|$dst, $src}",
2155 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2156 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2157 "movdqu\t{$src, $dst|$dst, $src}",
2158 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2159 XS, Requires<[HasSSE2]>;
2162 // Intrinsic forms of MOVDQU load and store
2163 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2164 "vmovdqu\t{$src, $dst|$dst, $src}",
2165 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2166 XS, VEX, Requires<[HasAVX]>;
2168 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2169 "movdqu\t{$src, $dst|$dst, $src}",
2170 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2171 XS, Requires<[HasSSE2]>;
2173 } // ExeDomain = SSEPackedInt
2175 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2176 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2177 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2179 //===---------------------------------------------------------------------===//
2180 // SSE2 - Packed Integer Arithmetic Instructions
2181 //===---------------------------------------------------------------------===//
2183 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2185 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2186 bit IsCommutable = 0, bit Is2Addr = 1> {
2187 let isCommutable = IsCommutable in
2188 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2189 (ins VR128:$src1, VR128:$src2),
2191 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2192 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2193 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2194 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2195 (ins VR128:$src1, i128mem:$src2),
2197 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2198 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2199 [(set VR128:$dst, (IntId VR128:$src1,
2200 (bitconvert (memopv2i64 addr:$src2))))]>;
2203 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2204 string OpcodeStr, Intrinsic IntId,
2205 Intrinsic IntId2, bit Is2Addr = 1> {
2206 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2207 (ins VR128:$src1, VR128:$src2),
2209 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2210 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2211 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2212 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2213 (ins VR128:$src1, i128mem:$src2),
2215 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2216 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2217 [(set VR128:$dst, (IntId VR128:$src1,
2218 (bitconvert (memopv2i64 addr:$src2))))]>;
2219 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2220 (ins VR128:$src1, i32i8imm:$src2),
2222 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2223 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2224 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2227 /// PDI_binop_rm - Simple SSE2 binary operator.
2228 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2229 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2230 let isCommutable = IsCommutable in
2231 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2232 (ins VR128:$src1, VR128:$src2),
2234 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2235 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2236 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2237 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2238 (ins VR128:$src1, i128mem:$src2),
2240 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2241 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2242 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2243 (bitconvert (memopv2i64 addr:$src2)))))]>;
2246 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2248 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2249 /// to collapse (bitconvert VT to VT) into its operand.
2251 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2252 bit IsCommutable = 0, bit Is2Addr = 1> {
2253 let isCommutable = IsCommutable in
2254 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2255 (ins VR128:$src1, VR128:$src2),
2257 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2258 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2259 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2260 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2261 (ins VR128:$src1, i128mem:$src2),
2263 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2264 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2265 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2268 } // ExeDomain = SSEPackedInt
2270 // 128-bit Integer Arithmetic
2272 let Predicates = [HasAVX] in {
2273 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2274 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2275 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2276 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2277 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2278 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2279 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2280 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2281 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2284 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2286 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2288 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2290 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2292 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2294 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2296 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2298 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2300 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2302 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2304 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2306 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2308 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2310 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2312 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2314 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2316 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2318 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2320 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2324 let Constraints = "$src1 = $dst" in {
2325 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2326 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2327 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2328 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2329 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2330 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2331 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2332 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2333 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2336 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2337 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2338 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2339 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2340 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2341 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2342 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2343 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2344 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2345 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2346 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2347 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2348 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2349 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2350 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2351 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2352 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2353 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2354 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2356 } // Constraints = "$src1 = $dst"
2358 //===---------------------------------------------------------------------===//
2359 // SSE2 - Packed Integer Logical Instructions
2360 //===---------------------------------------------------------------------===//
2362 let Predicates = [HasAVX] in {
2363 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2364 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2366 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2367 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2369 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2370 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2373 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2374 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2376 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2377 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2379 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2380 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2383 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2384 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2386 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2387 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2390 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2391 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2392 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2394 let ExeDomain = SSEPackedInt in {
2395 let neverHasSideEffects = 1 in {
2396 // 128-bit logical shifts.
2397 def VPSLLDQri : PDIi8<0x73, MRM7r,
2398 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2399 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2401 def VPSRLDQri : PDIi8<0x73, MRM3r,
2402 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2403 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2405 // PSRADQri doesn't exist in SSE[1-3].
2407 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2408 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2409 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2410 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2411 VR128:$src2)))]>, VEX_4V;
2413 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2414 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2415 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2416 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2417 (memopv2i64 addr:$src2))))]>,
2422 let Constraints = "$src1 = $dst" in {
2423 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2424 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2425 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2426 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2427 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2428 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2430 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2431 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2432 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2433 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2434 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2435 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2437 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2438 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2439 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2440 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2442 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2443 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2444 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2446 let ExeDomain = SSEPackedInt in {
2447 let neverHasSideEffects = 1 in {
2448 // 128-bit logical shifts.
2449 def PSLLDQri : PDIi8<0x73, MRM7r,
2450 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2451 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2452 def PSRLDQri : PDIi8<0x73, MRM3r,
2453 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2454 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2455 // PSRADQri doesn't exist in SSE[1-3].
2457 def PANDNrr : PDI<0xDF, MRMSrcReg,
2458 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2459 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2461 def PANDNrm : PDI<0xDF, MRMSrcMem,
2462 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2463 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2465 } // Constraints = "$src1 = $dst"
2467 let Predicates = [HasAVX] in {
2468 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2469 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2470 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2471 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2472 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2473 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2474 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2475 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2476 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2477 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2479 // Shift up / down and insert zero's.
2480 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2481 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2482 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2483 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2486 let Predicates = [HasSSE2] in {
2487 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2488 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2489 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2490 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2491 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2492 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2493 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2494 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2495 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2496 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2498 // Shift up / down and insert zero's.
2499 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2500 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2501 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2502 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2505 //===---------------------------------------------------------------------===//
2506 // SSE2 - Packed Integer Comparison Instructions
2507 //===---------------------------------------------------------------------===//
2509 let Predicates = [HasAVX] in {
2510 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2512 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2514 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2516 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2518 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2520 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2524 let Constraints = "$src1 = $dst" in {
2525 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2526 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2527 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2528 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2529 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2530 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2531 } // Constraints = "$src1 = $dst"
2533 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2534 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2535 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2536 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2537 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2538 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2539 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2540 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2541 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2542 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2543 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2544 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2546 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2547 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2548 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2549 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2550 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2551 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2552 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2553 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2554 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2555 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2556 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2557 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2559 //===---------------------------------------------------------------------===//
2560 // SSE2 - Packed Integer Pack Instructions
2561 //===---------------------------------------------------------------------===//
2563 let Predicates = [HasAVX] in {
2564 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2566 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2568 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2572 let Constraints = "$src1 = $dst" in {
2573 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2574 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2575 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2576 } // Constraints = "$src1 = $dst"
2578 //===---------------------------------------------------------------------===//
2579 // SSE2 - Packed Integer Shuffle Instructions
2580 //===---------------------------------------------------------------------===//
2582 let ExeDomain = SSEPackedInt in {
2583 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2585 def ri : Ii8<0x70, MRMSrcReg,
2586 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2587 !strconcat(OpcodeStr,
2588 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2589 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2591 def mi : Ii8<0x70, MRMSrcMem,
2592 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2593 !strconcat(OpcodeStr,
2594 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2595 [(set VR128:$dst, (vt (pshuf_frag:$src2
2596 (bc_frag (memopv2i64 addr:$src1)),
2599 } // ExeDomain = SSEPackedInt
2601 let Predicates = [HasAVX] in {
2602 let AddedComplexity = 5 in
2603 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2606 // SSE2 with ImmT == Imm8 and XS prefix.
2607 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2610 // SSE2 with ImmT == Imm8 and XD prefix.
2611 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2615 let Predicates = [HasSSE2] in {
2616 let AddedComplexity = 5 in
2617 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2619 // SSE2 with ImmT == Imm8 and XS prefix.
2620 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2622 // SSE2 with ImmT == Imm8 and XD prefix.
2623 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2626 //===---------------------------------------------------------------------===//
2627 // SSE2 - Packed Integer Unpack Instructions
2628 //===---------------------------------------------------------------------===//
2630 let ExeDomain = SSEPackedInt in {
2631 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2632 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2633 def rr : PDI<opc, MRMSrcReg,
2634 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2636 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2637 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2638 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2639 def rm : PDI<opc, MRMSrcMem,
2640 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2642 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2643 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2644 [(set VR128:$dst, (unp_frag VR128:$src1,
2645 (bc_frag (memopv2i64
2649 let Predicates = [HasAVX] in {
2650 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2652 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2654 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2657 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2658 /// knew to collapse (bitconvert VT to VT) into its operand.
2659 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2660 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2661 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2663 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2664 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2665 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2666 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2668 (v2i64 (unpckl VR128:$src1,
2669 (memopv2i64 addr:$src2))))]>, VEX_4V;
2671 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2673 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2675 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2678 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2679 /// knew to collapse (bitconvert VT to VT) into its operand.
2680 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2681 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2682 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2684 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2685 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2686 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2687 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2689 (v2i64 (unpckh VR128:$src1,
2690 (memopv2i64 addr:$src2))))]>, VEX_4V;
2693 let Constraints = "$src1 = $dst" in {
2694 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2695 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2696 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2698 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2699 /// knew to collapse (bitconvert VT to VT) into its operand.
2700 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2701 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2702 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2704 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2705 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2706 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2707 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2709 (v2i64 (unpckl VR128:$src1,
2710 (memopv2i64 addr:$src2))))]>;
2712 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2713 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2714 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2716 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2717 /// knew to collapse (bitconvert VT to VT) into its operand.
2718 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2719 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2720 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2722 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2723 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2724 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2725 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2727 (v2i64 (unpckh VR128:$src1,
2728 (memopv2i64 addr:$src2))))]>;
2731 } // ExeDomain = SSEPackedInt
2733 //===---------------------------------------------------------------------===//
2734 // SSE2 - Packed Integer Extract and Insert
2735 //===---------------------------------------------------------------------===//
2737 let ExeDomain = SSEPackedInt in {
2738 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2739 def rri : Ii8<0xC4, MRMSrcReg,
2740 (outs VR128:$dst), (ins VR128:$src1,
2741 GR32:$src2, i32i8imm:$src3),
2743 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2744 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2746 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2747 def rmi : Ii8<0xC4, MRMSrcMem,
2748 (outs VR128:$dst), (ins VR128:$src1,
2749 i16mem:$src2, i32i8imm:$src3),
2751 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2752 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2754 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2759 let Predicates = [HasAVX] in
2760 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2761 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2762 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2763 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2764 imm:$src2))]>, OpSize, VEX;
2765 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2766 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2767 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2768 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2772 let Predicates = [HasAVX] in {
2773 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2774 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2775 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2776 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2777 []>, OpSize, VEX_4V;
2780 let Constraints = "$src1 = $dst" in
2781 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2783 } // ExeDomain = SSEPackedInt
2785 //===---------------------------------------------------------------------===//
2786 // SSE2 - Packed Mask Creation
2787 //===---------------------------------------------------------------------===//
2789 let ExeDomain = SSEPackedInt in {
2791 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2792 "pmovmskb\t{$src, $dst|$dst, $src}",
2793 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2794 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2795 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2796 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2797 "pmovmskb\t{$src, $dst|$dst, $src}",
2798 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2800 } // ExeDomain = SSEPackedInt
2802 //===---------------------------------------------------------------------===//
2803 // SSE2 - Conditional Store
2804 //===---------------------------------------------------------------------===//
2806 let ExeDomain = SSEPackedInt in {
2809 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2810 (ins VR128:$src, VR128:$mask),
2811 "maskmovdqu\t{$mask, $src|$src, $mask}",
2812 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2814 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2815 (ins VR128:$src, VR128:$mask),
2816 "maskmovdqu\t{$mask, $src|$src, $mask}",
2817 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2820 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2821 "maskmovdqu\t{$mask, $src|$src, $mask}",
2822 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2824 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2825 "maskmovdqu\t{$mask, $src|$src, $mask}",
2826 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2828 } // ExeDomain = SSEPackedInt
2830 //===---------------------------------------------------------------------===//
2831 // SSE2 - Move Doubleword
2832 //===---------------------------------------------------------------------===//
2834 // Move Int Doubleword to Packed Double Int
2835 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2836 "movd\t{$src, $dst|$dst, $src}",
2838 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2839 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2840 "movd\t{$src, $dst|$dst, $src}",
2842 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2844 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2845 "movd\t{$src, $dst|$dst, $src}",
2847 (v4i32 (scalar_to_vector GR32:$src)))]>;
2848 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2849 "movd\t{$src, $dst|$dst, $src}",
2851 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2852 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2853 "mov{d|q}\t{$src, $dst|$dst, $src}",
2855 (v2i64 (scalar_to_vector GR64:$src)))]>;
2856 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2857 "mov{d|q}\t{$src, $dst|$dst, $src}",
2858 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2861 // Move Int Doubleword to Single Scalar
2862 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2863 "movd\t{$src, $dst|$dst, $src}",
2864 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2866 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2867 "movd\t{$src, $dst|$dst, $src}",
2868 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2870 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2871 "movd\t{$src, $dst|$dst, $src}",
2872 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2874 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2875 "movd\t{$src, $dst|$dst, $src}",
2876 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2878 // Move Packed Doubleword Int to Packed Double Int
2879 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2880 "movd\t{$src, $dst|$dst, $src}",
2881 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2883 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2884 (ins i32mem:$dst, VR128:$src),
2885 "movd\t{$src, $dst|$dst, $src}",
2886 [(store (i32 (vector_extract (v4i32 VR128:$src),
2887 (iPTR 0))), addr:$dst)]>, VEX;
2888 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2889 "movd\t{$src, $dst|$dst, $src}",
2890 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2892 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2893 "movd\t{$src, $dst|$dst, $src}",
2894 [(store (i32 (vector_extract (v4i32 VR128:$src),
2895 (iPTR 0))), addr:$dst)]>;
2897 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2898 "mov{d|q}\t{$src, $dst|$dst, $src}",
2899 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2901 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2902 "movq\t{$src, $dst|$dst, $src}",
2903 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2905 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2906 "mov{d|q}\t{$src, $dst|$dst, $src}",
2907 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2908 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2909 "movq\t{$src, $dst|$dst, $src}",
2910 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2912 // Move Scalar Single to Double Int
2913 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2914 "movd\t{$src, $dst|$dst, $src}",
2915 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2916 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2917 "movd\t{$src, $dst|$dst, $src}",
2918 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2919 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2920 "movd\t{$src, $dst|$dst, $src}",
2921 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2922 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2923 "movd\t{$src, $dst|$dst, $src}",
2924 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2926 // movd / movq to XMM register zero-extends
2927 let AddedComplexity = 15 in {
2928 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2929 "movd\t{$src, $dst|$dst, $src}",
2930 [(set VR128:$dst, (v4i32 (X86vzmovl
2931 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2933 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2934 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2935 [(set VR128:$dst, (v2i64 (X86vzmovl
2936 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2939 let AddedComplexity = 15 in {
2940 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2941 "movd\t{$src, $dst|$dst, $src}",
2942 [(set VR128:$dst, (v4i32 (X86vzmovl
2943 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2944 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2945 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2946 [(set VR128:$dst, (v2i64 (X86vzmovl
2947 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2950 let AddedComplexity = 20 in {
2951 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2954 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2955 (loadi32 addr:$src))))))]>,
2957 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2958 "movd\t{$src, $dst|$dst, $src}",
2960 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2961 (loadi32 addr:$src))))))]>;
2963 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2964 (MOVZDI2PDIrm addr:$src)>;
2965 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2966 (MOVZDI2PDIrm addr:$src)>;
2967 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2968 (MOVZDI2PDIrm addr:$src)>;
2971 // These are the correct encodings of the instructions so that we know how to
2972 // read correct assembly, even though we continue to emit the wrong ones for
2973 // compatibility with Darwin's buggy assembler.
2974 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2975 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
2976 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2977 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
2978 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2979 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
2980 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2981 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
2982 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2983 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2984 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2985 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2987 //===---------------------------------------------------------------------===//
2988 // SSE2 - Move Quadword
2989 //===---------------------------------------------------------------------===//
2991 // Move Quadword Int to Packed Quadword Int
2992 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2993 "vmovq\t{$src, $dst|$dst, $src}",
2995 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2996 VEX, Requires<[HasAVX]>;
2997 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2998 "movq\t{$src, $dst|$dst, $src}",
3000 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3001 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3003 // Move Packed Quadword Int to Quadword Int
3004 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3005 "movq\t{$src, $dst|$dst, $src}",
3006 [(store (i64 (vector_extract (v2i64 VR128:$src),
3007 (iPTR 0))), addr:$dst)]>, VEX;
3008 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3009 "movq\t{$src, $dst|$dst, $src}",
3010 [(store (i64 (vector_extract (v2i64 VR128:$src),
3011 (iPTR 0))), addr:$dst)]>;
3013 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3014 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3016 // Store / copy lower 64-bits of a XMM register.
3017 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3018 "movq\t{$src, $dst|$dst, $src}",
3019 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3020 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3021 "movq\t{$src, $dst|$dst, $src}",
3022 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3024 let AddedComplexity = 20 in
3025 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3026 "vmovq\t{$src, $dst|$dst, $src}",
3028 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3029 (loadi64 addr:$src))))))]>,
3030 XS, VEX, Requires<[HasAVX]>;
3032 let AddedComplexity = 20 in {
3033 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3034 "movq\t{$src, $dst|$dst, $src}",
3036 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3037 (loadi64 addr:$src))))))]>,
3038 XS, Requires<[HasSSE2]>;
3040 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3041 (MOVZQI2PQIrm addr:$src)>;
3042 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3043 (MOVZQI2PQIrm addr:$src)>;
3044 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3047 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3048 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3049 let AddedComplexity = 15 in
3050 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3051 "vmovq\t{$src, $dst|$dst, $src}",
3052 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3053 XS, VEX, Requires<[HasAVX]>;
3054 let AddedComplexity = 15 in
3055 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3056 "movq\t{$src, $dst|$dst, $src}",
3057 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3058 XS, Requires<[HasSSE2]>;
3060 let AddedComplexity = 20 in
3061 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3062 "vmovq\t{$src, $dst|$dst, $src}",
3063 [(set VR128:$dst, (v2i64 (X86vzmovl
3064 (loadv2i64 addr:$src))))]>,
3065 XS, VEX, Requires<[HasAVX]>;
3066 let AddedComplexity = 20 in {
3067 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3068 "movq\t{$src, $dst|$dst, $src}",
3069 [(set VR128:$dst, (v2i64 (X86vzmovl
3070 (loadv2i64 addr:$src))))]>,
3071 XS, Requires<[HasSSE2]>;
3073 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3074 (MOVZPQILo2PQIrm addr:$src)>;
3077 // Instructions to match in the assembler
3078 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3079 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3080 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3081 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3082 // Recognize "movd" with GR64 destination, but encode as a "movq"
3083 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3084 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3086 // Instructions for the disassembler
3087 // xr = XMM register
3090 let Predicates = [HasAVX] in
3091 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3092 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3093 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3094 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3096 //===---------------------------------------------------------------------===//
3097 // SSE2 - Misc Instructions
3098 //===---------------------------------------------------------------------===//
3101 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3102 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3103 TB, Requires<[HasSSE2]>;
3105 // Load, store, and memory fence
3106 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3107 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3108 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3109 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3110 def : Pat<(X86LFence), (LFENCE)>;
3111 def : Pat<(X86MFence), (MFENCE)>;
3114 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3115 // was introduced with SSE2, it's backward compatible.
3116 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3118 // Alias instructions that map zero vector to pxor / xorp* for sse.
3119 // We set canFoldAsLoad because this can be converted to a constant-pool
3120 // load of an all-ones value if folding it would be beneficial.
3121 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3122 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3123 // FIXME: Change encoding to pseudo.
3124 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3125 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3127 //===---------------------------------------------------------------------===//
3128 // SSE3 - Conversion Instructions
3129 //===---------------------------------------------------------------------===//
3131 // Convert Packed Double FP to Packed DW Integers
3132 let Predicates = [HasAVX] in {
3133 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3134 // register, but the same isn't true when using memory operands instead.
3135 // Provide other assembly rr and rm forms to address this explicitly.
3136 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3137 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3138 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3139 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3142 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3143 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3144 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3145 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3148 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3149 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3150 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3151 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3154 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3155 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3156 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3157 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3159 // Convert Packed DW Integers to Packed Double FP
3160 let Predicates = [HasAVX] in {
3161 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3162 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3163 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3164 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3165 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3166 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3167 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3168 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3171 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3172 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3173 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3174 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3176 // AVX 256-bit register conversion intrinsics
3177 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3178 (VCVTDQ2PDYrr VR128:$src)>;
3179 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3180 (VCVTDQ2PDYrm addr:$src)>;
3182 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3183 (VCVTPD2DQYrr VR256:$src)>;
3184 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3185 (VCVTPD2DQYrm addr:$src)>;
3187 //===---------------------------------------------------------------------===//
3188 // SSE3 - Move Instructions
3189 //===---------------------------------------------------------------------===//
3191 // Replicate Single FP
3192 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3193 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3194 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3195 [(set VR128:$dst, (v4f32 (rep_frag
3196 VR128:$src, (undef))))]>;
3197 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3198 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3199 [(set VR128:$dst, (rep_frag
3200 (memopv4f32 addr:$src), (undef)))]>;
3203 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3205 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3206 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3207 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3208 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3211 let Predicates = [HasAVX] in {
3212 // FIXME: Merge above classes when we have patterns for the ymm version
3213 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3214 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3215 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3216 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3218 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3219 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3221 // Replicate Double FP
3222 multiclass sse3_replicate_dfp<string OpcodeStr> {
3223 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3224 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3225 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3226 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3227 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3229 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3233 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3234 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3235 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3237 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3238 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3242 let Predicates = [HasAVX] in {
3243 // FIXME: Merge above classes when we have patterns for the ymm version
3244 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3245 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3247 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3249 // Move Unaligned Integer
3250 let Predicates = [HasAVX] in {
3251 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3252 "vlddqu\t{$src, $dst|$dst, $src}",
3253 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3254 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3255 "vlddqu\t{$src, $dst|$dst, $src}",
3256 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3258 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3259 "lddqu\t{$src, $dst|$dst, $src}",
3260 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3262 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3264 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3266 // Several Move patterns
3267 let AddedComplexity = 5 in {
3268 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3269 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3270 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3271 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3272 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3273 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3274 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3275 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3278 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3279 let AddedComplexity = 15 in
3280 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3281 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3282 let AddedComplexity = 20 in
3283 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3284 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3286 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3287 let AddedComplexity = 15 in
3288 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3289 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3290 let AddedComplexity = 20 in
3291 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3292 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3294 //===---------------------------------------------------------------------===//
3295 // SSE3 - Arithmetic
3296 //===---------------------------------------------------------------------===//
3298 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3299 X86MemOperand x86memop, bit Is2Addr = 1> {
3300 def rr : I<0xD0, MRMSrcReg,
3301 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3303 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3304 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3305 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3306 def rm : I<0xD0, MRMSrcMem,
3307 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3309 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3310 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3311 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3314 let Predicates = [HasAVX],
3315 ExeDomain = SSEPackedDouble in {
3316 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3317 f128mem, 0>, TB, XD, VEX_4V;
3318 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3319 f128mem, 0>, TB, OpSize, VEX_4V;
3320 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3321 f256mem, 0>, TB, XD, VEX_4V;
3322 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3323 f256mem, 0>, TB, OpSize, VEX_4V;
3325 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3326 ExeDomain = SSEPackedDouble in {
3327 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3329 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3330 f128mem>, TB, OpSize;
3333 //===---------------------------------------------------------------------===//
3334 // SSE3 Instructions
3335 //===---------------------------------------------------------------------===//
3338 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3339 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3340 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3342 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3343 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3344 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3346 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3348 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3350 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3352 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3353 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3354 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3356 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3357 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3358 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3360 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3362 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3363 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3364 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3367 let Predicates = [HasAVX] in {
3368 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3369 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3370 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3371 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3372 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3373 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3374 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3375 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3376 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3377 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3378 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3379 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3380 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3381 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3382 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3383 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3386 let Constraints = "$src1 = $dst" in {
3387 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3388 int_x86_sse3_hadd_ps>;
3389 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3390 int_x86_sse3_hadd_pd>;
3391 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3392 int_x86_sse3_hsub_ps>;
3393 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3394 int_x86_sse3_hsub_pd>;
3397 //===---------------------------------------------------------------------===//
3398 // SSSE3 - Packed Absolute Instructions
3399 //===---------------------------------------------------------------------===//
3402 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3403 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3404 PatFrag mem_frag128, Intrinsic IntId128> {
3405 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3407 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3408 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3411 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3413 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3416 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3419 let Predicates = [HasAVX] in {
3420 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3421 int_x86_ssse3_pabs_b_128>, VEX;
3422 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3423 int_x86_ssse3_pabs_w_128>, VEX;
3424 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3425 int_x86_ssse3_pabs_d_128>, VEX;
3428 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3429 int_x86_ssse3_pabs_b_128>;
3430 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3431 int_x86_ssse3_pabs_w_128>;
3432 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3433 int_x86_ssse3_pabs_d_128>;
3435 //===---------------------------------------------------------------------===//
3436 // SSSE3 - Packed Binary Operator Instructions
3437 //===---------------------------------------------------------------------===//
3439 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3440 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3441 PatFrag mem_frag128, Intrinsic IntId128,
3443 let isCommutable = 1 in
3444 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3445 (ins VR128:$src1, VR128:$src2),
3447 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3449 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3451 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3452 (ins VR128:$src1, i128mem:$src2),
3454 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3455 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3457 (IntId128 VR128:$src1,
3458 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3461 let Predicates = [HasAVX] in {
3462 let isCommutable = 0 in {
3463 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3464 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3465 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3466 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3467 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3468 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3469 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3470 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3471 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3472 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3473 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3474 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3475 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3476 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3477 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3478 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3479 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3480 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3481 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3482 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3483 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3484 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3486 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3487 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3490 // None of these have i8 immediate fields.
3491 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3492 let isCommutable = 0 in {
3493 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3494 int_x86_ssse3_phadd_w_128>;
3495 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3496 int_x86_ssse3_phadd_d_128>;
3497 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3498 int_x86_ssse3_phadd_sw_128>;
3499 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3500 int_x86_ssse3_phsub_w_128>;
3501 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3502 int_x86_ssse3_phsub_d_128>;
3503 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3504 int_x86_ssse3_phsub_sw_128>;
3505 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3506 int_x86_ssse3_pmadd_ub_sw_128>;
3507 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3508 int_x86_ssse3_pshuf_b_128>;
3509 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3510 int_x86_ssse3_psign_b_128>;
3511 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3512 int_x86_ssse3_psign_w_128>;
3513 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3514 int_x86_ssse3_psign_d_128>;
3516 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3517 int_x86_ssse3_pmul_hr_sw_128>;
3520 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3521 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3522 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3523 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3525 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3526 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3527 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3528 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3529 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3530 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3532 //===---------------------------------------------------------------------===//
3533 // SSSE3 - Packed Align Instruction Patterns
3534 //===---------------------------------------------------------------------===//
3536 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3537 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3538 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3540 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3542 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3544 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3545 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3547 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3549 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3553 let Predicates = [HasAVX] in
3554 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3555 let Constraints = "$src1 = $dst" in
3556 defm PALIGN : ssse3_palign<"palignr">;
3558 let AddedComplexity = 5 in {
3559 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3560 (PALIGNR128rr VR128:$src2, VR128:$src1,
3561 (SHUFFLE_get_palign_imm VR128:$src3))>,
3562 Requires<[HasSSSE3]>;
3563 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3564 (PALIGNR128rr VR128:$src2, VR128:$src1,
3565 (SHUFFLE_get_palign_imm VR128:$src3))>,
3566 Requires<[HasSSSE3]>;
3567 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3568 (PALIGNR128rr VR128:$src2, VR128:$src1,
3569 (SHUFFLE_get_palign_imm VR128:$src3))>,
3570 Requires<[HasSSSE3]>;
3571 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3572 (PALIGNR128rr VR128:$src2, VR128:$src1,
3573 (SHUFFLE_get_palign_imm VR128:$src3))>,
3574 Requires<[HasSSSE3]>;
3577 //===---------------------------------------------------------------------===//
3578 // SSSE3 Misc Instructions
3579 //===---------------------------------------------------------------------===//
3581 // Thread synchronization
3582 let usesCustomInserter = 1 in {
3583 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3584 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3585 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3586 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3589 let Uses = [EAX, ECX, EDX] in
3590 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3591 Requires<[HasSSE3]>;
3592 let Uses = [ECX, EAX] in
3593 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3594 Requires<[HasSSE3]>;
3596 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3597 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3599 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3600 Requires<[In32BitMode]>;
3601 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3602 Requires<[In64BitMode]>;
3604 //===---------------------------------------------------------------------===//
3605 // Non-Instruction Patterns
3606 //===---------------------------------------------------------------------===//
3608 // extload f32 -> f64. This matches load+fextend because we have a hack in
3609 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3611 // Since these loads aren't folded into the fextend, we have to match it
3613 let Predicates = [HasSSE2] in
3614 def : Pat<(fextend (loadf32 addr:$src)),
3615 (CVTSS2SDrm addr:$src)>;
3617 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
3618 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
3619 // represent this instead of always zeroing SRC1. One possible solution is
3620 // to represent the instruction w/ something similar as the "$src1 = $dst"
3621 // constraint but without the tied operands.
3622 let Predicates = [HasAVX] in
3623 def : Pat<(fextend (loadf32 addr:$src)),
3624 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)),
3628 let Predicates = [HasXMMInt] in {
3629 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3630 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3631 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3632 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3633 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3634 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3635 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3636 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3637 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3638 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3639 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3640 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3641 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3642 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3643 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3644 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3645 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3646 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3647 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3648 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3649 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3650 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3651 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3652 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3653 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3654 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3655 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3656 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3657 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3658 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3661 let Predicates = [HasAVX] in {
3662 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3665 // Move scalar to XMM zero-extended
3666 // movd to XMM register zero-extends
3667 let AddedComplexity = 15 in {
3668 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3669 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3670 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3671 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3672 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3673 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3674 (MOVSSrr (v4f32 (V_SET0PS)),
3675 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3676 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3677 (MOVSSrr (v4i32 (V_SET0PI)),
3678 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3681 // Splat v2f64 / v2i64
3682 let AddedComplexity = 10 in {
3683 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3684 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3685 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3686 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3687 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3688 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3689 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3690 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3693 // Special unary SHUFPSrri case.
3694 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3695 (SHUFPSrri VR128:$src1, VR128:$src1,
3696 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3697 let AddedComplexity = 5 in
3698 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3699 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3700 Requires<[HasSSE2]>;
3701 // Special unary SHUFPDrri case.
3702 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3703 (SHUFPDrri VR128:$src1, VR128:$src1,
3704 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3705 Requires<[HasSSE2]>;
3706 // Special unary SHUFPDrri case.
3707 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3708 (SHUFPDrri VR128:$src1, VR128:$src1,
3709 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3710 Requires<[HasSSE2]>;
3711 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3712 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3713 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3714 Requires<[HasSSE2]>;
3716 // Special binary v4i32 shuffle cases with SHUFPS.
3717 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3718 (SHUFPSrri VR128:$src1, VR128:$src2,
3719 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3720 Requires<[HasSSE2]>;
3721 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3722 (SHUFPSrmi VR128:$src1, addr:$src2,
3723 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3724 Requires<[HasSSE2]>;
3725 // Special binary v2i64 shuffle cases using SHUFPDrri.
3726 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3727 (SHUFPDrri VR128:$src1, VR128:$src2,
3728 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3729 Requires<[HasSSE2]>;
3731 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3732 let AddedComplexity = 15 in {
3733 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3734 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3735 Requires<[OptForSpeed, HasSSE2]>;
3736 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3737 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3738 Requires<[OptForSpeed, HasSSE2]>;
3740 let AddedComplexity = 10 in {
3741 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3742 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3743 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3744 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3745 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3746 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3747 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3748 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3751 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3752 let AddedComplexity = 15 in {
3753 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3754 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3755 Requires<[OptForSpeed, HasSSE2]>;
3756 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3757 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3758 Requires<[OptForSpeed, HasSSE2]>;
3760 let AddedComplexity = 10 in {
3761 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3762 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3763 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3764 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3765 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3766 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3767 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3768 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3771 let AddedComplexity = 20 in {
3772 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3773 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3774 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3776 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3777 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3778 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3780 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3781 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3782 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3783 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3784 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3787 let AddedComplexity = 20 in {
3788 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3789 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3790 (MOVLPSrm VR128:$src1, addr:$src2)>;
3791 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3792 (MOVLPDrm VR128:$src1, addr:$src2)>;
3793 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3794 (MOVLPSrm VR128:$src1, addr:$src2)>;
3795 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3796 (MOVLPDrm VR128:$src1, addr:$src2)>;
3799 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3800 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3801 (MOVLPSmr addr:$src1, VR128:$src2)>;
3802 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3803 (MOVLPDmr addr:$src1, VR128:$src2)>;
3804 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3806 (MOVLPSmr addr:$src1, VR128:$src2)>;
3807 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3808 (MOVLPDmr addr:$src1, VR128:$src2)>;
3810 let AddedComplexity = 15 in {
3811 // Setting the lowest element in the vector.
3812 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3813 (MOVSSrr (v4i32 VR128:$src1),
3814 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3815 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3816 (MOVSDrr (v2i64 VR128:$src1),
3817 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3819 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3820 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3821 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3822 Requires<[HasSSE2]>;
3823 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3824 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3825 Requires<[HasSSE2]>;
3828 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3829 // fall back to this for SSE1)
3830 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3831 (SHUFPSrri VR128:$src2, VR128:$src1,
3832 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3834 // Set lowest element and zero upper elements.
3835 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3836 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3838 // vector -> vector casts
3839 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3840 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3841 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3842 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3844 // Use movaps / movups for SSE integer load / store (one byte shorter).
3845 let Predicates = [HasSSE1] in {
3846 def : Pat<(alignedloadv4i32 addr:$src),
3847 (MOVAPSrm addr:$src)>;
3848 def : Pat<(loadv4i32 addr:$src),
3849 (MOVUPSrm addr:$src)>;
3850 def : Pat<(alignedloadv2i64 addr:$src),
3851 (MOVAPSrm addr:$src)>;
3852 def : Pat<(loadv2i64 addr:$src),
3853 (MOVUPSrm addr:$src)>;
3855 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3856 (MOVAPSmr addr:$dst, VR128:$src)>;
3857 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3858 (MOVAPSmr addr:$dst, VR128:$src)>;
3859 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3860 (MOVAPSmr addr:$dst, VR128:$src)>;
3861 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3862 (MOVAPSmr addr:$dst, VR128:$src)>;
3863 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3864 (MOVUPSmr addr:$dst, VR128:$src)>;
3865 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3866 (MOVUPSmr addr:$dst, VR128:$src)>;
3867 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3868 (MOVUPSmr addr:$dst, VR128:$src)>;
3869 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3870 (MOVUPSmr addr:$dst, VR128:$src)>;
3873 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3874 let Predicates = [HasAVX] in {
3875 def : Pat<(alignedloadv4i32 addr:$src),
3876 (VMOVAPSrm addr:$src)>;
3877 def : Pat<(loadv4i32 addr:$src),
3878 (VMOVUPSrm addr:$src)>;
3879 def : Pat<(alignedloadv2i64 addr:$src),
3880 (VMOVAPSrm addr:$src)>;
3881 def : Pat<(loadv2i64 addr:$src),
3882 (VMOVUPSrm addr:$src)>;
3884 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3885 (VMOVAPSmr addr:$dst, VR128:$src)>;
3886 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3887 (VMOVAPSmr addr:$dst, VR128:$src)>;
3888 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3889 (VMOVAPSmr addr:$dst, VR128:$src)>;
3890 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3891 (VMOVAPSmr addr:$dst, VR128:$src)>;
3892 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3893 (VMOVUPSmr addr:$dst, VR128:$src)>;
3894 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3895 (VMOVUPSmr addr:$dst, VR128:$src)>;
3896 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3897 (VMOVUPSmr addr:$dst, VR128:$src)>;
3898 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3899 (VMOVUPSmr addr:$dst, VR128:$src)>;
3902 //===----------------------------------------------------------------------===//
3903 // SSE4.1 - Packed Move with Sign/Zero Extend
3904 //===----------------------------------------------------------------------===//
3906 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3907 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3908 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3909 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3911 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3912 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3914 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3918 let Predicates = [HasAVX] in {
3919 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3921 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3923 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3925 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3927 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3929 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3933 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3934 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3935 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3936 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3937 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3938 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3940 // Common patterns involving scalar load.
3941 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3942 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3943 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3944 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3946 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3947 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3948 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3949 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3951 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3952 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3953 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3954 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3956 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3957 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3958 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3959 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3961 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3962 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3963 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3964 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3966 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3967 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3968 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3969 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3972 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3973 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3974 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3975 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3977 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3978 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3980 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
3984 let Predicates = [HasAVX] in {
3985 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
3987 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
3989 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
3991 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
3995 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
3996 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
3997 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
3998 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4000 // Common patterns involving scalar load
4001 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4002 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4003 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4004 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4006 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4007 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4008 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4009 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4012 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4013 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4014 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4015 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4017 // Expecting a i16 load any extended to i32 value.
4018 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4019 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4020 [(set VR128:$dst, (IntId (bitconvert
4021 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4025 let Predicates = [HasAVX] in {
4026 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4028 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4031 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4032 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4034 // Common patterns involving scalar load
4035 def : Pat<(int_x86_sse41_pmovsxbq
4036 (bitconvert (v4i32 (X86vzmovl
4037 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4038 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4040 def : Pat<(int_x86_sse41_pmovzxbq
4041 (bitconvert (v4i32 (X86vzmovl
4042 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4043 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4045 //===----------------------------------------------------------------------===//
4046 // SSE4.1 - Extract Instructions
4047 //===----------------------------------------------------------------------===//
4049 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4050 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4051 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4052 (ins VR128:$src1, i32i8imm:$src2),
4053 !strconcat(OpcodeStr,
4054 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4055 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4057 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4058 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4059 !strconcat(OpcodeStr,
4060 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4063 // There's an AssertZext in the way of writing the store pattern
4064 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4067 let Predicates = [HasAVX] in {
4068 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4069 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4070 (ins VR128:$src1, i32i8imm:$src2),
4071 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4074 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4077 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4078 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4079 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4080 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4081 !strconcat(OpcodeStr,
4082 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4085 // There's an AssertZext in the way of writing the store pattern
4086 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4089 let Predicates = [HasAVX] in
4090 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4092 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4095 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4096 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4097 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4098 (ins VR128:$src1, i32i8imm:$src2),
4099 !strconcat(OpcodeStr,
4100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4102 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4103 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4104 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4105 !strconcat(OpcodeStr,
4106 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4107 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4108 addr:$dst)]>, OpSize;
4111 let Predicates = [HasAVX] in
4112 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4114 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4116 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4117 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4118 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4119 (ins VR128:$src1, i32i8imm:$src2),
4120 !strconcat(OpcodeStr,
4121 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4123 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4124 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4125 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4126 !strconcat(OpcodeStr,
4127 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4128 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4129 addr:$dst)]>, OpSize, REX_W;
4132 let Predicates = [HasAVX] in
4133 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4135 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4137 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4139 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4140 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4141 (ins VR128:$src1, i32i8imm:$src2),
4142 !strconcat(OpcodeStr,
4143 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4145 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4147 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4148 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4149 !strconcat(OpcodeStr,
4150 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4151 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4152 addr:$dst)]>, OpSize;
4155 let Predicates = [HasAVX] in {
4156 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4157 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4158 (ins VR128:$src1, i32i8imm:$src2),
4159 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4162 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4164 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4165 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4168 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4169 Requires<[HasSSE41]>;
4171 //===----------------------------------------------------------------------===//
4172 // SSE4.1 - Insert Instructions
4173 //===----------------------------------------------------------------------===//
4175 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4176 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4177 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4179 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4181 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4183 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4184 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4185 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4187 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4189 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4191 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4192 imm:$src3))]>, OpSize;
4195 let Predicates = [HasAVX] in
4196 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4197 let Constraints = "$src1 = $dst" in
4198 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4200 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4201 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4202 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4204 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4206 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4208 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4210 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4211 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4213 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4215 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4217 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4218 imm:$src3)))]>, OpSize;
4221 let Predicates = [HasAVX] in
4222 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4223 let Constraints = "$src1 = $dst" in
4224 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4226 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4227 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4228 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4230 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4232 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4234 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4236 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4237 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4239 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4241 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4243 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4244 imm:$src3)))]>, OpSize;
4247 let Predicates = [HasAVX] in
4248 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4249 let Constraints = "$src1 = $dst" in
4250 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4252 // insertps has a few different modes, there's the first two here below which
4253 // are optimized inserts that won't zero arbitrary elements in the destination
4254 // vector. The next one matches the intrinsic and could zero arbitrary elements
4255 // in the target vector.
4256 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4257 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4258 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4260 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4262 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4264 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4266 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4267 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4269 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4271 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4273 (X86insrtps VR128:$src1,
4274 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4275 imm:$src3))]>, OpSize;
4278 let Constraints = "$src1 = $dst" in
4279 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4280 let Predicates = [HasAVX] in
4281 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4283 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4284 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4286 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4287 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4288 Requires<[HasSSE41]>;
4290 //===----------------------------------------------------------------------===//
4291 // SSE4.1 - Round Instructions
4292 //===----------------------------------------------------------------------===//
4294 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4295 X86MemOperand x86memop, RegisterClass RC,
4296 PatFrag mem_frag32, PatFrag mem_frag64,
4297 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4298 // Intrinsic operation, reg.
4299 // Vector intrinsic operation, reg
4300 def PSr : SS4AIi8<opcps, MRMSrcReg,
4301 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4302 !strconcat(OpcodeStr,
4303 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4304 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4307 // Vector intrinsic operation, mem
4308 def PSm : Ii8<opcps, MRMSrcMem,
4309 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4310 !strconcat(OpcodeStr,
4311 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4313 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4315 Requires<[HasSSE41]>;
4317 // Vector intrinsic operation, reg
4318 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4319 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4320 !strconcat(OpcodeStr,
4321 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4322 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4325 // Vector intrinsic operation, mem
4326 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4327 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4328 !strconcat(OpcodeStr,
4329 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4331 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4335 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4336 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4337 // Intrinsic operation, reg.
4338 // Vector intrinsic operation, reg
4339 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4340 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4341 !strconcat(OpcodeStr,
4342 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4345 // Vector intrinsic operation, mem
4346 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4347 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4348 !strconcat(OpcodeStr,
4349 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4350 []>, TA, OpSize, Requires<[HasSSE41]>;
4352 // Vector intrinsic operation, reg
4353 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4354 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4355 !strconcat(OpcodeStr,
4356 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4359 // Vector intrinsic operation, mem
4360 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4361 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4362 !strconcat(OpcodeStr,
4363 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4367 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4370 Intrinsic F64Int, bit Is2Addr = 1> {
4371 // Intrinsic operation, reg.
4372 def SSr : SS4AIi8<opcss, MRMSrcReg,
4373 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4375 !strconcat(OpcodeStr,
4376 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4377 !strconcat(OpcodeStr,
4378 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4379 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4382 // Intrinsic operation, mem.
4383 def SSm : SS4AIi8<opcss, MRMSrcMem,
4384 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4386 !strconcat(OpcodeStr,
4387 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4388 !strconcat(OpcodeStr,
4389 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4391 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4394 // Intrinsic operation, reg.
4395 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4396 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4398 !strconcat(OpcodeStr,
4399 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4400 !strconcat(OpcodeStr,
4401 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4402 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4405 // Intrinsic operation, mem.
4406 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4407 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4409 !strconcat(OpcodeStr,
4410 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4411 !strconcat(OpcodeStr,
4412 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4414 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4418 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4420 // Intrinsic operation, reg.
4421 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4422 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4423 !strconcat(OpcodeStr,
4424 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4427 // Intrinsic operation, mem.
4428 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4429 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4430 !strconcat(OpcodeStr,
4431 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4434 // Intrinsic operation, reg.
4435 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4436 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4437 !strconcat(OpcodeStr,
4438 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4441 // Intrinsic operation, mem.
4442 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4443 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4444 !strconcat(OpcodeStr,
4445 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4449 // FP round - roundss, roundps, roundsd, roundpd
4450 let Predicates = [HasAVX] in {
4452 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4453 memopv4f32, memopv2f64,
4454 int_x86_sse41_round_ps,
4455 int_x86_sse41_round_pd>, VEX;
4456 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4457 memopv8f32, memopv4f64,
4458 int_x86_avx_round_ps_256,
4459 int_x86_avx_round_pd_256>, VEX;
4460 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4461 int_x86_sse41_round_ss,
4462 int_x86_sse41_round_sd, 0>, VEX_4V;
4464 // Instructions for the assembler
4465 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4467 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4469 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4472 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4473 memopv4f32, memopv2f64,
4474 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4475 let Constraints = "$src1 = $dst" in
4476 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4477 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4479 //===----------------------------------------------------------------------===//
4480 // SSE4.1 - Packed Bit Test
4481 //===----------------------------------------------------------------------===//
4483 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4484 // the intel intrinsic that corresponds to this.
4485 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4486 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4487 "vptest\t{$src2, $src1|$src1, $src2}",
4488 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4490 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4491 "vptest\t{$src2, $src1|$src1, $src2}",
4492 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4495 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4496 "vptest\t{$src2, $src1|$src1, $src2}",
4497 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4499 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4500 "vptest\t{$src2, $src1|$src1, $src2}",
4501 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4505 let Defs = [EFLAGS] in {
4506 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4507 "ptest \t{$src2, $src1|$src1, $src2}",
4508 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4510 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4511 "ptest \t{$src2, $src1|$src1, $src2}",
4512 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4516 // The bit test instructions below are AVX only
4517 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4518 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4519 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4520 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4521 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4522 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4523 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4524 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4528 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4529 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4530 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4531 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4532 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4535 //===----------------------------------------------------------------------===//
4536 // SSE4.1 - Misc Instructions
4537 //===----------------------------------------------------------------------===//
4539 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4540 "popcnt{w}\t{$src, $dst|$dst, $src}",
4541 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4542 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4543 "popcnt{w}\t{$src, $dst|$dst, $src}",
4544 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4546 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4547 "popcnt{l}\t{$src, $dst|$dst, $src}",
4548 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4549 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4550 "popcnt{l}\t{$src, $dst|$dst, $src}",
4551 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4553 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4554 "popcnt{q}\t{$src, $dst|$dst, $src}",
4555 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4556 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4557 "popcnt{q}\t{$src, $dst|$dst, $src}",
4558 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4562 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4563 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4564 Intrinsic IntId128> {
4565 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4567 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4568 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4569 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4571 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4574 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4577 let Predicates = [HasAVX] in
4578 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4579 int_x86_sse41_phminposuw>, VEX;
4580 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4581 int_x86_sse41_phminposuw>;
4583 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4584 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4585 Intrinsic IntId128, bit Is2Addr = 1> {
4586 let isCommutable = 1 in
4587 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4588 (ins VR128:$src1, VR128:$src2),
4590 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4591 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4592 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4593 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4594 (ins VR128:$src1, i128mem:$src2),
4596 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4597 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4599 (IntId128 VR128:$src1,
4600 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4603 let Predicates = [HasAVX] in {
4604 let isCommutable = 0 in
4605 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4607 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4609 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4611 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4613 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4615 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4617 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4619 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4621 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4623 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4625 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4629 let Constraints = "$src1 = $dst" in {
4630 let isCommutable = 0 in
4631 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4632 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4633 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4634 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4635 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4636 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4637 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4638 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4639 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4640 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4641 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4644 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4645 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4646 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4647 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4649 /// SS48I_binop_rm - Simple SSE41 binary operator.
4650 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4651 ValueType OpVT, bit Is2Addr = 1> {
4652 let isCommutable = 1 in
4653 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4654 (ins VR128:$src1, VR128:$src2),
4656 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4657 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4658 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4660 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4661 (ins VR128:$src1, i128mem:$src2),
4663 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4664 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4665 [(set VR128:$dst, (OpNode VR128:$src1,
4666 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4670 let Predicates = [HasAVX] in
4671 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4672 let Constraints = "$src1 = $dst" in
4673 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4675 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4676 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4677 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4678 X86MemOperand x86memop, bit Is2Addr = 1> {
4679 let isCommutable = 1 in
4680 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4681 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4683 !strconcat(OpcodeStr,
4684 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4685 !strconcat(OpcodeStr,
4686 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4687 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4689 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4690 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4692 !strconcat(OpcodeStr,
4693 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4694 !strconcat(OpcodeStr,
4695 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4698 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4702 let Predicates = [HasAVX] in {
4703 let isCommutable = 0 in {
4704 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4705 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4706 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4707 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4708 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4709 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4710 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4711 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4712 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4713 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4714 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4715 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4717 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4718 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4719 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4720 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4721 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4722 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4725 let Constraints = "$src1 = $dst" in {
4726 let isCommutable = 0 in {
4727 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4728 VR128, memopv16i8, i128mem>;
4729 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4730 VR128, memopv16i8, i128mem>;
4731 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4732 VR128, memopv16i8, i128mem>;
4733 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4734 VR128, memopv16i8, i128mem>;
4736 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4737 VR128, memopv16i8, i128mem>;
4738 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4739 VR128, memopv16i8, i128mem>;
4742 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4743 let Predicates = [HasAVX] in {
4744 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4745 RegisterClass RC, X86MemOperand x86memop,
4746 PatFrag mem_frag, Intrinsic IntId> {
4747 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4748 (ins RC:$src1, RC:$src2, RC:$src3),
4749 !strconcat(OpcodeStr,
4750 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4751 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4752 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4754 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4755 (ins RC:$src1, x86memop:$src2, RC:$src3),
4756 !strconcat(OpcodeStr,
4757 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4759 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4761 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4765 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4766 memopv16i8, int_x86_sse41_blendvpd>;
4767 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4768 memopv16i8, int_x86_sse41_blendvps>;
4769 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4770 memopv16i8, int_x86_sse41_pblendvb>;
4771 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4772 memopv32i8, int_x86_avx_blendv_pd_256>;
4773 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4774 memopv32i8, int_x86_avx_blendv_ps_256>;
4776 /// SS41I_ternary_int - SSE 4.1 ternary operator
4777 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4778 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4779 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4780 (ins VR128:$src1, VR128:$src2),
4781 !strconcat(OpcodeStr,
4782 "\t{$src2, $dst|$dst, $src2}"),
4783 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4786 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4787 (ins VR128:$src1, i128mem:$src2),
4788 !strconcat(OpcodeStr,
4789 "\t{$src2, $dst|$dst, $src2}"),
4792 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4796 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4797 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4798 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4800 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4801 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4803 let Predicates = [HasAVX] in
4804 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4805 "vmovntdqa\t{$src, $dst|$dst, $src}",
4806 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4808 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4809 "movntdqa\t{$src, $dst|$dst, $src}",
4810 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4813 //===----------------------------------------------------------------------===//
4814 // SSE4.2 - Compare Instructions
4815 //===----------------------------------------------------------------------===//
4817 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4818 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4819 Intrinsic IntId128, bit Is2Addr = 1> {
4820 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4821 (ins VR128:$src1, VR128:$src2),
4823 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4824 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4825 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4827 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4828 (ins VR128:$src1, i128mem:$src2),
4830 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4831 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4833 (IntId128 VR128:$src1,
4834 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4837 let Predicates = [HasAVX] in
4838 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4840 let Constraints = "$src1 = $dst" in
4841 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4843 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4844 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4845 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4846 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4848 //===----------------------------------------------------------------------===//
4849 // SSE4.2 - String/text Processing Instructions
4850 //===----------------------------------------------------------------------===//
4852 // Packed Compare Implicit Length Strings, Return Mask
4853 multiclass pseudo_pcmpistrm<string asm> {
4854 def REG : PseudoI<(outs VR128:$dst),
4855 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4856 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4858 def MEM : PseudoI<(outs VR128:$dst),
4859 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4860 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4861 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4864 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4865 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4866 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4869 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4870 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4871 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4872 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4873 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4874 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4875 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4878 let Defs = [XMM0, EFLAGS] in {
4879 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4880 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4881 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4882 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4883 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4884 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4887 // Packed Compare Explicit Length Strings, Return Mask
4888 multiclass pseudo_pcmpestrm<string asm> {
4889 def REG : PseudoI<(outs VR128:$dst),
4890 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4891 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4892 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4893 def MEM : PseudoI<(outs VR128:$dst),
4894 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4895 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4896 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4899 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4900 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4901 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4904 let Predicates = [HasAVX],
4905 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4906 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4907 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4908 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4909 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4910 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4911 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4914 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4915 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4916 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4917 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4918 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4919 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4920 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4923 // Packed Compare Implicit Length Strings, Return Index
4924 let Defs = [ECX, EFLAGS] in {
4925 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4926 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4927 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4928 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4929 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4930 (implicit EFLAGS)]>, OpSize;
4931 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4932 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4933 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4934 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4935 (implicit EFLAGS)]>, OpSize;
4939 let Predicates = [HasAVX] in {
4940 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4942 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4944 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4946 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4948 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4950 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4954 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4955 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4956 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4957 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4958 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4959 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4961 // Packed Compare Explicit Length Strings, Return Index
4962 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4963 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4964 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4965 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4966 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4967 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4968 (implicit EFLAGS)]>, OpSize;
4969 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4970 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4971 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4973 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4974 (implicit EFLAGS)]>, OpSize;
4978 let Predicates = [HasAVX] in {
4979 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4981 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
4983 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
4985 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
4987 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
4989 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
4993 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
4994 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
4995 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
4996 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
4997 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
4998 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5000 //===----------------------------------------------------------------------===//
5001 // SSE4.2 - CRC Instructions
5002 //===----------------------------------------------------------------------===//
5004 // No CRC instructions have AVX equivalents
5006 // crc intrinsic instruction
5007 // This set of instructions are only rm, the only difference is the size
5009 let Constraints = "$src1 = $dst" in {
5010 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5011 (ins GR32:$src1, i8mem:$src2),
5012 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5014 (int_x86_sse42_crc32_32_8 GR32:$src1,
5015 (load addr:$src2)))]>;
5016 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5017 (ins GR32:$src1, GR8:$src2),
5018 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5020 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5021 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5022 (ins GR32:$src1, i16mem:$src2),
5023 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5025 (int_x86_sse42_crc32_32_16 GR32:$src1,
5026 (load addr:$src2)))]>,
5028 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5029 (ins GR32:$src1, GR16:$src2),
5030 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5032 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5034 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5035 (ins GR32:$src1, i32mem:$src2),
5036 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5038 (int_x86_sse42_crc32_32_32 GR32:$src1,
5039 (load addr:$src2)))]>;
5040 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5041 (ins GR32:$src1, GR32:$src2),
5042 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5044 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5045 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5046 (ins GR64:$src1, i8mem:$src2),
5047 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5049 (int_x86_sse42_crc32_64_8 GR64:$src1,
5050 (load addr:$src2)))]>,
5052 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5053 (ins GR64:$src1, GR8:$src2),
5054 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5056 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5058 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5059 (ins GR64:$src1, i64mem:$src2),
5060 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5062 (int_x86_sse42_crc32_64_64 GR64:$src1,
5063 (load addr:$src2)))]>,
5065 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5066 (ins GR64:$src1, GR64:$src2),
5067 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5069 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5073 //===----------------------------------------------------------------------===//
5074 // AES-NI Instructions
5075 //===----------------------------------------------------------------------===//
5077 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5078 Intrinsic IntId128, bit Is2Addr = 1> {
5079 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5080 (ins VR128:$src1, VR128:$src2),
5082 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5083 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5084 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5086 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5087 (ins VR128:$src1, i128mem:$src2),
5089 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5092 (IntId128 VR128:$src1,
5093 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5096 // Perform One Round of an AES Encryption/Decryption Flow
5097 let Predicates = [HasAVX, HasAES] in {
5098 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5099 int_x86_aesni_aesenc, 0>, VEX_4V;
5100 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5101 int_x86_aesni_aesenclast, 0>, VEX_4V;
5102 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5103 int_x86_aesni_aesdec, 0>, VEX_4V;
5104 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5105 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5108 let Constraints = "$src1 = $dst" in {
5109 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5110 int_x86_aesni_aesenc>;
5111 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5112 int_x86_aesni_aesenclast>;
5113 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5114 int_x86_aesni_aesdec>;
5115 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5116 int_x86_aesni_aesdeclast>;
5119 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5120 (AESENCrr VR128:$src1, VR128:$src2)>;
5121 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5122 (AESENCrm VR128:$src1, addr:$src2)>;
5123 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5124 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5125 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5126 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5127 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5128 (AESDECrr VR128:$src1, VR128:$src2)>;
5129 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5130 (AESDECrm VR128:$src1, addr:$src2)>;
5131 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5132 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5133 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5134 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5136 // Perform the AES InvMixColumn Transformation
5137 let Predicates = [HasAVX, HasAES] in {
5138 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5140 "vaesimc\t{$src1, $dst|$dst, $src1}",
5142 (int_x86_aesni_aesimc VR128:$src1))]>,
5144 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5145 (ins i128mem:$src1),
5146 "vaesimc\t{$src1, $dst|$dst, $src1}",
5148 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5151 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5153 "aesimc\t{$src1, $dst|$dst, $src1}",
5155 (int_x86_aesni_aesimc VR128:$src1))]>,
5157 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5158 (ins i128mem:$src1),
5159 "aesimc\t{$src1, $dst|$dst, $src1}",
5161 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5164 // AES Round Key Generation Assist
5165 let Predicates = [HasAVX, HasAES] in {
5166 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5167 (ins VR128:$src1, i8imm:$src2),
5168 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5170 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5172 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5173 (ins i128mem:$src1, i8imm:$src2),
5174 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5176 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5180 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5181 (ins VR128:$src1, i8imm:$src2),
5182 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5184 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5186 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5187 (ins i128mem:$src1, i8imm:$src2),
5188 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5190 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5194 //===----------------------------------------------------------------------===//
5195 // CLMUL Instructions
5196 //===----------------------------------------------------------------------===//
5198 // Carry-less Multiplication instructions
5199 let Constraints = "$src1 = $dst" in {
5200 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5201 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5202 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5205 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5206 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5207 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5211 // AVX carry-less Multiplication instructions
5212 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5213 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5214 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5217 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5218 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5219 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5223 multiclass pclmul_alias<string asm, int immop> {
5224 def : InstAlias<!strconcat("pclmul", asm,
5225 "dq {$src, $dst|$dst, $src}"),
5226 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5228 def : InstAlias<!strconcat("pclmul", asm,
5229 "dq {$src, $dst|$dst, $src}"),
5230 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5232 def : InstAlias<!strconcat("vpclmul", asm,
5233 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5234 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5236 def : InstAlias<!strconcat("vpclmul", asm,
5237 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5238 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5240 defm : pclmul_alias<"hqhq", 0x11>;
5241 defm : pclmul_alias<"hqlq", 0x01>;
5242 defm : pclmul_alias<"lqhq", 0x10>;
5243 defm : pclmul_alias<"lqlq", 0x00>;
5245 //===----------------------------------------------------------------------===//
5247 //===----------------------------------------------------------------------===//
5250 // Load from memory and broadcast to all elements of the destination operand
5251 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5252 X86MemOperand x86memop, Intrinsic Int> :
5253 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5254 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5255 [(set RC:$dst, (Int addr:$src))]>, VEX;
5257 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5258 int_x86_avx_vbroadcastss>;
5259 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5260 int_x86_avx_vbroadcastss_256>;
5261 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5262 int_x86_avx_vbroadcast_sd_256>;
5263 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5264 int_x86_avx_vbroadcastf128_pd_256>;
5266 // Insert packed floating-point values
5267 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5268 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5269 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5271 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5272 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5273 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5276 // Extract packed floating-point values
5277 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5278 (ins VR256:$src1, i8imm:$src2),
5279 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5281 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5282 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5283 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5286 // Conditional SIMD Packed Loads and Stores
5287 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5288 Intrinsic IntLd, Intrinsic IntLd256,
5289 Intrinsic IntSt, Intrinsic IntSt256,
5290 PatFrag pf128, PatFrag pf256> {
5291 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5292 (ins VR128:$src1, f128mem:$src2),
5293 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5294 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5296 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5297 (ins VR256:$src1, f256mem:$src2),
5298 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5299 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5301 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5302 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5303 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5304 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5305 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5306 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5307 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5308 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5311 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5312 int_x86_avx_maskload_ps,
5313 int_x86_avx_maskload_ps_256,
5314 int_x86_avx_maskstore_ps,
5315 int_x86_avx_maskstore_ps_256,
5316 memopv4f32, memopv8f32>;
5317 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5318 int_x86_avx_maskload_pd,
5319 int_x86_avx_maskload_pd_256,
5320 int_x86_avx_maskstore_pd,
5321 int_x86_avx_maskstore_pd_256,
5322 memopv2f64, memopv4f64>;
5324 // Permute Floating-Point Values
5325 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5326 RegisterClass RC, X86MemOperand x86memop_f,
5327 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5328 Intrinsic IntVar, Intrinsic IntImm> {
5329 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5330 (ins RC:$src1, RC:$src2),
5331 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5332 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5333 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5334 (ins RC:$src1, x86memop_i:$src2),
5335 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5336 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5338 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5339 (ins RC:$src1, i8imm:$src2),
5340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5341 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5342 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5343 (ins x86memop_f:$src1, i8imm:$src2),
5344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5345 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5348 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5349 memopv4f32, memopv4i32,
5350 int_x86_avx_vpermilvar_ps,
5351 int_x86_avx_vpermil_ps>;
5352 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5353 memopv8f32, memopv8i32,
5354 int_x86_avx_vpermilvar_ps_256,
5355 int_x86_avx_vpermil_ps_256>;
5356 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5357 memopv2f64, memopv2i64,
5358 int_x86_avx_vpermilvar_pd,
5359 int_x86_avx_vpermil_pd>;
5360 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5361 memopv4f64, memopv4i64,
5362 int_x86_avx_vpermilvar_pd_256,
5363 int_x86_avx_vpermil_pd_256>;
5365 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5366 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5367 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5369 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5370 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5371 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5374 // Zero All YMM registers
5375 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5376 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5378 // Zero Upper bits of YMM registers
5379 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5380 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5382 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5383 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5384 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5385 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5386 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5387 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5389 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5391 (VINSERTF128rr VR256:$src1, VR128:$src2,
5392 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5393 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5395 (VINSERTF128rr VR256:$src1, VR128:$src2,
5396 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5397 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5399 (VINSERTF128rr VR256:$src1, VR128:$src2,
5400 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5401 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5403 (VINSERTF128rr VR256:$src1, VR128:$src2,
5404 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5406 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5407 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5408 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5409 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5410 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5411 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5413 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5414 (v4f32 (VEXTRACTF128rr
5415 (v8f32 VR256:$src1),
5416 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5417 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5418 (v2f64 (VEXTRACTF128rr
5419 (v4f64 VR256:$src1),
5420 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5421 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5422 (v4i32 (VEXTRACTF128rr
5423 (v8i32 VR256:$src1),
5424 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5425 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5426 (v2i64 (VEXTRACTF128rr
5427 (v4i64 VR256:$src1),
5428 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5430 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5431 (VBROADCASTF128 addr:$src)>;
5433 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5434 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5435 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5436 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5437 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5438 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5440 def : Pat<(int_x86_avx_vperm2f128_ps_256
5441 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5442 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5443 def : Pat<(int_x86_avx_vperm2f128_pd_256
5444 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5445 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5446 def : Pat<(int_x86_avx_vperm2f128_si_256
5447 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5448 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5450 //===----------------------------------------------------------------------===//
5451 // SSE Shuffle pattern fragments
5452 //===----------------------------------------------------------------------===//
5454 // This is part of a "work in progress" refactoring. The idea is that all
5455 // vector shuffles are going to be translated into target specific nodes and
5456 // directly matched by the patterns below (which can be changed along the way)
5457 // The AVX version of some but not all of them are described here, and more
5458 // should come in a near future.
5460 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5461 // SSE2 loads, which are always promoted to v2i64. The last one should match
5462 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5463 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5464 // we investigate further.
5465 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5467 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5468 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5470 (PSHUFDmi addr:$src1, imm:$imm)>;
5471 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5473 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5475 // Shuffle with PSHUFD instruction.
5476 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5477 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5478 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5479 (PSHUFDri VR128:$src1, imm:$imm)>;
5481 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5482 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5483 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5484 (PSHUFDri VR128:$src1, imm:$imm)>;
5486 // Shuffle with SHUFPD instruction.
5487 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5488 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5489 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5490 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5491 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5492 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5494 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5495 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5496 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5497 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5499 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5500 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5501 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5502 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5504 // Shuffle with SHUFPS instruction.
5505 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5506 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5507 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5508 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5509 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5510 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5512 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5513 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5514 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5515 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5517 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5518 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5519 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5520 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5521 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5522 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5524 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5525 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5526 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5527 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5529 // Shuffle with MOVHLPS instruction
5530 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5531 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5532 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5533 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5535 // Shuffle with MOVDDUP instruction
5536 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5537 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5538 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5539 (MOVDDUPrm addr:$src)>;
5541 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5542 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5543 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5544 (MOVDDUPrm addr:$src)>;
5546 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5547 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5548 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5549 (MOVDDUPrm addr:$src)>;
5551 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5552 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5553 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5554 (MOVDDUPrm addr:$src)>;
5556 def : Pat<(X86Movddup (bc_v2f64
5557 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5558 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5559 def : Pat<(X86Movddup (bc_v2f64
5560 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5561 (MOVDDUPrm addr:$src)>;
5564 // Shuffle with UNPCKLPS
5565 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5566 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5567 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5568 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5569 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5570 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5572 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5573 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5574 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5575 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5576 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5577 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5579 // Shuffle with UNPCKHPS
5580 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5581 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5582 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5583 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5585 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5586 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5587 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5588 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5590 // Shuffle with UNPCKLPD
5591 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5592 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5593 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5594 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5595 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5596 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5598 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5599 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5600 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5601 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5602 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5603 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5605 // Shuffle with UNPCKHPD
5606 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5607 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5608 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5609 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5611 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5612 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5613 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5614 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5616 // Shuffle with PUNPCKLBW
5617 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5618 (bc_v16i8 (memopv2i64 addr:$src2)))),
5619 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5620 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5621 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5623 // Shuffle with PUNPCKLWD
5624 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5625 (bc_v8i16 (memopv2i64 addr:$src2)))),
5626 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5627 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5628 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5630 // Shuffle with PUNPCKLDQ
5631 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5632 (bc_v4i32 (memopv2i64 addr:$src2)))),
5633 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5634 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5635 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5637 // Shuffle with PUNPCKLQDQ
5638 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5639 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5640 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5641 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5643 // Shuffle with PUNPCKHBW
5644 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5645 (bc_v16i8 (memopv2i64 addr:$src2)))),
5646 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5647 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5648 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5650 // Shuffle with PUNPCKHWD
5651 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5652 (bc_v8i16 (memopv2i64 addr:$src2)))),
5653 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5654 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5655 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5657 // Shuffle with PUNPCKHDQ
5658 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5659 (bc_v4i32 (memopv2i64 addr:$src2)))),
5660 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5661 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5662 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5664 // Shuffle with PUNPCKHQDQ
5665 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5666 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5667 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5668 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5670 // Shuffle with MOVLHPS
5671 def : Pat<(X86Movlhps VR128:$src1,
5672 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5673 (MOVHPSrm VR128:$src1, addr:$src2)>;
5674 def : Pat<(X86Movlhps VR128:$src1,
5675 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5676 (MOVHPSrm VR128:$src1, addr:$src2)>;
5677 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5678 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5679 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5680 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5681 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5682 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5684 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5685 // is during lowering, where it's not possible to recognize the load fold cause
5686 // it has two uses through a bitcast. One use disappears at isel time and the
5687 // fold opportunity reappears.
5688 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5689 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5691 // Shuffle with MOVLHPD
5692 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5693 (scalar_to_vector (loadf64 addr:$src2)))),
5694 (MOVHPDrm VR128:$src1, addr:$src2)>;
5696 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5697 // is during lowering, where it's not possible to recognize the load fold cause
5698 // it has two uses through a bitcast. One use disappears at isel time and the
5699 // fold opportunity reappears.
5700 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5701 (scalar_to_vector (loadf64 addr:$src2)))),
5702 (MOVHPDrm VR128:$src1, addr:$src2)>;
5704 // Shuffle with MOVSS
5705 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5706 (MOVSSrr VR128:$src1, FR32:$src2)>;
5707 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5708 (MOVSSrr (v4i32 VR128:$src1),
5709 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5710 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5711 (MOVSSrr (v4f32 VR128:$src1),
5712 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5713 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5714 // is during lowering, where it's not possible to recognize the load fold cause
5715 // it has two uses through a bitcast. One use disappears at isel time and the
5716 // fold opportunity reappears.
5717 def : Pat<(X86Movss VR128:$src1,
5718 (bc_v4i32 (v2i64 (load addr:$src2)))),
5719 (MOVLPSrm VR128:$src1, addr:$src2)>;
5721 // Shuffle with MOVSD
5722 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5723 (MOVSDrr VR128:$src1, FR64:$src2)>;
5724 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5725 (MOVSDrr (v2i64 VR128:$src1),
5726 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5727 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5728 (MOVSDrr (v2f64 VR128:$src1),
5729 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5730 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5731 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5732 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5733 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5735 // Shuffle with MOVSHDUP
5736 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5737 (MOVSHDUPrr VR128:$src)>;
5738 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5739 (MOVSHDUPrm addr:$src)>;
5741 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5742 (MOVSHDUPrr VR128:$src)>;
5743 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5744 (MOVSHDUPrm addr:$src)>;
5746 // Shuffle with MOVSLDUP
5747 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5748 (MOVSLDUPrr VR128:$src)>;
5749 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5750 (MOVSLDUPrm addr:$src)>;
5752 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5753 (MOVSLDUPrr VR128:$src)>;
5754 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5755 (MOVSLDUPrm addr:$src)>;
5757 // Shuffle with PSHUFHW
5758 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5759 (PSHUFHWri VR128:$src, imm:$imm)>;
5760 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5761 (PSHUFHWmi addr:$src, imm:$imm)>;
5763 // Shuffle with PSHUFLW
5764 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5765 (PSHUFLWri VR128:$src, imm:$imm)>;
5766 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5767 (PSHUFLWmi addr:$src, imm:$imm)>;
5769 // Shuffle with PALIGN
5770 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5771 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5772 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5773 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5774 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5775 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5776 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5777 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5779 // Shuffle with MOVLPS
5780 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5781 (MOVLPSrm VR128:$src1, addr:$src2)>;
5782 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5783 (MOVLPSrm VR128:$src1, addr:$src2)>;
5784 def : Pat<(X86Movlps VR128:$src1,
5785 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5786 (MOVLPSrm VR128:$src1, addr:$src2)>;
5787 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5788 // is during lowering, where it's not possible to recognize the load fold cause
5789 // it has two uses through a bitcast. One use disappears at isel time and the
5790 // fold opportunity reappears.
5791 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5792 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5794 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5795 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5797 // Shuffle with MOVLPD
5798 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5799 (MOVLPDrm VR128:$src1, addr:$src2)>;
5800 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5801 (MOVLPDrm VR128:$src1, addr:$src2)>;
5802 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5803 (scalar_to_vector (loadf64 addr:$src2)))),
5804 (MOVLPDrm VR128:$src1, addr:$src2)>;
5806 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5807 def : Pat<(store (f64 (vector_extract
5808 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5809 (MOVHPSmr addr:$dst, VR128:$src)>;
5810 def : Pat<(store (f64 (vector_extract
5811 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5812 (MOVHPDmr addr:$dst, VR128:$src)>;
5814 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5815 (MOVLPSmr addr:$src1, VR128:$src2)>;
5816 def : Pat<(store (v4i32 (X86Movlps
5817 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5818 (MOVLPSmr addr:$src1, VR128:$src2)>;
5820 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5821 (MOVLPDmr addr:$src1, VR128:$src2)>;
5822 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5823 (MOVLPDmr addr:$src1, VR128:$src2)>;