1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // SSE 1 & 2 - Move Instructions
120 //===----------------------------------------------------------------------===//
122 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
123 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
124 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
126 // Loading from memory automatically zeroing upper bits.
127 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
128 PatFrag mem_pat, string OpcodeStr> :
129 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
131 [(set RC:$dst, (mem_pat addr:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
139 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
140 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
141 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
143 let canFoldAsLoad = 1, isReMaterializable = 1 in {
144 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
146 let AddedComplexity = 20 in
147 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
150 let Constraints = "$src1 = $dst" in {
151 def MOVSSrr : sse12_move_rr<FR32, v4f32,
152 "movss\t{$src2, $dst|$dst, $src2}">, XS;
153 def MOVSDrr : sse12_move_rr<FR64, v2f64,
154 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
157 let canFoldAsLoad = 1, isReMaterializable = 1 in {
158 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
160 let AddedComplexity = 20 in
161 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
164 let AddedComplexity = 15 in {
165 // Extract the low 32-bit value from one vector and insert it into another.
166 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
167 (MOVSSrr (v4f32 VR128:$src1),
168 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
169 // Extract the low 64-bit value from one vector and insert it into another.
170 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
171 (MOVSDrr (v2f64 VR128:$src1),
172 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
175 // Implicitly promote a 32-bit scalar to a vector.
176 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
177 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
178 // Implicitly promote a 64-bit scalar to a vector.
179 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
180 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
181 // Implicitly promote a 32-bit scalar to a vector.
182 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
183 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
184 // Implicitly promote a 64-bit scalar to a vector.
185 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
186 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
188 let AddedComplexity = 20 in {
189 // MOVSSrm zeros the high parts of the register; represent this
190 // with SUBREG_TO_REG.
191 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
192 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
193 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 // MOVSDrm zeros the high parts of the register; represent this
198 // with SUBREG_TO_REG.
199 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
200 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
201 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzload addr:$src)),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
211 // Store scalar value to memory.
212 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
213 "movss\t{$src, $dst|$dst, $src}",
214 [(store FR32:$src, addr:$dst)]>;
215 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
216 "movsd\t{$src, $dst|$dst, $src}",
217 [(store FR64:$src, addr:$dst)]>;
219 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
220 "movss\t{$src, $dst|$dst, $src}",
221 [(store FR32:$src, addr:$dst)]>, XS, VEX;
222 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
223 "movsd\t{$src, $dst|$dst, $src}",
224 [(store FR64:$src, addr:$dst)]>, XD, VEX;
226 // Extract and store.
227 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
230 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
231 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
236 // Move Aligned/Unaligned floating point values
237 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
238 X86MemOperand x86memop, PatFrag ld_frag,
239 string asm, Domain d,
240 bit IsReMaterializable = 1> {
241 let neverHasSideEffects = 1 in
242 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
243 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
244 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
245 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
246 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
247 [(set RC:$dst, (ld_frag addr:$src))], d>;
250 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
251 "movaps", SSEPackedSingle>, VEX;
252 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
253 "movapd", SSEPackedDouble>, OpSize, VEX;
254 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
255 "movups", SSEPackedSingle>, VEX;
256 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
257 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
259 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
260 "movaps", SSEPackedSingle>, VEX;
261 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
262 "movapd", SSEPackedDouble>, OpSize, VEX;
263 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
264 "movups", SSEPackedSingle>, VEX;
265 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
266 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
267 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
268 "movaps", SSEPackedSingle>, TB;
269 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
270 "movapd", SSEPackedDouble>, TB, OpSize;
271 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
272 "movups", SSEPackedSingle>, TB;
273 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
274 "movupd", SSEPackedDouble, 0>, TB, OpSize;
276 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
277 "movaps\t{$src, $dst|$dst, $src}",
278 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
279 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
280 "movapd\t{$src, $dst|$dst, $src}",
281 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
282 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
283 "movups\t{$src, $dst|$dst, $src}",
284 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
285 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
286 "movupd\t{$src, $dst|$dst, $src}",
287 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
288 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
289 "movaps\t{$src, $dst|$dst, $src}",
290 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
291 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
292 "movapd\t{$src, $dst|$dst, $src}",
293 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
294 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
295 "movups\t{$src, $dst|$dst, $src}",
296 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
297 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
298 "movupd\t{$src, $dst|$dst, $src}",
299 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
302 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
303 (VMOVUPSYmr addr:$dst, VR256:$src)>;
305 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
306 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
307 (VMOVUPDYmr addr:$dst, VR256:$src)>;
309 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
310 "movaps\t{$src, $dst|$dst, $src}",
311 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
312 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
313 "movapd\t{$src, $dst|$dst, $src}",
314 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
315 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
316 "movups\t{$src, $dst|$dst, $src}",
317 [(store (v4f32 VR128:$src), addr:$dst)]>;
318 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movupd\t{$src, $dst|$dst, $src}",
320 [(store (v2f64 VR128:$src), addr:$dst)]>;
322 // Intrinsic forms of MOVUPS/D load and store
323 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
324 (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
327 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
328 (ins f128mem:$dst, VR128:$src),
329 "movupd\t{$src, $dst|$dst, $src}",
330 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
332 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
333 "movups\t{$src, $dst|$dst, $src}",
334 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
335 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
336 "movupd\t{$src, $dst|$dst, $src}",
337 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
339 // Move Low/High packed floating point values
340 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
341 PatFrag mov_frag, string base_opc,
343 def PSrm : PI<opc, MRMSrcMem,
344 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
345 !strconcat(base_opc, "s", asm_opr),
348 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
349 SSEPackedSingle>, TB;
351 def PDrm : PI<opc, MRMSrcMem,
352 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
353 !strconcat(base_opc, "d", asm_opr),
354 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
355 (scalar_to_vector (loadf64 addr:$src2)))))],
356 SSEPackedDouble>, TB, OpSize;
359 let AddedComplexity = 20 in {
360 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
361 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
362 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
363 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
365 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
366 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
367 "\t{$src2, $dst|$dst, $src2}">;
368 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
369 "\t{$src2, $dst|$dst, $src2}">;
372 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
373 "movlps\t{$src, $dst|$dst, $src}",
374 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
375 (iPTR 0))), addr:$dst)]>, VEX;
376 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
377 "movlpd\t{$src, $dst|$dst, $src}",
378 [(store (f64 (vector_extract (v2f64 VR128:$src),
379 (iPTR 0))), addr:$dst)]>, VEX;
380 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
381 "movlps\t{$src, $dst|$dst, $src}",
382 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
383 (iPTR 0))), addr:$dst)]>;
384 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
385 "movlpd\t{$src, $dst|$dst, $src}",
386 [(store (f64 (vector_extract (v2f64 VR128:$src),
387 (iPTR 0))), addr:$dst)]>;
389 // v2f64 extract element 1 is always custom lowered to unpack high to low
390 // and extract element 0 so the non-store version isn't too horrible.
391 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
392 "movhps\t{$src, $dst|$dst, $src}",
393 [(store (f64 (vector_extract
394 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
395 (undef)), (iPTR 0))), addr:$dst)]>,
397 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
398 "movhpd\t{$src, $dst|$dst, $src}",
399 [(store (f64 (vector_extract
400 (v2f64 (unpckh VR128:$src, (undef))),
401 (iPTR 0))), addr:$dst)]>,
403 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movhps\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract
406 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
407 (undef)), (iPTR 0))), addr:$dst)]>;
408 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movhpd\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract
411 (v2f64 (unpckh VR128:$src, (undef))),
412 (iPTR 0))), addr:$dst)]>;
414 let AddedComplexity = 20 in {
415 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
416 (ins VR128:$src1, VR128:$src2),
417 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
419 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
421 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
422 (ins VR128:$src1, VR128:$src2),
423 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
425 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
428 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
429 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
430 (ins VR128:$src1, VR128:$src2),
431 "movlhps\t{$src2, $dst|$dst, $src2}",
433 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
434 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
435 (ins VR128:$src1, VR128:$src2),
436 "movhlps\t{$src2, $dst|$dst, $src2}",
438 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
441 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
442 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
443 let AddedComplexity = 20 in {
444 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
445 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
446 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
447 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
450 //===----------------------------------------------------------------------===//
451 // SSE 1 & 2 - Conversion Instructions
452 //===----------------------------------------------------------------------===//
454 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
455 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
457 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
458 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
460 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
463 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
464 X86MemOperand x86memop, string asm> {
465 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
467 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
471 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
472 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
473 string asm, Domain d> {
474 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
475 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
476 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
477 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
480 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
481 X86MemOperand x86memop, string asm> {
482 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
483 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
484 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
485 (ins DstRC:$src1, x86memop:$src),
486 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
489 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
490 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
491 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
492 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
494 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
495 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
496 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
497 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
500 // The assembler can recognize rr 64-bit instructions by seeing a rxx
501 // register, but the same isn't true when only using memory operands,
502 // provide other assembly "l" and "q" forms to address this explicitly
503 // where appropriate to do so.
504 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
506 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
508 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
510 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
512 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
515 let Predicates = [HasAVX] in {
516 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
517 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
518 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
519 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
520 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
521 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
522 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
523 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
525 def : Pat<(f32 (sint_to_fp GR32:$src)),
526 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
527 def : Pat<(f32 (sint_to_fp GR64:$src)),
528 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
529 def : Pat<(f64 (sint_to_fp GR32:$src)),
530 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
531 def : Pat<(f64 (sint_to_fp GR64:$src)),
532 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
535 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
536 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
537 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
538 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
539 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
540 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
541 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
542 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
543 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
544 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
545 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
546 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
547 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
548 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
549 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
550 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
552 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
553 // and/or XMM operand(s).
555 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
556 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
558 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
559 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
560 [(set DstRC:$dst, (Int SrcRC:$src))]>;
561 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
562 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
563 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
566 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
567 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
568 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
569 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
571 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
572 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
573 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
574 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
575 (ins DstRC:$src1, x86memop:$src2),
577 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
578 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
579 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
582 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
583 f32mem, load, "cvtss2si">, XS, VEX;
584 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
585 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
587 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
588 f128mem, load, "cvtsd2si">, XD, VEX;
589 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
590 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
593 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
594 // Get rid of this hack or rename the intrinsics, there are several
595 // intructions that only match with the intrinsic form, why create duplicates
596 // to let them be recognized by the assembler?
597 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
598 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
599 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
600 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
601 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
602 f32mem, load, "cvtss2si">, XS;
603 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
604 f32mem, load, "cvtss2si{q}">, XS, REX_W;
605 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
606 f128mem, load, "cvtsd2si{l}">, XD;
607 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
608 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
611 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
612 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
613 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
614 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
616 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
617 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
618 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
619 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
622 let Constraints = "$src1 = $dst" in {
623 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
624 int_x86_sse_cvtsi2ss, i32mem, loadi32,
626 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
627 int_x86_sse_cvtsi642ss, i64mem, loadi64,
628 "cvtsi2ss{q}">, XS, REX_W;
629 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
630 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
632 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
633 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
634 "cvtsi2sd">, XD, REX_W;
639 // Aliases for intrinsics
640 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
641 f32mem, load, "cvttss2si">, XS, VEX;
642 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
643 int_x86_sse_cvttss2si64, f32mem, load,
644 "cvttss2si">, XS, VEX, VEX_W;
645 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
646 f128mem, load, "cvttsd2si">, XD, VEX;
647 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
648 int_x86_sse2_cvttsd2si64, f128mem, load,
649 "cvttsd2si">, XD, VEX, VEX_W;
650 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
651 f32mem, load, "cvttss2si">, XS;
652 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
653 int_x86_sse_cvttss2si64, f32mem, load,
654 "cvttss2si{q}">, XS, REX_W;
655 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
656 f128mem, load, "cvttsd2si">, XD;
657 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
658 int_x86_sse2_cvttsd2si64, f128mem, load,
659 "cvttsd2si{q}">, XD, REX_W;
661 let Pattern = []<dag> in {
662 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
663 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
664 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
665 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
667 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
668 "cvtdq2ps\t{$src, $dst|$dst, $src}",
669 SSEPackedSingle>, TB, VEX;
670 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
671 "cvtdq2ps\t{$src, $dst|$dst, $src}",
672 SSEPackedSingle>, TB, VEX;
674 let Pattern = []<dag> in {
675 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
676 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
677 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
678 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
679 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
680 "cvtdq2ps\t{$src, $dst|$dst, $src}",
681 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
686 // Convert scalar double to scalar single
687 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
688 (ins FR64:$src1, FR64:$src2),
689 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
691 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
692 (ins FR64:$src1, f64mem:$src2),
693 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
694 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
695 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
698 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
699 "cvtsd2ss\t{$src, $dst|$dst, $src}",
700 [(set FR32:$dst, (fround FR64:$src))]>;
701 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
702 "cvtsd2ss\t{$src, $dst|$dst, $src}",
703 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
704 Requires<[HasSSE2, OptForSize]>;
706 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
707 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
709 let Constraints = "$src1 = $dst" in
710 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
711 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
713 // Convert scalar single to scalar double
714 // SSE2 instructions with XS prefix
715 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
716 (ins FR32:$src1, FR32:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 []>, XS, Requires<[HasAVX]>, VEX_4V;
719 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
720 (ins FR32:$src1, f32mem:$src2),
721 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
722 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
723 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
726 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
727 "cvtss2sd\t{$src, $dst|$dst, $src}",
728 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
730 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
731 "cvtss2sd\t{$src, $dst|$dst, $src}",
732 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
733 Requires<[HasSSE2, OptForSize]>;
735 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
736 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
737 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
738 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
739 VR128:$src2))]>, XS, VEX_4V,
741 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
742 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
743 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
744 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
745 (load addr:$src2)))]>, XS, VEX_4V,
747 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
748 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
749 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
750 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
751 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
754 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
755 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
756 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
757 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
758 (load addr:$src2)))]>, XS,
762 def : Pat<(extloadf32 addr:$src),
763 (CVTSS2SDrr (MOVSSrm addr:$src))>,
764 Requires<[HasSSE2, OptForSpeed]>;
766 // Convert doubleword to packed single/double fp
767 // SSE2 instructions without OpSize prefix
768 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
769 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
770 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
771 TB, VEX, Requires<[HasAVX]>;
772 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
773 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
774 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
775 (bitconvert (memopv2i64 addr:$src))))]>,
776 TB, VEX, Requires<[HasAVX]>;
777 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
778 "cvtdq2ps\t{$src, $dst|$dst, $src}",
779 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
780 TB, Requires<[HasSSE2]>;
781 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
782 "cvtdq2ps\t{$src, $dst|$dst, $src}",
783 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
784 (bitconvert (memopv2i64 addr:$src))))]>,
785 TB, Requires<[HasSSE2]>;
787 // FIXME: why the non-intrinsic version is described as SSE3?
788 // SSE2 instructions with XS prefix
789 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
790 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
791 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
792 XS, VEX, Requires<[HasAVX]>;
793 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
794 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
795 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
796 (bitconvert (memopv2i64 addr:$src))))]>,
797 XS, VEX, Requires<[HasAVX]>;
798 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
799 "cvtdq2pd\t{$src, $dst|$dst, $src}",
800 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
801 XS, Requires<[HasSSE2]>;
802 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
803 "cvtdq2pd\t{$src, $dst|$dst, $src}",
804 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
805 (bitconvert (memopv2i64 addr:$src))))]>,
806 XS, Requires<[HasSSE2]>;
809 // Convert packed single/double fp to doubleword
810 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
811 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
812 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
813 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
814 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
815 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
816 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
817 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
818 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
819 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
820 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
821 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
823 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
824 "cvtps2dq\t{$src, $dst|$dst, $src}",
825 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
827 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
829 "cvtps2dq\t{$src, $dst|$dst, $src}",
830 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
831 (memop addr:$src)))]>, VEX;
832 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
833 "cvtps2dq\t{$src, $dst|$dst, $src}",
834 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
835 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
836 "cvtps2dq\t{$src, $dst|$dst, $src}",
837 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
838 (memop addr:$src)))]>;
840 // SSE2 packed instructions with XD prefix
841 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
842 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
843 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
844 XD, VEX, Requires<[HasAVX]>;
845 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
846 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
847 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
848 (memop addr:$src)))]>,
849 XD, VEX, Requires<[HasAVX]>;
850 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
851 "cvtpd2dq\t{$src, $dst|$dst, $src}",
852 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
853 XD, Requires<[HasSSE2]>;
854 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
855 "cvtpd2dq\t{$src, $dst|$dst, $src}",
856 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
857 (memop addr:$src)))]>,
858 XD, Requires<[HasSSE2]>;
861 // Convert with truncation packed single/double fp to doubleword
862 // SSE2 packed instructions with XS prefix
863 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
864 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
865 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
866 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
867 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
868 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
869 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
870 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
871 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
872 "cvttps2dq\t{$src, $dst|$dst, $src}",
874 (int_x86_sse2_cvttps2dq VR128:$src))]>;
875 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
876 "cvttps2dq\t{$src, $dst|$dst, $src}",
878 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
881 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
882 "vcvttps2dq\t{$src, $dst|$dst, $src}",
884 (int_x86_sse2_cvttps2dq VR128:$src))]>,
885 XS, VEX, Requires<[HasAVX]>;
886 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
887 "vcvttps2dq\t{$src, $dst|$dst, $src}",
888 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
889 (memop addr:$src)))]>,
890 XS, VEX, Requires<[HasAVX]>;
892 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
894 "cvttpd2dq\t{$src, $dst|$dst, $src}",
895 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
897 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
899 "cvttpd2dq\t{$src, $dst|$dst, $src}",
900 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
901 (memop addr:$src)))]>, VEX;
902 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
903 "cvttpd2dq\t{$src, $dst|$dst, $src}",
904 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
905 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
906 "cvttpd2dq\t{$src, $dst|$dst, $src}",
907 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
908 (memop addr:$src)))]>;
910 // The assembler can recognize rr 256-bit instructions by seeing a ymm
911 // register, but the same isn't true when using memory operands instead.
912 // Provide other assembly rr and rm forms to address this explicitly.
913 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
914 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
915 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
916 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
919 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
920 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
921 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
922 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
925 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
926 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
927 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
928 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
930 // Convert packed single to packed double
931 let Predicates = [HasAVX] in {
932 // SSE2 instructions without OpSize prefix
933 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
934 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
935 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
936 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
937 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
938 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
939 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
940 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
942 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
943 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
944 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
945 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
947 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
948 "vcvtps2pd\t{$src, $dst|$dst, $src}",
949 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
950 VEX, Requires<[HasAVX]>;
951 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
952 "vcvtps2pd\t{$src, $dst|$dst, $src}",
953 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
954 (load addr:$src)))]>,
955 VEX, Requires<[HasAVX]>;
956 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvtps2pd\t{$src, $dst|$dst, $src}",
958 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
959 TB, Requires<[HasSSE2]>;
960 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
961 "cvtps2pd\t{$src, $dst|$dst, $src}",
962 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
963 (load addr:$src)))]>,
964 TB, Requires<[HasSSE2]>;
966 // Convert packed double to packed single
967 // The assembler can recognize rr 256-bit instructions by seeing a ymm
968 // register, but the same isn't true when using memory operands instead.
969 // Provide other assembly rr and rm forms to address this explicitly.
970 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
971 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
972 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
973 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
976 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
977 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
978 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
979 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
982 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
983 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
984 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
985 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
986 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
987 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
988 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
989 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
992 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
993 "cvtpd2ps\t{$src, $dst|$dst, $src}",
994 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
995 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
997 "cvtpd2ps\t{$src, $dst|$dst, $src}",
998 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
999 (memop addr:$src)))]>;
1000 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1001 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1002 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1003 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1004 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1005 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1006 (memop addr:$src)))]>;
1008 // AVX 256-bit register conversion intrinsics
1009 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1010 // whenever possible to avoid declaring two versions of each one.
1011 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1012 (VCVTDQ2PSYrr VR256:$src)>;
1013 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1014 (VCVTDQ2PSYrm addr:$src)>;
1016 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1017 (VCVTPD2PSYrr VR256:$src)>;
1018 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1019 (VCVTPD2PSYrm addr:$src)>;
1021 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1022 (VCVTPS2DQYrr VR256:$src)>;
1023 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1024 (VCVTPS2DQYrm addr:$src)>;
1026 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1027 (VCVTPS2PDYrr VR128:$src)>;
1028 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1029 (VCVTPS2PDYrm addr:$src)>;
1031 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1032 (VCVTTPD2DQYrr VR256:$src)>;
1033 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1034 (VCVTTPD2DQYrm addr:$src)>;
1036 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1037 (VCVTTPS2DQYrr VR256:$src)>;
1038 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1039 (VCVTTPS2DQYrm addr:$src)>;
1041 //===----------------------------------------------------------------------===//
1042 // SSE 1 & 2 - Compare Instructions
1043 //===----------------------------------------------------------------------===//
1045 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1046 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1047 string asm, string asm_alt> {
1048 let isAsmParserOnly = 1 in {
1049 def rr : SIi8<0xC2, MRMSrcReg,
1050 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1053 def rm : SIi8<0xC2, MRMSrcMem,
1054 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1058 // Accept explicit immediate argument form instead of comparison code.
1059 def rr_alt : SIi8<0xC2, MRMSrcReg,
1060 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1063 def rm_alt : SIi8<0xC2, MRMSrcMem,
1064 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1068 let neverHasSideEffects = 1 in {
1069 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1070 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1071 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1073 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1074 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1075 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1079 let Constraints = "$src1 = $dst" in {
1080 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1081 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1082 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1083 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1084 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1085 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1086 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1087 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1088 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1089 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1090 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1091 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1092 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1093 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1094 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1095 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1097 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1098 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1099 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1100 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1101 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1102 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1103 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1104 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1105 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1106 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1107 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1108 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1109 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1112 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1113 Intrinsic Int, string asm> {
1114 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1115 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1116 [(set VR128:$dst, (Int VR128:$src1,
1117 VR128:$src, imm:$cc))]>;
1118 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1119 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1120 [(set VR128:$dst, (Int VR128:$src1,
1121 (load addr:$src), imm:$cc))]>;
1124 // Aliases to match intrinsics which expect XMM operand(s).
1125 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1126 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1128 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1129 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1131 let Constraints = "$src1 = $dst" in {
1132 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1133 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1134 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1135 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1139 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1140 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1141 ValueType vt, X86MemOperand x86memop,
1142 PatFrag ld_frag, string OpcodeStr, Domain d> {
1143 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1144 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1145 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1146 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1147 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1148 [(set EFLAGS, (OpNode (vt RC:$src1),
1149 (ld_frag addr:$src2)))], d>;
1152 let Defs = [EFLAGS] in {
1153 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1154 "ucomiss", SSEPackedSingle>, VEX;
1155 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1156 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1157 let Pattern = []<dag> in {
1158 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1159 "comiss", SSEPackedSingle>, VEX;
1160 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1161 "comisd", SSEPackedDouble>, OpSize, VEX;
1164 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1165 load, "ucomiss", SSEPackedSingle>, VEX;
1166 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1167 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1169 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1170 load, "comiss", SSEPackedSingle>, VEX;
1171 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1172 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1173 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1174 "ucomiss", SSEPackedSingle>, TB;
1175 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1176 "ucomisd", SSEPackedDouble>, TB, OpSize;
1178 let Pattern = []<dag> in {
1179 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1180 "comiss", SSEPackedSingle>, TB;
1181 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1182 "comisd", SSEPackedDouble>, TB, OpSize;
1185 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1186 load, "ucomiss", SSEPackedSingle>, TB;
1187 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1188 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1190 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1191 "comiss", SSEPackedSingle>, TB;
1192 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1193 "comisd", SSEPackedDouble>, TB, OpSize;
1194 } // Defs = [EFLAGS]
1196 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1197 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1198 Intrinsic Int, string asm, string asm_alt,
1200 let isAsmParserOnly = 1 in {
1201 def rri : PIi8<0xC2, MRMSrcReg,
1202 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1203 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1204 def rmi : PIi8<0xC2, MRMSrcMem,
1205 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1206 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1209 // Accept explicit immediate argument form instead of comparison code.
1210 def rri_alt : PIi8<0xC2, MRMSrcReg,
1211 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1213 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1214 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1218 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1219 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1220 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1221 SSEPackedSingle>, VEX_4V;
1222 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1223 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1224 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1225 SSEPackedDouble>, OpSize, VEX_4V;
1226 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1227 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1228 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1229 SSEPackedSingle>, VEX_4V;
1230 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1231 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1232 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1233 SSEPackedDouble>, OpSize, VEX_4V;
1234 let Constraints = "$src1 = $dst" in {
1235 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1236 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1237 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1238 SSEPackedSingle>, TB;
1239 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1240 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1241 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1242 SSEPackedDouble>, TB, OpSize;
1245 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1246 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1247 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1248 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1249 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1250 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1251 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1252 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1254 //===----------------------------------------------------------------------===//
1255 // SSE 1 & 2 - Shuffle Instructions
1256 //===----------------------------------------------------------------------===//
1258 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1259 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1260 ValueType vt, string asm, PatFrag mem_frag,
1261 Domain d, bit IsConvertibleToThreeAddress = 0> {
1262 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1263 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1264 [(set RC:$dst, (vt (shufp:$src3
1265 RC:$src1, (mem_frag addr:$src2))))], d>;
1266 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1267 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1268 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1270 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1273 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1274 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1275 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1276 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1277 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1278 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1279 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1280 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1281 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1282 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1283 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1284 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1286 let Constraints = "$src1 = $dst" in {
1287 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1288 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1289 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1291 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1292 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1293 memopv2f64, SSEPackedDouble>, TB, OpSize;
1296 //===----------------------------------------------------------------------===//
1297 // SSE 1 & 2 - Unpack Instructions
1298 //===----------------------------------------------------------------------===//
1300 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1301 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1302 PatFrag mem_frag, RegisterClass RC,
1303 X86MemOperand x86memop, string asm,
1305 def rr : PI<opc, MRMSrcReg,
1306 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1308 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1309 def rm : PI<opc, MRMSrcMem,
1310 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1312 (vt (OpNode RC:$src1,
1313 (mem_frag addr:$src2))))], d>;
1316 let AddedComplexity = 10 in {
1317 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1318 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1319 SSEPackedSingle>, VEX_4V;
1320 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1321 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1322 SSEPackedDouble>, OpSize, VEX_4V;
1323 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1324 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1325 SSEPackedSingle>, VEX_4V;
1326 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1327 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1328 SSEPackedDouble>, OpSize, VEX_4V;
1330 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1331 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1332 SSEPackedSingle>, VEX_4V;
1333 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1334 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1335 SSEPackedDouble>, OpSize, VEX_4V;
1336 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1337 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1338 SSEPackedSingle>, VEX_4V;
1339 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1340 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1341 SSEPackedDouble>, OpSize, VEX_4V;
1343 let Constraints = "$src1 = $dst" in {
1344 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1345 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1346 SSEPackedSingle>, TB;
1347 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1348 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1349 SSEPackedDouble>, TB, OpSize;
1350 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1351 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1352 SSEPackedSingle>, TB;
1353 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1354 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1355 SSEPackedDouble>, TB, OpSize;
1356 } // Constraints = "$src1 = $dst"
1357 } // AddedComplexity
1359 //===----------------------------------------------------------------------===//
1360 // SSE 1 & 2 - Extract Floating-Point Sign mask
1361 //===----------------------------------------------------------------------===//
1363 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1364 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1366 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1367 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1368 [(set GR32:$dst, (Int RC:$src))], d>;
1369 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1370 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1374 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1375 "movmskps", SSEPackedSingle>, VEX;
1376 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1377 "movmskpd", SSEPackedDouble>, OpSize,
1379 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1380 "movmskps", SSEPackedSingle>, VEX;
1381 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1382 "movmskpd", SSEPackedDouble>, OpSize,
1384 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1385 SSEPackedSingle>, TB;
1386 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1387 SSEPackedDouble>, TB, OpSize;
1390 def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
1391 "movmskpd\t{$src, $dst|$dst, $src}",
1392 [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1393 def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1394 "movmskpd\t{$src, $dst|$dst, $src}",
1395 [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1396 def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
1397 "movmskps\t{$src, $dst|$dst, $src}",
1398 [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1399 def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1400 "movmskps\t{$src, $dst|$dst, $src}",
1401 [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1404 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1405 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1406 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1407 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1409 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1410 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1411 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1412 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1415 //===----------------------------------------------------------------------===//
1416 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1417 //===----------------------------------------------------------------------===//
1419 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1420 // names that start with 'Fs'.
1422 // Alias instructions that map fld0 to pxor for sse.
1423 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1424 canFoldAsLoad = 1 in {
1425 // FIXME: Set encoding to pseudo!
1426 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1427 [(set FR32:$dst, fp32imm0)]>,
1428 Requires<[HasSSE1]>, TB, OpSize;
1429 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1430 [(set FR64:$dst, fpimm0)]>,
1431 Requires<[HasSSE2]>, TB, OpSize;
1432 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1433 [(set FR32:$dst, fp32imm0)]>,
1434 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1435 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1436 [(set FR64:$dst, fpimm0)]>,
1437 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1440 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1441 // bits are disregarded.
1442 let neverHasSideEffects = 1 in {
1443 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1444 "movaps\t{$src, $dst|$dst, $src}", []>;
1445 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1446 "movapd\t{$src, $dst|$dst, $src}", []>;
1449 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1450 // bits are disregarded.
1451 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1452 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1453 "movaps\t{$src, $dst|$dst, $src}",
1454 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1455 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1456 "movapd\t{$src, $dst|$dst, $src}",
1457 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1460 //===----------------------------------------------------------------------===//
1461 // SSE 1 & 2 - Logical Instructions
1462 //===----------------------------------------------------------------------===//
1464 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1466 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1468 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1469 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1471 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1472 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1474 let Constraints = "$src1 = $dst" in {
1475 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1476 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1478 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1479 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1483 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1484 let mayLoad = 0 in {
1485 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1486 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1487 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1490 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1491 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1493 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1495 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1497 let Pattern = []<dag> in {
1498 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1499 !strconcat(OpcodeStr, "ps"), f128mem,
1500 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1501 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1502 (memopv2i64 addr:$src2)))], 0>, VEX_4V;
1504 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1505 !strconcat(OpcodeStr, "pd"), f128mem,
1506 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1507 (bc_v2i64 (v2f64 VR128:$src2))))],
1508 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1509 (memopv2i64 addr:$src2)))], 0>,
1512 let Constraints = "$src1 = $dst" in {
1513 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1514 !strconcat(OpcodeStr, "ps"), f128mem,
1515 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1516 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1517 (memopv2i64 addr:$src2)))]>, TB;
1519 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1520 !strconcat(OpcodeStr, "pd"), f128mem,
1521 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1522 (bc_v2i64 (v2f64 VR128:$src2))))],
1523 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1524 (memopv2i64 addr:$src2)))]>, TB, OpSize;
1528 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1530 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1532 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1533 !strconcat(OpcodeStr, "ps"), f256mem,
1534 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
1535 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1536 (memopv4i64 addr:$src2)))], 0>, VEX_4V;
1538 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1539 !strconcat(OpcodeStr, "pd"), f256mem,
1540 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1541 (bc_v4i64 (v4f64 VR256:$src2))))],
1542 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1543 (memopv4i64 addr:$src2)))], 0>,
1547 // AVX 256-bit packed logical ops forms
1548 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1549 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1550 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1551 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
1553 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1554 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1555 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1556 let isCommutable = 0 in
1557 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
1559 //===----------------------------------------------------------------------===//
1560 // SSE 1 & 2 - Arithmetic Instructions
1561 //===----------------------------------------------------------------------===//
1563 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1566 /// In addition, we also have a special variant of the scalar form here to
1567 /// represent the associated intrinsic operation. This form is unlike the
1568 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1569 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1571 /// These three forms can each be reg+reg or reg+mem.
1574 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1576 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1578 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1579 OpNode, FR32, f32mem, Is2Addr>, XS;
1580 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1581 OpNode, FR64, f64mem, Is2Addr>, XD;
1584 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1586 let mayLoad = 0 in {
1587 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1588 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1589 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1590 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1594 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1596 let mayLoad = 0 in {
1597 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1598 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1599 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1600 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1604 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1606 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1607 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1608 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1609 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1612 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1614 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1615 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1616 SSEPackedSingle, Is2Addr>, TB;
1618 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1619 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1620 SSEPackedDouble, Is2Addr>, TB, OpSize;
1623 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1624 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1625 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1626 SSEPackedSingle, 0>, TB;
1628 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1629 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1630 SSEPackedDouble, 0>, TB, OpSize;
1633 // Binary Arithmetic instructions
1634 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1635 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1636 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1637 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1638 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1639 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1640 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1641 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1643 let isCommutable = 0 in {
1644 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1645 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1646 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1647 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1648 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1649 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1650 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1651 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1652 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1653 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1654 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1655 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1656 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1657 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1658 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1659 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1660 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1661 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1662 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1663 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1666 let Constraints = "$src1 = $dst" in {
1667 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1668 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1669 basic_sse12_fp_binop_s_int<0x58, "add">;
1670 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1671 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1672 basic_sse12_fp_binop_s_int<0x59, "mul">;
1674 let isCommutable = 0 in {
1675 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1676 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1677 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1678 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1679 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1680 basic_sse12_fp_binop_s_int<0x5E, "div">;
1681 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1682 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1683 basic_sse12_fp_binop_s_int<0x5F, "max">,
1684 basic_sse12_fp_binop_p_int<0x5F, "max">;
1685 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1686 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1687 basic_sse12_fp_binop_s_int<0x5D, "min">,
1688 basic_sse12_fp_binop_p_int<0x5D, "min">;
1693 /// In addition, we also have a special variant of the scalar form here to
1694 /// represent the associated intrinsic operation. This form is unlike the
1695 /// plain scalar form, in that it takes an entire vector (instead of a
1696 /// scalar) and leaves the top elements undefined.
1698 /// And, we have a special variant form for a full-vector intrinsic form.
1700 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1701 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1702 SDNode OpNode, Intrinsic F32Int> {
1703 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1704 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1705 [(set FR32:$dst, (OpNode FR32:$src))]>;
1706 // For scalar unary operations, fold a load into the operation
1707 // only in OptForSize mode. It eliminates an instruction, but it also
1708 // eliminates a whole-register clobber (the load), so it introduces a
1709 // partial register update condition.
1710 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1711 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1712 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1713 Requires<[HasSSE1, OptForSize]>;
1714 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1715 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1716 [(set VR128:$dst, (F32Int VR128:$src))]>;
1717 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1718 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1719 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1722 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1723 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1724 SDNode OpNode, Intrinsic F32Int> {
1725 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1726 !strconcat(OpcodeStr,
1727 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1728 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1729 !strconcat(OpcodeStr,
1730 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1731 []>, XS, Requires<[HasAVX, OptForSize]>;
1732 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1733 !strconcat(OpcodeStr,
1734 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1735 [(set VR128:$dst, (F32Int VR128:$src))]>;
1736 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1737 !strconcat(OpcodeStr,
1738 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1739 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1742 /// sse1_fp_unop_p - SSE1 unops in packed form.
1743 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1744 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1745 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1746 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1747 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1748 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1749 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1752 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1753 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1754 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1755 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1756 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1757 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1758 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1759 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1762 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1763 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1764 Intrinsic V4F32Int> {
1765 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1766 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1767 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1768 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1769 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1770 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1773 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1774 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1775 Intrinsic V4F32Int> {
1776 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1777 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1778 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1779 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1780 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1781 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1784 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1785 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1786 SDNode OpNode, Intrinsic F64Int> {
1787 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1788 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1789 [(set FR64:$dst, (OpNode FR64:$src))]>;
1790 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1791 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1792 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1793 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1794 Requires<[HasSSE2, OptForSize]>;
1795 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1796 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1797 [(set VR128:$dst, (F64Int VR128:$src))]>;
1798 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1799 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1800 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1803 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1804 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1805 SDNode OpNode, Intrinsic F64Int> {
1806 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1807 !strconcat(OpcodeStr,
1808 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1809 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1810 (ins FR64:$src1, f64mem:$src2),
1811 !strconcat(OpcodeStr,
1812 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1813 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1814 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1815 [(set VR128:$dst, (F64Int VR128:$src))]>;
1816 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1817 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1818 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1821 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1822 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1824 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1825 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1826 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1827 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1828 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1829 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1832 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1833 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1834 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1835 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1836 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1837 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1838 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1839 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1842 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1843 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1844 Intrinsic V2F64Int> {
1845 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1847 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1848 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1849 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1850 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1853 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1854 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1855 Intrinsic V2F64Int> {
1856 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1857 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1858 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1859 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1860 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1861 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1864 let Predicates = [HasAVX] in {
1866 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1867 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1870 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1871 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1872 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1873 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1874 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1875 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1876 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1877 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1880 // Reciprocal approximations. Note that these typically require refinement
1881 // in order to obtain suitable precision.
1882 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1883 int_x86_sse_rsqrt_ss>, VEX_4V;
1884 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1885 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1886 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1887 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1889 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1891 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1892 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1893 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1894 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1898 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1899 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1900 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1901 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1902 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1903 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1905 // Reciprocal approximations. Note that these typically require refinement
1906 // in order to obtain suitable precision.
1907 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1908 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1909 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1910 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1911 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1912 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1914 // There is no f64 version of the reciprocal approximation instructions.
1916 //===----------------------------------------------------------------------===//
1917 // SSE 1 & 2 - Non-temporal stores
1918 //===----------------------------------------------------------------------===//
1920 let AddedComplexity = 400 in { // Prefer non-temporal versions
1921 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1922 (ins f128mem:$dst, VR128:$src),
1923 "movntps\t{$src, $dst|$dst, $src}",
1924 [(alignednontemporalstore (v4f32 VR128:$src),
1926 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1927 (ins f128mem:$dst, VR128:$src),
1928 "movntpd\t{$src, $dst|$dst, $src}",
1929 [(alignednontemporalstore (v2f64 VR128:$src),
1931 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1932 (ins f128mem:$dst, VR128:$src),
1933 "movntdq\t{$src, $dst|$dst, $src}",
1934 [(alignednontemporalstore (v2f64 VR128:$src),
1937 let ExeDomain = SSEPackedInt in
1938 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1939 (ins f128mem:$dst, VR128:$src),
1940 "movntdq\t{$src, $dst|$dst, $src}",
1941 [(alignednontemporalstore (v4f32 VR128:$src),
1944 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1945 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
1947 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1948 (ins f256mem:$dst, VR256:$src),
1949 "movntps\t{$src, $dst|$dst, $src}",
1950 [(alignednontemporalstore (v8f32 VR256:$src),
1952 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1953 (ins f256mem:$dst, VR256:$src),
1954 "movntpd\t{$src, $dst|$dst, $src}",
1955 [(alignednontemporalstore (v4f64 VR256:$src),
1957 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1958 (ins f256mem:$dst, VR256:$src),
1959 "movntdq\t{$src, $dst|$dst, $src}",
1960 [(alignednontemporalstore (v4f64 VR256:$src),
1962 let ExeDomain = SSEPackedInt in
1963 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1964 (ins f256mem:$dst, VR256:$src),
1965 "movntdq\t{$src, $dst|$dst, $src}",
1966 [(alignednontemporalstore (v8f32 VR256:$src),
1970 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1971 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1972 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
1973 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
1974 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
1975 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
1977 let AddedComplexity = 400 in { // Prefer non-temporal versions
1978 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1979 "movntps\t{$src, $dst|$dst, $src}",
1980 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1981 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1982 "movntpd\t{$src, $dst|$dst, $src}",
1983 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1985 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1986 "movntdq\t{$src, $dst|$dst, $src}",
1987 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1989 let ExeDomain = SSEPackedInt in
1990 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1991 "movntdq\t{$src, $dst|$dst, $src}",
1992 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1994 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1995 (MOVNTDQmr addr:$dst, VR128:$src)>;
1997 // There is no AVX form for instructions below this point
1998 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1999 "movnti{l}\t{$src, $dst|$dst, $src}",
2000 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2001 TB, Requires<[HasSSE2]>;
2002 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2003 "movnti{q}\t{$src, $dst|$dst, $src}",
2004 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2005 TB, Requires<[HasSSE2]>;
2008 //===----------------------------------------------------------------------===//
2009 // SSE 1 & 2 - Misc Instructions (No AVX form)
2010 //===----------------------------------------------------------------------===//
2012 // Prefetch intrinsic.
2013 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2014 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2015 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2016 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2017 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2018 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2019 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2020 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2022 // Load, store, and memory fence
2023 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2024 TB, Requires<[HasSSE1]>;
2025 def : Pat<(X86SFence), (SFENCE)>;
2027 // Alias instructions that map zero vector to pxor / xorp* for sse.
2028 // We set canFoldAsLoad because this can be converted to a constant-pool
2029 // load of an all-zeros value if folding it would be beneficial.
2030 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2031 // JIT implementation, it does not expand the instructions below like
2032 // X86MCInstLower does.
2033 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2034 isCodeGenOnly = 1 in {
2035 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2036 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2037 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2038 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2039 let ExeDomain = SSEPackedInt in
2040 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2041 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2044 // The same as done above but for AVX. The 128-bit versions are the
2045 // same, but re-encoded. The 256-bit does not support PI version, and
2046 // doesn't need it because on sandy bridge the register is set to zero
2047 // at the rename stage without using any execution unit, so SET0PSY
2048 // and SET0PDY can be used for vector int instructions without penalty
2049 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2050 // JIT implementatioan, it does not expand the instructions below like
2051 // X86MCInstLower does.
2052 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2053 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2054 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2055 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2056 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2057 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2058 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2059 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2060 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2061 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2062 let ExeDomain = SSEPackedInt in
2063 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2064 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2067 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2068 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2069 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2071 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2072 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2074 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
2075 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
2076 // represent this instead of always zeroing SRC1. One possible solution is
2077 // to represent the instruction w/ something similar as the "$src1 = $dst"
2078 // constraint but without the tied operands.
2079 def : Pat<(extloadf32 addr:$src),
2080 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)), addr:$src)>,
2081 Requires<[HasAVX, OptForSpeed]>;
2083 //===----------------------------------------------------------------------===//
2084 // SSE 1 & 2 - Load/Store XCSR register
2085 //===----------------------------------------------------------------------===//
2087 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2088 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2089 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2090 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2092 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2093 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2094 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2095 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2097 //===---------------------------------------------------------------------===//
2098 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2099 //===---------------------------------------------------------------------===//
2101 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2103 let neverHasSideEffects = 1 in {
2104 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2105 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2106 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2107 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2109 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2110 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2111 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2112 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2114 let canFoldAsLoad = 1, mayLoad = 1 in {
2115 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2116 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2117 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2118 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2119 let Predicates = [HasAVX] in {
2120 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2121 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2122 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2123 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2127 let mayStore = 1 in {
2128 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2129 (ins i128mem:$dst, VR128:$src),
2130 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2131 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2132 (ins i256mem:$dst, VR256:$src),
2133 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2134 let Predicates = [HasAVX] in {
2135 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2136 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2137 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2138 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2142 let neverHasSideEffects = 1 in
2143 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2144 "movdqa\t{$src, $dst|$dst, $src}", []>;
2146 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2147 "movdqu\t{$src, $dst|$dst, $src}",
2148 []>, XS, Requires<[HasSSE2]>;
2150 let canFoldAsLoad = 1, mayLoad = 1 in {
2151 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2152 "movdqa\t{$src, $dst|$dst, $src}",
2153 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2154 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2155 "movdqu\t{$src, $dst|$dst, $src}",
2156 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2157 XS, Requires<[HasSSE2]>;
2160 let mayStore = 1 in {
2161 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2162 "movdqa\t{$src, $dst|$dst, $src}",
2163 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2164 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2165 "movdqu\t{$src, $dst|$dst, $src}",
2166 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2167 XS, Requires<[HasSSE2]>;
2170 // Intrinsic forms of MOVDQU load and store
2171 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2172 "vmovdqu\t{$src, $dst|$dst, $src}",
2173 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2174 XS, VEX, Requires<[HasAVX]>;
2176 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2177 "movdqu\t{$src, $dst|$dst, $src}",
2178 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2179 XS, Requires<[HasSSE2]>;
2181 } // ExeDomain = SSEPackedInt
2183 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2184 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2185 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2187 //===---------------------------------------------------------------------===//
2188 // SSE2 - Packed Integer Arithmetic Instructions
2189 //===---------------------------------------------------------------------===//
2191 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2193 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2194 bit IsCommutable = 0, bit Is2Addr = 1> {
2195 let isCommutable = IsCommutable in
2196 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2197 (ins VR128:$src1, VR128:$src2),
2199 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2200 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2201 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2202 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2203 (ins VR128:$src1, i128mem:$src2),
2205 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2206 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2207 [(set VR128:$dst, (IntId VR128:$src1,
2208 (bitconvert (memopv2i64 addr:$src2))))]>;
2211 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2212 string OpcodeStr, Intrinsic IntId,
2213 Intrinsic IntId2, bit Is2Addr = 1> {
2214 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2215 (ins VR128:$src1, VR128:$src2),
2217 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2219 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2220 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2221 (ins VR128:$src1, i128mem:$src2),
2223 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2224 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2225 [(set VR128:$dst, (IntId VR128:$src1,
2226 (bitconvert (memopv2i64 addr:$src2))))]>;
2227 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2228 (ins VR128:$src1, i32i8imm:$src2),
2230 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2231 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2232 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2235 /// PDI_binop_rm - Simple SSE2 binary operator.
2236 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2237 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2238 let isCommutable = IsCommutable in
2239 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2240 (ins VR128:$src1, VR128:$src2),
2242 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2243 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2244 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2245 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2246 (ins VR128:$src1, i128mem:$src2),
2248 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2249 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2250 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2251 (bitconvert (memopv2i64 addr:$src2)))))]>;
2254 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2256 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2257 /// to collapse (bitconvert VT to VT) into its operand.
2259 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2260 bit IsCommutable = 0, bit Is2Addr = 1> {
2261 let isCommutable = IsCommutable in
2262 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2263 (ins VR128:$src1, VR128:$src2),
2265 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2266 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2267 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2268 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2269 (ins VR128:$src1, i128mem:$src2),
2271 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2272 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2273 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2276 } // ExeDomain = SSEPackedInt
2278 // 128-bit Integer Arithmetic
2280 let Predicates = [HasAVX] in {
2281 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2282 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2283 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2284 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2285 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2286 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2287 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2288 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2289 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2292 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2294 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2296 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2298 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2300 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2302 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2304 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2306 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2308 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2310 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2312 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2314 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2316 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2318 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2320 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2322 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2324 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2326 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2328 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2332 let Constraints = "$src1 = $dst" in {
2333 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2334 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2335 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2336 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2337 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2338 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2339 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2340 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2341 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2344 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2345 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2346 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2347 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2348 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2349 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2350 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2351 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2352 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2353 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2354 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2355 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2356 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2357 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2358 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2359 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2360 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2361 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2362 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2364 } // Constraints = "$src1 = $dst"
2366 //===---------------------------------------------------------------------===//
2367 // SSE2 - Packed Integer Logical Instructions
2368 //===---------------------------------------------------------------------===//
2370 let Predicates = [HasAVX] in {
2371 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2372 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2374 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2375 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2377 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2378 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2381 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2382 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2384 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2385 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2387 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2388 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2391 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2392 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2394 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2395 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2398 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2399 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2400 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2402 let ExeDomain = SSEPackedInt in {
2403 let neverHasSideEffects = 1 in {
2404 // 128-bit logical shifts.
2405 def VPSLLDQri : PDIi8<0x73, MRM7r,
2406 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2407 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2409 def VPSRLDQri : PDIi8<0x73, MRM3r,
2410 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2411 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2413 // PSRADQri doesn't exist in SSE[1-3].
2415 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2416 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2417 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2418 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2419 VR128:$src2)))]>, VEX_4V;
2421 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2422 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2423 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2424 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2425 (memopv2i64 addr:$src2))))]>,
2430 let Constraints = "$src1 = $dst" in {
2431 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2432 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2433 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2434 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2435 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2436 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2438 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2439 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2440 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2441 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2442 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2443 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2445 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2446 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2447 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2448 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2450 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2451 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2452 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2454 let ExeDomain = SSEPackedInt in {
2455 let neverHasSideEffects = 1 in {
2456 // 128-bit logical shifts.
2457 def PSLLDQri : PDIi8<0x73, MRM7r,
2458 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2459 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2460 def PSRLDQri : PDIi8<0x73, MRM3r,
2461 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2462 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2463 // PSRADQri doesn't exist in SSE[1-3].
2465 def PANDNrr : PDI<0xDF, MRMSrcReg,
2466 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2467 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2469 def PANDNrm : PDI<0xDF, MRMSrcMem,
2470 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2471 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2473 } // Constraints = "$src1 = $dst"
2475 let Predicates = [HasAVX] in {
2476 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2477 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2478 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2479 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2480 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2481 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2482 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2483 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2484 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2485 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2487 // Shift up / down and insert zero's.
2488 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2489 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2490 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2491 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2494 let Predicates = [HasSSE2] in {
2495 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2496 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2497 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2498 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2499 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2500 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2501 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2502 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2503 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2504 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2506 // Shift up / down and insert zero's.
2507 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2508 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2509 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2510 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2513 //===---------------------------------------------------------------------===//
2514 // SSE2 - Packed Integer Comparison Instructions
2515 //===---------------------------------------------------------------------===//
2517 let Predicates = [HasAVX] in {
2518 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2520 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2522 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2524 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2526 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2528 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2532 let Constraints = "$src1 = $dst" in {
2533 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2534 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2535 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2536 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2537 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2538 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2539 } // Constraints = "$src1 = $dst"
2541 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2542 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2543 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2544 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2545 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2546 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2547 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2548 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2549 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2550 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2551 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2552 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2554 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2555 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2556 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2557 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2558 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2559 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2560 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2561 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2562 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2563 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2564 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2565 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2567 //===---------------------------------------------------------------------===//
2568 // SSE2 - Packed Integer Pack Instructions
2569 //===---------------------------------------------------------------------===//
2571 let Predicates = [HasAVX] in {
2572 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2574 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2576 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2580 let Constraints = "$src1 = $dst" in {
2581 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2582 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2583 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2584 } // Constraints = "$src1 = $dst"
2586 //===---------------------------------------------------------------------===//
2587 // SSE2 - Packed Integer Shuffle Instructions
2588 //===---------------------------------------------------------------------===//
2590 let ExeDomain = SSEPackedInt in {
2591 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2593 def ri : Ii8<0x70, MRMSrcReg,
2594 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2595 !strconcat(OpcodeStr,
2596 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2597 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2599 def mi : Ii8<0x70, MRMSrcMem,
2600 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2601 !strconcat(OpcodeStr,
2602 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2603 [(set VR128:$dst, (vt (pshuf_frag:$src2
2604 (bc_frag (memopv2i64 addr:$src1)),
2607 } // ExeDomain = SSEPackedInt
2609 let Predicates = [HasAVX] in {
2610 let AddedComplexity = 5 in
2611 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2614 // SSE2 with ImmT == Imm8 and XS prefix.
2615 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2618 // SSE2 with ImmT == Imm8 and XD prefix.
2619 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2623 let Predicates = [HasSSE2] in {
2624 let AddedComplexity = 5 in
2625 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2627 // SSE2 with ImmT == Imm8 and XS prefix.
2628 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2630 // SSE2 with ImmT == Imm8 and XD prefix.
2631 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2634 //===---------------------------------------------------------------------===//
2635 // SSE2 - Packed Integer Unpack Instructions
2636 //===---------------------------------------------------------------------===//
2638 let ExeDomain = SSEPackedInt in {
2639 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2640 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2641 def rr : PDI<opc, MRMSrcReg,
2642 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2644 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2645 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2646 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2647 def rm : PDI<opc, MRMSrcMem,
2648 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2650 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2651 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2652 [(set VR128:$dst, (unp_frag VR128:$src1,
2653 (bc_frag (memopv2i64
2657 let Predicates = [HasAVX] in {
2658 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2660 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2662 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2665 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2666 /// knew to collapse (bitconvert VT to VT) into its operand.
2667 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2668 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2669 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2671 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2672 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2673 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2674 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2676 (v2i64 (unpckl VR128:$src1,
2677 (memopv2i64 addr:$src2))))]>, VEX_4V;
2679 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2681 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2683 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2686 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2687 /// knew to collapse (bitconvert VT to VT) into its operand.
2688 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2689 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2690 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2692 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2693 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2694 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2695 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2697 (v2i64 (unpckh VR128:$src1,
2698 (memopv2i64 addr:$src2))))]>, VEX_4V;
2701 let Constraints = "$src1 = $dst" in {
2702 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2703 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2704 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2706 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2707 /// knew to collapse (bitconvert VT to VT) into its operand.
2708 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2709 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2710 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2712 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2713 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2714 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2715 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2717 (v2i64 (unpckl VR128:$src1,
2718 (memopv2i64 addr:$src2))))]>;
2720 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2721 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2722 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2724 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2725 /// knew to collapse (bitconvert VT to VT) into its operand.
2726 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2727 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2728 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2730 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2731 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2732 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2733 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2735 (v2i64 (unpckh VR128:$src1,
2736 (memopv2i64 addr:$src2))))]>;
2739 } // ExeDomain = SSEPackedInt
2741 //===---------------------------------------------------------------------===//
2742 // SSE2 - Packed Integer Extract and Insert
2743 //===---------------------------------------------------------------------===//
2745 let ExeDomain = SSEPackedInt in {
2746 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2747 def rri : Ii8<0xC4, MRMSrcReg,
2748 (outs VR128:$dst), (ins VR128:$src1,
2749 GR32:$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, GR32:$src2, imm:$src3))]>;
2755 def rmi : Ii8<0xC4, MRMSrcMem,
2756 (outs VR128:$dst), (ins VR128:$src1,
2757 i16mem:$src2, i32i8imm:$src3),
2759 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2760 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2762 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2767 let Predicates = [HasAVX] in
2768 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2769 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2770 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2771 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2772 imm:$src2))]>, OpSize, VEX;
2773 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2774 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2775 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2776 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2780 let Predicates = [HasAVX] in {
2781 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2782 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2783 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2784 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2785 []>, OpSize, VEX_4V;
2788 let Constraints = "$src1 = $dst" in
2789 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2791 } // ExeDomain = SSEPackedInt
2793 //===---------------------------------------------------------------------===//
2794 // SSE2 - Packed Mask Creation
2795 //===---------------------------------------------------------------------===//
2797 let ExeDomain = SSEPackedInt in {
2799 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2800 "pmovmskb\t{$src, $dst|$dst, $src}",
2801 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2802 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2803 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2804 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2805 "pmovmskb\t{$src, $dst|$dst, $src}",
2806 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2808 } // ExeDomain = SSEPackedInt
2810 //===---------------------------------------------------------------------===//
2811 // SSE2 - Conditional Store
2812 //===---------------------------------------------------------------------===//
2814 let ExeDomain = SSEPackedInt in {
2817 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2818 (ins VR128:$src, VR128:$mask),
2819 "maskmovdqu\t{$mask, $src|$src, $mask}",
2820 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2822 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2823 (ins VR128:$src, VR128:$mask),
2824 "maskmovdqu\t{$mask, $src|$src, $mask}",
2825 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2828 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2829 "maskmovdqu\t{$mask, $src|$src, $mask}",
2830 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2832 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2833 "maskmovdqu\t{$mask, $src|$src, $mask}",
2834 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2836 } // ExeDomain = SSEPackedInt
2838 //===---------------------------------------------------------------------===//
2839 // SSE2 - Move Doubleword
2840 //===---------------------------------------------------------------------===//
2842 // Move Int Doubleword to Packed Double Int
2843 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2844 "movd\t{$src, $dst|$dst, $src}",
2846 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2847 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2848 "movd\t{$src, $dst|$dst, $src}",
2850 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2852 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2853 "movd\t{$src, $dst|$dst, $src}",
2855 (v4i32 (scalar_to_vector GR32:$src)))]>;
2856 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2857 "movd\t{$src, $dst|$dst, $src}",
2859 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2860 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2861 "mov{d|q}\t{$src, $dst|$dst, $src}",
2863 (v2i64 (scalar_to_vector GR64:$src)))]>;
2864 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2865 "mov{d|q}\t{$src, $dst|$dst, $src}",
2866 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2869 // Move Int Doubleword to Single Scalar
2870 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2871 "movd\t{$src, $dst|$dst, $src}",
2872 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2874 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2875 "movd\t{$src, $dst|$dst, $src}",
2876 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2878 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2879 "movd\t{$src, $dst|$dst, $src}",
2880 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2882 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2883 "movd\t{$src, $dst|$dst, $src}",
2884 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2886 // Move Packed Doubleword Int to Packed Double Int
2887 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2888 "movd\t{$src, $dst|$dst, $src}",
2889 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2891 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2892 (ins i32mem:$dst, VR128:$src),
2893 "movd\t{$src, $dst|$dst, $src}",
2894 [(store (i32 (vector_extract (v4i32 VR128:$src),
2895 (iPTR 0))), addr:$dst)]>, VEX;
2896 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2897 "movd\t{$src, $dst|$dst, $src}",
2898 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2900 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2901 "movd\t{$src, $dst|$dst, $src}",
2902 [(store (i32 (vector_extract (v4i32 VR128:$src),
2903 (iPTR 0))), addr:$dst)]>;
2905 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2906 "mov{d|q}\t{$src, $dst|$dst, $src}",
2907 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2909 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2910 "movq\t{$src, $dst|$dst, $src}",
2911 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2913 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2914 "mov{d|q}\t{$src, $dst|$dst, $src}",
2915 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2916 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2917 "movq\t{$src, $dst|$dst, $src}",
2918 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2920 // Move Scalar Single to Double Int
2921 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2922 "movd\t{$src, $dst|$dst, $src}",
2923 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2924 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2925 "movd\t{$src, $dst|$dst, $src}",
2926 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2927 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2928 "movd\t{$src, $dst|$dst, $src}",
2929 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2930 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2931 "movd\t{$src, $dst|$dst, $src}",
2932 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2934 // movd / movq to XMM register zero-extends
2935 let AddedComplexity = 15 in {
2936 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2937 "movd\t{$src, $dst|$dst, $src}",
2938 [(set VR128:$dst, (v4i32 (X86vzmovl
2939 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2941 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2942 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2943 [(set VR128:$dst, (v2i64 (X86vzmovl
2944 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2947 let AddedComplexity = 15 in {
2948 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2949 "movd\t{$src, $dst|$dst, $src}",
2950 [(set VR128:$dst, (v4i32 (X86vzmovl
2951 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2952 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2953 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2954 [(set VR128:$dst, (v2i64 (X86vzmovl
2955 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2958 let AddedComplexity = 20 in {
2959 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2960 "movd\t{$src, $dst|$dst, $src}",
2962 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2963 (loadi32 addr:$src))))))]>,
2965 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2966 "movd\t{$src, $dst|$dst, $src}",
2968 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2969 (loadi32 addr:$src))))))]>;
2971 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2972 (MOVZDI2PDIrm addr:$src)>;
2973 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2974 (MOVZDI2PDIrm addr:$src)>;
2975 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2976 (MOVZDI2PDIrm addr:$src)>;
2979 // These are the correct encodings of the instructions so that we know how to
2980 // read correct assembly, even though we continue to emit the wrong ones for
2981 // compatibility with Darwin's buggy assembler.
2982 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2983 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
2984 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2985 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
2986 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2987 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
2988 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2989 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
2990 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2991 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2992 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2993 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2995 //===---------------------------------------------------------------------===//
2996 // SSE2 - Move Quadword
2997 //===---------------------------------------------------------------------===//
2999 // Move Quadword Int to Packed Quadword Int
3000 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3001 "vmovq\t{$src, $dst|$dst, $src}",
3003 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3004 VEX, Requires<[HasAVX]>;
3005 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3006 "movq\t{$src, $dst|$dst, $src}",
3008 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3009 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3011 // Move Packed Quadword Int to Quadword Int
3012 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3013 "movq\t{$src, $dst|$dst, $src}",
3014 [(store (i64 (vector_extract (v2i64 VR128:$src),
3015 (iPTR 0))), addr:$dst)]>, VEX;
3016 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3017 "movq\t{$src, $dst|$dst, $src}",
3018 [(store (i64 (vector_extract (v2i64 VR128:$src),
3019 (iPTR 0))), addr:$dst)]>;
3021 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3022 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3024 // Store / copy lower 64-bits of a XMM register.
3025 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3026 "movq\t{$src, $dst|$dst, $src}",
3027 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3028 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3029 "movq\t{$src, $dst|$dst, $src}",
3030 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3032 let AddedComplexity = 20 in
3033 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3034 "vmovq\t{$src, $dst|$dst, $src}",
3036 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3037 (loadi64 addr:$src))))))]>,
3038 XS, VEX, Requires<[HasAVX]>;
3040 let AddedComplexity = 20 in {
3041 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3042 "movq\t{$src, $dst|$dst, $src}",
3044 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3045 (loadi64 addr:$src))))))]>,
3046 XS, Requires<[HasSSE2]>;
3048 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3049 (MOVZQI2PQIrm addr:$src)>;
3050 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3051 (MOVZQI2PQIrm addr:$src)>;
3052 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3055 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3056 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3057 let AddedComplexity = 15 in
3058 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3059 "vmovq\t{$src, $dst|$dst, $src}",
3060 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3061 XS, VEX, Requires<[HasAVX]>;
3062 let AddedComplexity = 15 in
3063 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3064 "movq\t{$src, $dst|$dst, $src}",
3065 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3066 XS, Requires<[HasSSE2]>;
3068 let AddedComplexity = 20 in
3069 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3070 "vmovq\t{$src, $dst|$dst, $src}",
3071 [(set VR128:$dst, (v2i64 (X86vzmovl
3072 (loadv2i64 addr:$src))))]>,
3073 XS, VEX, Requires<[HasAVX]>;
3074 let AddedComplexity = 20 in {
3075 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3076 "movq\t{$src, $dst|$dst, $src}",
3077 [(set VR128:$dst, (v2i64 (X86vzmovl
3078 (loadv2i64 addr:$src))))]>,
3079 XS, Requires<[HasSSE2]>;
3081 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3082 (MOVZPQILo2PQIrm addr:$src)>;
3085 // Instructions to match in the assembler
3086 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3087 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3088 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3089 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3090 // Recognize "movd" with GR64 destination, but encode as a "movq"
3091 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3092 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3094 // Instructions for the disassembler
3095 // xr = XMM register
3098 let Predicates = [HasAVX] in
3099 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3100 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3101 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3102 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3104 //===---------------------------------------------------------------------===//
3105 // SSE2 - Misc Instructions
3106 //===---------------------------------------------------------------------===//
3109 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3110 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3111 TB, Requires<[HasSSE2]>;
3113 // Load, store, and memory fence
3114 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3115 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3116 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3117 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3118 def : Pat<(X86LFence), (LFENCE)>;
3119 def : Pat<(X86MFence), (MFENCE)>;
3122 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3123 // was introduced with SSE2, it's backward compatible.
3124 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3126 // Alias instructions that map zero vector to pxor / xorp* for sse.
3127 // We set canFoldAsLoad because this can be converted to a constant-pool
3128 // load of an all-ones value if folding it would be beneficial.
3129 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3130 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3131 // FIXME: Change encoding to pseudo.
3132 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3133 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3135 //===---------------------------------------------------------------------===//
3136 // SSE3 - Conversion Instructions
3137 //===---------------------------------------------------------------------===//
3139 // Convert Packed Double FP to Packed DW Integers
3140 let Predicates = [HasAVX] in {
3141 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3142 // register, but the same isn't true when using memory operands instead.
3143 // Provide other assembly rr and rm forms to address this explicitly.
3144 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3145 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3146 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3147 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3150 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3151 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3152 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3153 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3156 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3157 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3158 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3159 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3162 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3163 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3164 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3165 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3167 // Convert Packed DW Integers to Packed Double FP
3168 let Predicates = [HasAVX] in {
3169 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3170 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3171 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3172 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3173 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3174 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3175 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3176 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3179 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3180 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3181 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3182 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3184 // AVX 256-bit register conversion intrinsics
3185 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3186 (VCVTDQ2PDYrr VR128:$src)>;
3187 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3188 (VCVTDQ2PDYrm addr:$src)>;
3190 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3191 (VCVTPD2DQYrr VR256:$src)>;
3192 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3193 (VCVTPD2DQYrm addr:$src)>;
3195 //===---------------------------------------------------------------------===//
3196 // SSE3 - Move Instructions
3197 //===---------------------------------------------------------------------===//
3199 // Replicate Single FP
3200 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3201 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3202 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3203 [(set VR128:$dst, (v4f32 (rep_frag
3204 VR128:$src, (undef))))]>;
3205 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3206 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3207 [(set VR128:$dst, (rep_frag
3208 (memopv4f32 addr:$src), (undef)))]>;
3211 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3213 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3214 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3215 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3216 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3219 let Predicates = [HasAVX] in {
3220 // FIXME: Merge above classes when we have patterns for the ymm version
3221 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3222 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3223 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3224 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3226 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3227 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3229 // Replicate Double FP
3230 multiclass sse3_replicate_dfp<string OpcodeStr> {
3231 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3232 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3233 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3234 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3235 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3237 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3241 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3242 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3243 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3245 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3246 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3250 let Predicates = [HasAVX] in {
3251 // FIXME: Merge above classes when we have patterns for the ymm version
3252 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3253 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3255 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3257 // Move Unaligned Integer
3258 let Predicates = [HasAVX] in {
3259 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3260 "vlddqu\t{$src, $dst|$dst, $src}",
3261 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3262 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3263 "vlddqu\t{$src, $dst|$dst, $src}",
3264 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3266 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3267 "lddqu\t{$src, $dst|$dst, $src}",
3268 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3270 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3272 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3274 // Several Move patterns
3275 let AddedComplexity = 5 in {
3276 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3277 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3278 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3279 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3280 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3281 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3282 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3283 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3286 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3287 let AddedComplexity = 15 in
3288 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3289 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3290 let AddedComplexity = 20 in
3291 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3292 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3294 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3295 let AddedComplexity = 15 in
3296 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3297 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3298 let AddedComplexity = 20 in
3299 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3300 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3302 //===---------------------------------------------------------------------===//
3303 // SSE3 - Arithmetic
3304 //===---------------------------------------------------------------------===//
3306 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3307 X86MemOperand x86memop, bit Is2Addr = 1> {
3308 def rr : I<0xD0, MRMSrcReg,
3309 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3311 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3312 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3313 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3314 def rm : I<0xD0, MRMSrcMem,
3315 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3317 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3318 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3319 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3322 let Predicates = [HasAVX],
3323 ExeDomain = SSEPackedDouble in {
3324 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3325 f128mem, 0>, TB, XD, VEX_4V;
3326 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3327 f128mem, 0>, TB, OpSize, VEX_4V;
3328 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3329 f256mem, 0>, TB, XD, VEX_4V;
3330 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3331 f256mem, 0>, TB, OpSize, VEX_4V;
3333 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3334 ExeDomain = SSEPackedDouble in {
3335 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3337 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3338 f128mem>, TB, OpSize;
3341 //===---------------------------------------------------------------------===//
3342 // SSE3 Instructions
3343 //===---------------------------------------------------------------------===//
3346 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3347 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3348 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3350 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3351 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3352 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3354 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3356 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3357 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3358 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3360 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3361 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3362 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3364 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3366 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3368 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3370 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3372 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3375 let Predicates = [HasAVX] in {
3376 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3377 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3378 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3379 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3380 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3381 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3382 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3383 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3384 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3385 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3386 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3387 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3388 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3389 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3390 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3391 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3394 let Constraints = "$src1 = $dst" in {
3395 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3396 int_x86_sse3_hadd_ps>;
3397 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3398 int_x86_sse3_hadd_pd>;
3399 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3400 int_x86_sse3_hsub_ps>;
3401 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3402 int_x86_sse3_hsub_pd>;
3405 //===---------------------------------------------------------------------===//
3406 // SSSE3 - Packed Absolute Instructions
3407 //===---------------------------------------------------------------------===//
3410 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3411 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3412 PatFrag mem_frag128, Intrinsic IntId128> {
3413 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3415 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3416 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3419 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3421 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3424 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3427 let Predicates = [HasAVX] in {
3428 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3429 int_x86_ssse3_pabs_b_128>, VEX;
3430 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3431 int_x86_ssse3_pabs_w_128>, VEX;
3432 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3433 int_x86_ssse3_pabs_d_128>, VEX;
3436 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3437 int_x86_ssse3_pabs_b_128>;
3438 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3439 int_x86_ssse3_pabs_w_128>;
3440 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3441 int_x86_ssse3_pabs_d_128>;
3443 //===---------------------------------------------------------------------===//
3444 // SSSE3 - Packed Binary Operator Instructions
3445 //===---------------------------------------------------------------------===//
3447 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3448 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3449 PatFrag mem_frag128, Intrinsic IntId128,
3451 let isCommutable = 1 in
3452 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3453 (ins VR128:$src1, VR128:$src2),
3455 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3456 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3457 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3459 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3460 (ins VR128:$src1, i128mem:$src2),
3462 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3463 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3465 (IntId128 VR128:$src1,
3466 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3469 let Predicates = [HasAVX] in {
3470 let isCommutable = 0 in {
3471 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3472 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3473 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3474 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3475 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3476 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3477 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3478 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3479 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3480 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3481 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3482 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3483 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3484 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3485 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3486 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3487 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3488 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3489 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3490 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3491 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3492 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3494 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3495 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3498 // None of these have i8 immediate fields.
3499 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3500 let isCommutable = 0 in {
3501 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3502 int_x86_ssse3_phadd_w_128>;
3503 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3504 int_x86_ssse3_phadd_d_128>;
3505 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3506 int_x86_ssse3_phadd_sw_128>;
3507 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3508 int_x86_ssse3_phsub_w_128>;
3509 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3510 int_x86_ssse3_phsub_d_128>;
3511 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3512 int_x86_ssse3_phsub_sw_128>;
3513 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3514 int_x86_ssse3_pmadd_ub_sw_128>;
3515 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3516 int_x86_ssse3_pshuf_b_128>;
3517 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3518 int_x86_ssse3_psign_b_128>;
3519 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3520 int_x86_ssse3_psign_w_128>;
3521 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3522 int_x86_ssse3_psign_d_128>;
3524 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3525 int_x86_ssse3_pmul_hr_sw_128>;
3528 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3529 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3530 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3531 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3533 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3534 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3535 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3536 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3537 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3538 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3540 //===---------------------------------------------------------------------===//
3541 // SSSE3 - Packed Align Instruction Patterns
3542 //===---------------------------------------------------------------------===//
3544 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3545 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3546 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3548 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3550 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3552 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3553 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3555 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3557 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3561 let Predicates = [HasAVX] in
3562 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3563 let Constraints = "$src1 = $dst" in
3564 defm PALIGN : ssse3_palign<"palignr">;
3566 let AddedComplexity = 5 in {
3567 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3568 (PALIGNR128rr VR128:$src2, VR128:$src1,
3569 (SHUFFLE_get_palign_imm VR128:$src3))>,
3570 Requires<[HasSSSE3]>;
3571 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3572 (PALIGNR128rr VR128:$src2, VR128:$src1,
3573 (SHUFFLE_get_palign_imm VR128:$src3))>,
3574 Requires<[HasSSSE3]>;
3575 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3576 (PALIGNR128rr VR128:$src2, VR128:$src1,
3577 (SHUFFLE_get_palign_imm VR128:$src3))>,
3578 Requires<[HasSSSE3]>;
3579 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3580 (PALIGNR128rr VR128:$src2, VR128:$src1,
3581 (SHUFFLE_get_palign_imm VR128:$src3))>,
3582 Requires<[HasSSSE3]>;
3585 //===---------------------------------------------------------------------===//
3586 // SSSE3 Misc Instructions
3587 //===---------------------------------------------------------------------===//
3589 // Thread synchronization
3590 let usesCustomInserter = 1 in {
3591 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3592 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3593 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3594 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3597 let Uses = [EAX, ECX, EDX] in
3598 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3599 Requires<[HasSSE3]>;
3600 let Uses = [ECX, EAX] in
3601 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3602 Requires<[HasSSE3]>;
3604 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3605 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3607 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3608 Requires<[In32BitMode]>;
3609 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3610 Requires<[In64BitMode]>;
3612 //===---------------------------------------------------------------------===//
3613 // Non-Instruction Patterns
3614 //===---------------------------------------------------------------------===//
3616 // extload f32 -> f64. This matches load+fextend because we have a hack in
3617 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3619 // Since these loads aren't folded into the fextend, we have to match it
3621 let Predicates = [HasSSE2] in
3622 def : Pat<(fextend (loadf32 addr:$src)),
3623 (CVTSS2SDrm addr:$src)>;
3625 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
3626 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
3627 // represent this instead of always zeroing SRC1. One possible solution is
3628 // to represent the instruction w/ something similar as the "$src1 = $dst"
3629 // constraint but without the tied operands.
3630 let Predicates = [HasAVX] in
3631 def : Pat<(fextend (loadf32 addr:$src)),
3632 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)),
3636 let Predicates = [HasXMMInt] in {
3637 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3638 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3639 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3640 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3641 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3642 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3643 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3644 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3645 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3646 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3647 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3648 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3649 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3650 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3651 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3652 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3653 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3654 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3655 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3656 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3657 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3658 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3659 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3660 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3661 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3662 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3663 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3664 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3665 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3666 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3669 let Predicates = [HasAVX] in {
3670 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3671 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
3672 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
3673 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
3674 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
3675 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
3676 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
3677 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
3678 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
3679 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
3680 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
3681 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
3682 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
3683 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
3686 // Move scalar to XMM zero-extended
3687 // movd to XMM register zero-extends
3688 let AddedComplexity = 15 in {
3689 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3690 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3691 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3692 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3693 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3694 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3695 (MOVSSrr (v4f32 (V_SET0PS)),
3696 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3697 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3698 (MOVSSrr (v4i32 (V_SET0PI)),
3699 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3702 // Splat v2f64 / v2i64
3703 let AddedComplexity = 10 in {
3704 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3705 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3706 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3707 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3708 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3709 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3710 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3711 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3714 // Special unary SHUFPSrri case.
3715 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3716 (SHUFPSrri VR128:$src1, VR128:$src1,
3717 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3718 let AddedComplexity = 5 in
3719 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3720 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3721 Requires<[HasSSE2]>;
3722 // Special unary SHUFPDrri case.
3723 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3724 (SHUFPDrri VR128:$src1, VR128:$src1,
3725 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3726 Requires<[HasSSE2]>;
3727 // Special unary SHUFPDrri case.
3728 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3729 (SHUFPDrri VR128:$src1, VR128:$src1,
3730 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3731 Requires<[HasSSE2]>;
3732 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3733 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3734 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3735 Requires<[HasSSE2]>;
3737 // Special binary v4i32 shuffle cases with SHUFPS.
3738 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3739 (SHUFPSrri VR128:$src1, VR128:$src2,
3740 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3741 Requires<[HasSSE2]>;
3742 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3743 (SHUFPSrmi VR128:$src1, addr:$src2,
3744 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3745 Requires<[HasSSE2]>;
3746 // Special binary v2i64 shuffle cases using SHUFPDrri.
3747 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3748 (SHUFPDrri VR128:$src1, VR128:$src2,
3749 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3750 Requires<[HasSSE2]>;
3752 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3753 let AddedComplexity = 15 in {
3754 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3755 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3756 Requires<[OptForSpeed, HasSSE2]>;
3757 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3758 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3759 Requires<[OptForSpeed, HasSSE2]>;
3761 let AddedComplexity = 10 in {
3762 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3763 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3764 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3765 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3766 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3767 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3768 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3769 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3772 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3773 let AddedComplexity = 15 in {
3774 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3775 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3776 Requires<[OptForSpeed, HasSSE2]>;
3777 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3778 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3779 Requires<[OptForSpeed, HasSSE2]>;
3781 let AddedComplexity = 10 in {
3782 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3783 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3784 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3785 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3786 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3787 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3788 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3789 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3792 let AddedComplexity = 20 in {
3793 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3794 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3795 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3797 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3798 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3799 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3801 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3802 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3803 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3804 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3805 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3808 let AddedComplexity = 20 in {
3809 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3810 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3811 (MOVLPSrm VR128:$src1, addr:$src2)>;
3812 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3813 (MOVLPDrm VR128:$src1, addr:$src2)>;
3814 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3815 (MOVLPSrm VR128:$src1, addr:$src2)>;
3816 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3817 (MOVLPDrm VR128:$src1, addr:$src2)>;
3820 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3821 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3822 (MOVLPSmr addr:$src1, VR128:$src2)>;
3823 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3824 (MOVLPDmr addr:$src1, VR128:$src2)>;
3825 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3827 (MOVLPSmr addr:$src1, VR128:$src2)>;
3828 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3829 (MOVLPDmr addr:$src1, VR128:$src2)>;
3831 let AddedComplexity = 15 in {
3832 // Setting the lowest element in the vector.
3833 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3834 (MOVSSrr (v4i32 VR128:$src1),
3835 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3836 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3837 (MOVSDrr (v2i64 VR128:$src1),
3838 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3840 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3841 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3842 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3843 Requires<[HasSSE2]>;
3844 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3845 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3846 Requires<[HasSSE2]>;
3849 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3850 // fall back to this for SSE1)
3851 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3852 (SHUFPSrri VR128:$src2, VR128:$src1,
3853 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3855 // Set lowest element and zero upper elements.
3856 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3857 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3859 // vector -> vector casts
3860 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3861 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3862 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3863 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3865 // Use movaps / movups for SSE integer load / store (one byte shorter).
3866 // The instructions selected below are then converted to MOVDQA/MOVDQU
3867 // during the SSE domain pass.
3868 let Predicates = [HasSSE1] in {
3869 def : Pat<(alignedloadv4i32 addr:$src),
3870 (MOVAPSrm addr:$src)>;
3871 def : Pat<(loadv4i32 addr:$src),
3872 (MOVUPSrm addr:$src)>;
3873 def : Pat<(alignedloadv2i64 addr:$src),
3874 (MOVAPSrm addr:$src)>;
3875 def : Pat<(loadv2i64 addr:$src),
3876 (MOVUPSrm addr:$src)>;
3878 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3879 (MOVAPSmr addr:$dst, VR128:$src)>;
3880 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3881 (MOVAPSmr addr:$dst, VR128:$src)>;
3882 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3883 (MOVAPSmr addr:$dst, VR128:$src)>;
3884 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3885 (MOVAPSmr addr:$dst, VR128:$src)>;
3886 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3887 (MOVUPSmr addr:$dst, VR128:$src)>;
3888 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3889 (MOVUPSmr addr:$dst, VR128:$src)>;
3890 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3891 (MOVUPSmr addr:$dst, VR128:$src)>;
3892 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3893 (MOVUPSmr addr:$dst, VR128:$src)>;
3896 // Use vmovaps/vmovups for AVX integer load/store.
3897 let Predicates = [HasAVX] in {
3898 // 128-bit load/store
3899 def : Pat<(alignedloadv4i32 addr:$src),
3900 (VMOVAPSrm addr:$src)>;
3901 def : Pat<(loadv4i32 addr:$src),
3902 (VMOVUPSrm addr:$src)>;
3903 def : Pat<(alignedloadv2i64 addr:$src),
3904 (VMOVAPSrm addr:$src)>;
3905 def : Pat<(loadv2i64 addr:$src),
3906 (VMOVUPSrm addr:$src)>;
3908 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3909 (VMOVAPSmr addr:$dst, VR128:$src)>;
3910 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3911 (VMOVAPSmr addr:$dst, VR128:$src)>;
3912 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3913 (VMOVAPSmr addr:$dst, VR128:$src)>;
3914 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3915 (VMOVAPSmr addr:$dst, VR128:$src)>;
3916 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3917 (VMOVUPSmr addr:$dst, VR128:$src)>;
3918 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3919 (VMOVUPSmr addr:$dst, VR128:$src)>;
3920 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3921 (VMOVUPSmr addr:$dst, VR128:$src)>;
3922 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3923 (VMOVUPSmr addr:$dst, VR128:$src)>;
3925 // 256-bit load/store
3926 def : Pat<(alignedloadv4i64 addr:$src),
3927 (VMOVAPSYrm addr:$src)>;
3928 def : Pat<(loadv4i64 addr:$src),
3929 (VMOVUPSYrm addr:$src)>;
3930 def : Pat<(alignedloadv8i32 addr:$src),
3931 (VMOVAPSYrm addr:$src)>;
3932 def : Pat<(loadv8i32 addr:$src),
3933 (VMOVUPSYrm addr:$src)>;
3934 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
3935 (VMOVAPSYmr addr:$dst, VR256:$src)>;
3936 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
3937 (VMOVAPSYmr addr:$dst, VR256:$src)>;
3938 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
3939 (VMOVUPSYmr addr:$dst, VR256:$src)>;
3940 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
3941 (VMOVUPSYmr addr:$dst, VR256:$src)>;
3944 //===----------------------------------------------------------------------===//
3945 // SSE4.1 - Packed Move with Sign/Zero Extend
3946 //===----------------------------------------------------------------------===//
3948 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3949 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3950 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3951 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3953 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3954 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3956 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3960 let Predicates = [HasAVX] in {
3961 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3963 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3965 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3967 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3969 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3971 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3975 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3976 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3977 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3978 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3979 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3980 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3982 // Common patterns involving scalar load.
3983 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3984 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3985 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3986 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3988 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3989 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3990 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3991 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3993 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3994 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3995 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3996 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3998 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3999 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4000 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4001 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4003 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4004 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4005 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4006 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4008 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4009 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4010 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4011 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4014 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4015 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4016 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4017 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4019 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4020 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4022 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4026 let Predicates = [HasAVX] in {
4027 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4029 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4031 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4033 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4037 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4038 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4039 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4040 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4042 // Common patterns involving scalar load
4043 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4044 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4045 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4046 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4048 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4049 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4050 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4051 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4054 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4055 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4056 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4057 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4059 // Expecting a i16 load any extended to i32 value.
4060 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4061 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4062 [(set VR128:$dst, (IntId (bitconvert
4063 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4067 let Predicates = [HasAVX] in {
4068 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4070 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4073 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4074 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4076 // Common patterns involving scalar load
4077 def : Pat<(int_x86_sse41_pmovsxbq
4078 (bitconvert (v4i32 (X86vzmovl
4079 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4080 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4082 def : Pat<(int_x86_sse41_pmovzxbq
4083 (bitconvert (v4i32 (X86vzmovl
4084 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4085 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4087 //===----------------------------------------------------------------------===//
4088 // SSE4.1 - Extract Instructions
4089 //===----------------------------------------------------------------------===//
4091 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4092 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4093 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4094 (ins VR128:$src1, i32i8imm:$src2),
4095 !strconcat(OpcodeStr,
4096 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4097 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4099 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4100 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4101 !strconcat(OpcodeStr,
4102 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4105 // There's an AssertZext in the way of writing the store pattern
4106 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4109 let Predicates = [HasAVX] in {
4110 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4111 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4112 (ins VR128:$src1, i32i8imm:$src2),
4113 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4116 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4119 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4120 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4121 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4122 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4123 !strconcat(OpcodeStr,
4124 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4127 // There's an AssertZext in the way of writing the store pattern
4128 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4131 let Predicates = [HasAVX] in
4132 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4134 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4137 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4138 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4139 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4140 (ins VR128:$src1, i32i8imm:$src2),
4141 !strconcat(OpcodeStr,
4142 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4144 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4145 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4146 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4147 !strconcat(OpcodeStr,
4148 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4149 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4150 addr:$dst)]>, OpSize;
4153 let Predicates = [HasAVX] in
4154 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4156 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4158 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4159 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4160 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4161 (ins VR128:$src1, i32i8imm:$src2),
4162 !strconcat(OpcodeStr,
4163 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4165 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4166 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4167 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4168 !strconcat(OpcodeStr,
4169 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4170 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4171 addr:$dst)]>, OpSize, REX_W;
4174 let Predicates = [HasAVX] in
4175 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4177 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4179 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4181 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4182 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4183 (ins VR128:$src1, i32i8imm:$src2),
4184 !strconcat(OpcodeStr,
4185 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4187 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4189 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4190 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4191 !strconcat(OpcodeStr,
4192 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4193 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4194 addr:$dst)]>, OpSize;
4197 let Predicates = [HasAVX] in {
4198 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4199 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4200 (ins VR128:$src1, i32i8imm:$src2),
4201 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4204 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4206 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4207 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4210 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4211 Requires<[HasSSE41]>;
4213 //===----------------------------------------------------------------------===//
4214 // SSE4.1 - Insert Instructions
4215 //===----------------------------------------------------------------------===//
4217 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4218 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4219 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4221 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4223 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4225 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4226 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4227 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4229 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4231 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4233 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4234 imm:$src3))]>, OpSize;
4237 let Predicates = [HasAVX] in
4238 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4239 let Constraints = "$src1 = $dst" in
4240 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4242 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4243 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4244 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4246 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4248 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4250 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4252 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4253 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4255 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4257 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4259 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4260 imm:$src3)))]>, OpSize;
4263 let Predicates = [HasAVX] in
4264 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4265 let Constraints = "$src1 = $dst" in
4266 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4268 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4269 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4270 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4272 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4274 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4276 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4278 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4279 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4281 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4283 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4285 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4286 imm:$src3)))]>, OpSize;
4289 let Predicates = [HasAVX] in
4290 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4291 let Constraints = "$src1 = $dst" in
4292 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4294 // insertps has a few different modes, there's the first two here below which
4295 // are optimized inserts that won't zero arbitrary elements in the destination
4296 // vector. The next one matches the intrinsic and could zero arbitrary elements
4297 // in the target vector.
4298 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4299 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4300 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4302 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4304 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4306 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4308 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4309 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4311 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4313 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4315 (X86insrtps VR128:$src1,
4316 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4317 imm:$src3))]>, OpSize;
4320 let Constraints = "$src1 = $dst" in
4321 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4322 let Predicates = [HasAVX] in
4323 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4325 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4326 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4328 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4329 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4330 Requires<[HasSSE41]>;
4332 //===----------------------------------------------------------------------===//
4333 // SSE4.1 - Round Instructions
4334 //===----------------------------------------------------------------------===//
4336 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4337 X86MemOperand x86memop, RegisterClass RC,
4338 PatFrag mem_frag32, PatFrag mem_frag64,
4339 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4340 // Intrinsic operation, reg.
4341 // Vector intrinsic operation, reg
4342 def PSr : SS4AIi8<opcps, MRMSrcReg,
4343 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4344 !strconcat(OpcodeStr,
4345 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4346 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4349 // Vector intrinsic operation, mem
4350 def PSm : Ii8<opcps, MRMSrcMem,
4351 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4352 !strconcat(OpcodeStr,
4353 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4355 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4357 Requires<[HasSSE41]>;
4359 // Vector intrinsic operation, reg
4360 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4361 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4362 !strconcat(OpcodeStr,
4363 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4364 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4367 // Vector intrinsic operation, mem
4368 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4369 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4370 !strconcat(OpcodeStr,
4371 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4373 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4377 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4378 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4379 // Intrinsic operation, reg.
4380 // Vector intrinsic operation, reg
4381 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4382 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4383 !strconcat(OpcodeStr,
4384 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4387 // Vector intrinsic operation, mem
4388 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4389 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4390 !strconcat(OpcodeStr,
4391 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4392 []>, TA, OpSize, Requires<[HasSSE41]>;
4394 // Vector intrinsic operation, reg
4395 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4396 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4397 !strconcat(OpcodeStr,
4398 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4401 // Vector intrinsic operation, mem
4402 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4403 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4404 !strconcat(OpcodeStr,
4405 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4409 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4412 Intrinsic F64Int, bit Is2Addr = 1> {
4413 // Intrinsic operation, reg.
4414 def SSr : SS4AIi8<opcss, MRMSrcReg,
4415 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4417 !strconcat(OpcodeStr,
4418 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4419 !strconcat(OpcodeStr,
4420 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4421 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4424 // Intrinsic operation, mem.
4425 def SSm : SS4AIi8<opcss, MRMSrcMem,
4426 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4428 !strconcat(OpcodeStr,
4429 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4430 !strconcat(OpcodeStr,
4431 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4433 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4436 // Intrinsic operation, reg.
4437 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4438 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4440 !strconcat(OpcodeStr,
4441 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4442 !strconcat(OpcodeStr,
4443 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4444 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4447 // Intrinsic operation, mem.
4448 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4449 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4451 !strconcat(OpcodeStr,
4452 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4453 !strconcat(OpcodeStr,
4454 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4456 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4460 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4462 // Intrinsic operation, reg.
4463 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4464 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4465 !strconcat(OpcodeStr,
4466 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4469 // Intrinsic operation, mem.
4470 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4471 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4472 !strconcat(OpcodeStr,
4473 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4476 // Intrinsic operation, reg.
4477 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4478 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4479 !strconcat(OpcodeStr,
4480 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4483 // Intrinsic operation, mem.
4484 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4485 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4486 !strconcat(OpcodeStr,
4487 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4491 // FP round - roundss, roundps, roundsd, roundpd
4492 let Predicates = [HasAVX] in {
4494 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4495 memopv4f32, memopv2f64,
4496 int_x86_sse41_round_ps,
4497 int_x86_sse41_round_pd>, VEX;
4498 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4499 memopv8f32, memopv4f64,
4500 int_x86_avx_round_ps_256,
4501 int_x86_avx_round_pd_256>, VEX;
4502 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4503 int_x86_sse41_round_ss,
4504 int_x86_sse41_round_sd, 0>, VEX_4V;
4506 // Instructions for the assembler
4507 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4509 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4511 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4514 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4515 memopv4f32, memopv2f64,
4516 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4517 let Constraints = "$src1 = $dst" in
4518 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4519 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4521 //===----------------------------------------------------------------------===//
4522 // SSE4.1 - Packed Bit Test
4523 //===----------------------------------------------------------------------===//
4525 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4526 // the intel intrinsic that corresponds to this.
4527 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4528 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4529 "vptest\t{$src2, $src1|$src1, $src2}",
4530 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4532 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4533 "vptest\t{$src2, $src1|$src1, $src2}",
4534 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4537 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4538 "vptest\t{$src2, $src1|$src1, $src2}",
4539 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4541 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4542 "vptest\t{$src2, $src1|$src1, $src2}",
4543 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4547 let Defs = [EFLAGS] in {
4548 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4549 "ptest \t{$src2, $src1|$src1, $src2}",
4550 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4552 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4553 "ptest \t{$src2, $src1|$src1, $src2}",
4554 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4558 // The bit test instructions below are AVX only
4559 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4560 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4561 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4562 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4563 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4564 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4565 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4566 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4570 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4571 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4572 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4573 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4574 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4577 //===----------------------------------------------------------------------===//
4578 // SSE4.1 - Misc Instructions
4579 //===----------------------------------------------------------------------===//
4581 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4582 "popcnt{w}\t{$src, $dst|$dst, $src}",
4583 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4584 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4585 "popcnt{w}\t{$src, $dst|$dst, $src}",
4586 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4588 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4589 "popcnt{l}\t{$src, $dst|$dst, $src}",
4590 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4591 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4592 "popcnt{l}\t{$src, $dst|$dst, $src}",
4593 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4595 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4596 "popcnt{q}\t{$src, $dst|$dst, $src}",
4597 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4598 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4599 "popcnt{q}\t{$src, $dst|$dst, $src}",
4600 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4604 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4605 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4606 Intrinsic IntId128> {
4607 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4609 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4610 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4611 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4616 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4619 let Predicates = [HasAVX] in
4620 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4621 int_x86_sse41_phminposuw>, VEX;
4622 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4623 int_x86_sse41_phminposuw>;
4625 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4626 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4627 Intrinsic IntId128, bit Is2Addr = 1> {
4628 let isCommutable = 1 in
4629 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4630 (ins VR128:$src1, VR128:$src2),
4632 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4634 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4635 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4636 (ins VR128:$src1, i128mem:$src2),
4638 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4639 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4641 (IntId128 VR128:$src1,
4642 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4645 let Predicates = [HasAVX] in {
4646 let isCommutable = 0 in
4647 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4649 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4651 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4653 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4655 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4657 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4659 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4661 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4663 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4665 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4667 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4671 let Constraints = "$src1 = $dst" in {
4672 let isCommutable = 0 in
4673 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4674 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4675 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4676 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4677 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4678 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4679 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4680 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4681 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4682 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4683 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4686 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4687 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4688 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4689 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4691 /// SS48I_binop_rm - Simple SSE41 binary operator.
4692 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4693 ValueType OpVT, bit Is2Addr = 1> {
4694 let isCommutable = 1 in
4695 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4696 (ins VR128:$src1, VR128:$src2),
4698 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4699 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4700 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4702 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4703 (ins VR128:$src1, i128mem:$src2),
4705 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4706 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4707 [(set VR128:$dst, (OpNode VR128:$src1,
4708 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4712 let Predicates = [HasAVX] in
4713 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4714 let Constraints = "$src1 = $dst" in
4715 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4717 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4718 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4719 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4720 X86MemOperand x86memop, bit Is2Addr = 1> {
4721 let isCommutable = 1 in
4722 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4723 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4725 !strconcat(OpcodeStr,
4726 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4727 !strconcat(OpcodeStr,
4728 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4729 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4731 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4732 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4734 !strconcat(OpcodeStr,
4735 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4736 !strconcat(OpcodeStr,
4737 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4740 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4744 let Predicates = [HasAVX] in {
4745 let isCommutable = 0 in {
4746 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4747 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4748 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4749 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4750 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4751 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4752 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4753 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4754 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4755 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4756 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4757 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4759 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4760 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4761 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4762 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4763 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4764 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4767 let Constraints = "$src1 = $dst" in {
4768 let isCommutable = 0 in {
4769 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4770 VR128, memopv16i8, i128mem>;
4771 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4772 VR128, memopv16i8, i128mem>;
4773 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4774 VR128, memopv16i8, i128mem>;
4775 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4776 VR128, memopv16i8, i128mem>;
4778 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4779 VR128, memopv16i8, i128mem>;
4780 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4781 VR128, memopv16i8, i128mem>;
4784 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4785 let Predicates = [HasAVX] in {
4786 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4787 RegisterClass RC, X86MemOperand x86memop,
4788 PatFrag mem_frag, Intrinsic IntId> {
4789 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4790 (ins RC:$src1, RC:$src2, RC:$src3),
4791 !strconcat(OpcodeStr,
4792 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4793 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4794 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4796 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4797 (ins RC:$src1, x86memop:$src2, RC:$src3),
4798 !strconcat(OpcodeStr,
4799 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4801 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4803 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4807 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4808 memopv16i8, int_x86_sse41_blendvpd>;
4809 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4810 memopv16i8, int_x86_sse41_blendvps>;
4811 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4812 memopv16i8, int_x86_sse41_pblendvb>;
4813 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4814 memopv32i8, int_x86_avx_blendv_pd_256>;
4815 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4816 memopv32i8, int_x86_avx_blendv_ps_256>;
4818 /// SS41I_ternary_int - SSE 4.1 ternary operator
4819 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4820 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4821 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4822 (ins VR128:$src1, VR128:$src2),
4823 !strconcat(OpcodeStr,
4824 "\t{$src2, $dst|$dst, $src2}"),
4825 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4828 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4829 (ins VR128:$src1, i128mem:$src2),
4830 !strconcat(OpcodeStr,
4831 "\t{$src2, $dst|$dst, $src2}"),
4834 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4838 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4839 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4840 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4842 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4843 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4845 let Predicates = [HasAVX] in
4846 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4847 "vmovntdqa\t{$src, $dst|$dst, $src}",
4848 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4850 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4851 "movntdqa\t{$src, $dst|$dst, $src}",
4852 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4855 //===----------------------------------------------------------------------===//
4856 // SSE4.2 - Compare Instructions
4857 //===----------------------------------------------------------------------===//
4859 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4860 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4861 Intrinsic IntId128, bit Is2Addr = 1> {
4862 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4863 (ins VR128:$src1, VR128:$src2),
4865 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4866 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4867 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4869 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4870 (ins VR128:$src1, i128mem:$src2),
4872 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4873 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4875 (IntId128 VR128:$src1,
4876 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4879 let Predicates = [HasAVX] in
4880 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4882 let Constraints = "$src1 = $dst" in
4883 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4885 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4886 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4887 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4888 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4890 //===----------------------------------------------------------------------===//
4891 // SSE4.2 - String/text Processing Instructions
4892 //===----------------------------------------------------------------------===//
4894 // Packed Compare Implicit Length Strings, Return Mask
4895 multiclass pseudo_pcmpistrm<string asm> {
4896 def REG : PseudoI<(outs VR128:$dst),
4897 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4898 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4900 def MEM : PseudoI<(outs VR128:$dst),
4901 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4902 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4903 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4906 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4907 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4908 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4911 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4912 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4913 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4914 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4915 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4916 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4917 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4920 let Defs = [XMM0, EFLAGS] in {
4921 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4922 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4923 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4924 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4925 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4926 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4929 // Packed Compare Explicit Length Strings, Return Mask
4930 multiclass pseudo_pcmpestrm<string asm> {
4931 def REG : PseudoI<(outs VR128:$dst),
4932 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4933 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4934 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4935 def MEM : PseudoI<(outs VR128:$dst),
4936 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4937 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4938 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4941 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4942 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4943 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4946 let Predicates = [HasAVX],
4947 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4948 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4949 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4950 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4951 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4952 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4953 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4956 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4957 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4958 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4959 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4960 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4961 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4962 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4965 // Packed Compare Implicit Length Strings, Return Index
4966 let Defs = [ECX, EFLAGS] in {
4967 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4968 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4969 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4970 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4971 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4972 (implicit EFLAGS)]>, OpSize;
4973 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4974 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4975 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4976 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4977 (implicit EFLAGS)]>, OpSize;
4981 let Predicates = [HasAVX] in {
4982 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4984 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4986 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4988 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4990 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4992 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4996 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4997 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4998 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4999 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5000 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5001 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5003 // Packed Compare Explicit Length Strings, Return Index
5004 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5005 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5006 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5007 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5008 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5009 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5010 (implicit EFLAGS)]>, OpSize;
5011 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5012 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5013 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5015 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5016 (implicit EFLAGS)]>, OpSize;
5020 let Predicates = [HasAVX] in {
5021 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5023 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5025 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5027 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5029 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5031 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5035 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5036 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5037 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5038 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5039 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5040 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5042 //===----------------------------------------------------------------------===//
5043 // SSE4.2 - CRC Instructions
5044 //===----------------------------------------------------------------------===//
5046 // No CRC instructions have AVX equivalents
5048 // crc intrinsic instruction
5049 // This set of instructions are only rm, the only difference is the size
5051 let Constraints = "$src1 = $dst" in {
5052 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5053 (ins GR32:$src1, i8mem:$src2),
5054 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5056 (int_x86_sse42_crc32_32_8 GR32:$src1,
5057 (load addr:$src2)))]>;
5058 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5059 (ins GR32:$src1, GR8:$src2),
5060 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5062 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5063 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5064 (ins GR32:$src1, i16mem:$src2),
5065 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5067 (int_x86_sse42_crc32_32_16 GR32:$src1,
5068 (load addr:$src2)))]>,
5070 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5071 (ins GR32:$src1, GR16:$src2),
5072 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5074 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5076 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5077 (ins GR32:$src1, i32mem:$src2),
5078 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5080 (int_x86_sse42_crc32_32_32 GR32:$src1,
5081 (load addr:$src2)))]>;
5082 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5083 (ins GR32:$src1, GR32:$src2),
5084 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5086 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5087 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5088 (ins GR64:$src1, i8mem:$src2),
5089 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5091 (int_x86_sse42_crc32_64_8 GR64:$src1,
5092 (load addr:$src2)))]>,
5094 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5095 (ins GR64:$src1, GR8:$src2),
5096 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5098 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5100 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5101 (ins GR64:$src1, i64mem:$src2),
5102 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5104 (int_x86_sse42_crc32_64_64 GR64:$src1,
5105 (load addr:$src2)))]>,
5107 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5108 (ins GR64:$src1, GR64:$src2),
5109 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5111 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5115 //===----------------------------------------------------------------------===//
5116 // AES-NI Instructions
5117 //===----------------------------------------------------------------------===//
5119 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5120 Intrinsic IntId128, bit Is2Addr = 1> {
5121 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5122 (ins VR128:$src1, VR128:$src2),
5124 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5125 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5126 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5128 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5129 (ins VR128:$src1, i128mem:$src2),
5131 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5132 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5134 (IntId128 VR128:$src1,
5135 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5138 // Perform One Round of an AES Encryption/Decryption Flow
5139 let Predicates = [HasAVX, HasAES] in {
5140 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5141 int_x86_aesni_aesenc, 0>, VEX_4V;
5142 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5143 int_x86_aesni_aesenclast, 0>, VEX_4V;
5144 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5145 int_x86_aesni_aesdec, 0>, VEX_4V;
5146 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5147 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5150 let Constraints = "$src1 = $dst" in {
5151 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5152 int_x86_aesni_aesenc>;
5153 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5154 int_x86_aesni_aesenclast>;
5155 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5156 int_x86_aesni_aesdec>;
5157 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5158 int_x86_aesni_aesdeclast>;
5161 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5162 (AESENCrr VR128:$src1, VR128:$src2)>;
5163 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5164 (AESENCrm VR128:$src1, addr:$src2)>;
5165 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5166 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5167 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5168 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5169 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5170 (AESDECrr VR128:$src1, VR128:$src2)>;
5171 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5172 (AESDECrm VR128:$src1, addr:$src2)>;
5173 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5174 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5175 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5176 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5178 // Perform the AES InvMixColumn Transformation
5179 let Predicates = [HasAVX, HasAES] in {
5180 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5182 "vaesimc\t{$src1, $dst|$dst, $src1}",
5184 (int_x86_aesni_aesimc VR128:$src1))]>,
5186 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5187 (ins i128mem:$src1),
5188 "vaesimc\t{$src1, $dst|$dst, $src1}",
5190 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5193 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5195 "aesimc\t{$src1, $dst|$dst, $src1}",
5197 (int_x86_aesni_aesimc VR128:$src1))]>,
5199 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5200 (ins i128mem:$src1),
5201 "aesimc\t{$src1, $dst|$dst, $src1}",
5203 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5206 // AES Round Key Generation Assist
5207 let Predicates = [HasAVX, HasAES] in {
5208 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5209 (ins VR128:$src1, i8imm:$src2),
5210 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5212 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5214 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5215 (ins i128mem:$src1, i8imm:$src2),
5216 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5218 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5222 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5223 (ins VR128:$src1, i8imm:$src2),
5224 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5226 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5228 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5229 (ins i128mem:$src1, i8imm:$src2),
5230 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5232 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5236 //===----------------------------------------------------------------------===//
5237 // CLMUL Instructions
5238 //===----------------------------------------------------------------------===//
5240 // Carry-less Multiplication instructions
5241 let Constraints = "$src1 = $dst" in {
5242 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5243 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5244 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5247 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5248 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5249 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5253 // AVX carry-less Multiplication instructions
5254 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5255 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5256 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5259 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5260 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5261 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5265 multiclass pclmul_alias<string asm, int immop> {
5266 def : InstAlias<!strconcat("pclmul", asm,
5267 "dq {$src, $dst|$dst, $src}"),
5268 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5270 def : InstAlias<!strconcat("pclmul", asm,
5271 "dq {$src, $dst|$dst, $src}"),
5272 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5274 def : InstAlias<!strconcat("vpclmul", asm,
5275 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5276 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5278 def : InstAlias<!strconcat("vpclmul", asm,
5279 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5280 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5282 defm : pclmul_alias<"hqhq", 0x11>;
5283 defm : pclmul_alias<"hqlq", 0x01>;
5284 defm : pclmul_alias<"lqhq", 0x10>;
5285 defm : pclmul_alias<"lqlq", 0x00>;
5287 //===----------------------------------------------------------------------===//
5289 //===----------------------------------------------------------------------===//
5292 // Load from memory and broadcast to all elements of the destination operand
5293 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5294 X86MemOperand x86memop, Intrinsic Int> :
5295 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5296 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5297 [(set RC:$dst, (Int addr:$src))]>, VEX;
5299 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5300 int_x86_avx_vbroadcastss>;
5301 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5302 int_x86_avx_vbroadcastss_256>;
5303 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5304 int_x86_avx_vbroadcast_sd_256>;
5305 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5306 int_x86_avx_vbroadcastf128_pd_256>;
5308 // Insert packed floating-point values
5309 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5310 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5311 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5313 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5314 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5315 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5318 // Extract packed floating-point values
5319 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5320 (ins VR256:$src1, i8imm:$src2),
5321 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5323 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5324 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5325 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5328 // Conditional SIMD Packed Loads and Stores
5329 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5330 Intrinsic IntLd, Intrinsic IntLd256,
5331 Intrinsic IntSt, Intrinsic IntSt256,
5332 PatFrag pf128, PatFrag pf256> {
5333 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5334 (ins VR128:$src1, f128mem:$src2),
5335 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5336 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5338 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5339 (ins VR256:$src1, f256mem:$src2),
5340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5341 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5343 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5344 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5345 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5346 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5347 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5348 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5350 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5353 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5354 int_x86_avx_maskload_ps,
5355 int_x86_avx_maskload_ps_256,
5356 int_x86_avx_maskstore_ps,
5357 int_x86_avx_maskstore_ps_256,
5358 memopv4f32, memopv8f32>;
5359 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5360 int_x86_avx_maskload_pd,
5361 int_x86_avx_maskload_pd_256,
5362 int_x86_avx_maskstore_pd,
5363 int_x86_avx_maskstore_pd_256,
5364 memopv2f64, memopv4f64>;
5366 // Permute Floating-Point Values
5367 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5368 RegisterClass RC, X86MemOperand x86memop_f,
5369 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5370 Intrinsic IntVar, Intrinsic IntImm> {
5371 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5372 (ins RC:$src1, RC:$src2),
5373 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5374 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5375 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5376 (ins RC:$src1, x86memop_i:$src2),
5377 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5378 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5380 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5381 (ins RC:$src1, i8imm:$src2),
5382 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5383 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5384 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5385 (ins x86memop_f:$src1, i8imm:$src2),
5386 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5387 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5390 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5391 memopv4f32, memopv4i32,
5392 int_x86_avx_vpermilvar_ps,
5393 int_x86_avx_vpermil_ps>;
5394 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5395 memopv8f32, memopv8i32,
5396 int_x86_avx_vpermilvar_ps_256,
5397 int_x86_avx_vpermil_ps_256>;
5398 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5399 memopv2f64, memopv2i64,
5400 int_x86_avx_vpermilvar_pd,
5401 int_x86_avx_vpermil_pd>;
5402 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5403 memopv4f64, memopv4i64,
5404 int_x86_avx_vpermilvar_pd_256,
5405 int_x86_avx_vpermil_pd_256>;
5407 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5408 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5409 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5411 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5412 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5413 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5416 // Zero All YMM registers
5417 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5418 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5420 // Zero Upper bits of YMM registers
5421 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5422 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5424 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5425 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5426 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5427 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5428 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5429 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5431 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5433 (VINSERTF128rr VR256:$src1, VR128:$src2,
5434 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5435 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5437 (VINSERTF128rr VR256:$src1, VR128:$src2,
5438 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5439 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5441 (VINSERTF128rr VR256:$src1, VR128:$src2,
5442 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5443 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5445 (VINSERTF128rr VR256:$src1, VR128:$src2,
5446 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5448 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5449 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5450 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5451 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5452 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5453 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5455 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5456 (v4f32 (VEXTRACTF128rr
5457 (v8f32 VR256:$src1),
5458 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5459 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5460 (v2f64 (VEXTRACTF128rr
5461 (v4f64 VR256:$src1),
5462 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5463 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5464 (v4i32 (VEXTRACTF128rr
5465 (v8i32 VR256:$src1),
5466 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5467 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5468 (v2i64 (VEXTRACTF128rr
5469 (v4i64 VR256:$src1),
5470 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5472 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5473 (VBROADCASTF128 addr:$src)>;
5475 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5476 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5477 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5478 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5479 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5480 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5482 def : Pat<(int_x86_avx_vperm2f128_ps_256
5483 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5484 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5485 def : Pat<(int_x86_avx_vperm2f128_pd_256
5486 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5487 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5488 def : Pat<(int_x86_avx_vperm2f128_si_256
5489 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5490 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5492 //===----------------------------------------------------------------------===//
5493 // SSE Shuffle pattern fragments
5494 //===----------------------------------------------------------------------===//
5496 // This is part of a "work in progress" refactoring. The idea is that all
5497 // vector shuffles are going to be translated into target specific nodes and
5498 // directly matched by the patterns below (which can be changed along the way)
5499 // The AVX version of some but not all of them are described here, and more
5500 // should come in a near future.
5502 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5503 // SSE2 loads, which are always promoted to v2i64. The last one should match
5504 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5505 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5506 // we investigate further.
5507 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5509 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5510 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5512 (PSHUFDmi addr:$src1, imm:$imm)>;
5513 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5515 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5517 // Shuffle with PSHUFD instruction.
5518 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5519 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5520 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5521 (PSHUFDri VR128:$src1, imm:$imm)>;
5523 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5524 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5525 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5526 (PSHUFDri VR128:$src1, imm:$imm)>;
5528 // Shuffle with SHUFPD instruction.
5529 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5530 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5531 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5532 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5533 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5534 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5536 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5537 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5538 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5539 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5541 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5542 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5543 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5544 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5546 // Shuffle with SHUFPS instruction.
5547 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5548 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5549 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5550 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5551 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5552 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5554 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5555 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5556 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5557 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5559 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5560 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5561 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5562 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5563 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5564 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5566 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5567 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5568 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5569 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5571 // Shuffle with MOVHLPS instruction
5572 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5573 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5574 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5575 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5577 // Shuffle with MOVDDUP instruction
5578 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5579 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5580 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5581 (MOVDDUPrm addr:$src)>;
5583 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5584 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5585 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5586 (MOVDDUPrm addr:$src)>;
5588 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5589 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5590 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5591 (MOVDDUPrm addr:$src)>;
5593 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5594 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5595 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5596 (MOVDDUPrm addr:$src)>;
5598 def : Pat<(X86Movddup (bc_v2f64
5599 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5600 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5601 def : Pat<(X86Movddup (bc_v2f64
5602 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5603 (MOVDDUPrm addr:$src)>;
5606 // Shuffle with UNPCKLPS
5607 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5608 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5609 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5610 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5611 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5612 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5614 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5615 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5616 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5617 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5618 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5619 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5621 // Shuffle with UNPCKHPS
5622 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5623 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5624 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5625 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5627 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5628 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5629 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5630 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5632 // Shuffle with UNPCKLPD
5633 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5634 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5635 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5636 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5637 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5638 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5640 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5641 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5642 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5643 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5644 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5645 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5647 // Shuffle with UNPCKHPD
5648 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5649 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5650 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5651 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5653 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5654 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5655 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5656 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5658 // Shuffle with PUNPCKLBW
5659 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5660 (bc_v16i8 (memopv2i64 addr:$src2)))),
5661 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5662 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5663 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5665 // Shuffle with PUNPCKLWD
5666 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5667 (bc_v8i16 (memopv2i64 addr:$src2)))),
5668 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5669 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5670 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5672 // Shuffle with PUNPCKLDQ
5673 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5674 (bc_v4i32 (memopv2i64 addr:$src2)))),
5675 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5676 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5677 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5679 // Shuffle with PUNPCKLQDQ
5680 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5681 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5682 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5683 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5685 // Shuffle with PUNPCKHBW
5686 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5687 (bc_v16i8 (memopv2i64 addr:$src2)))),
5688 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5689 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5690 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5692 // Shuffle with PUNPCKHWD
5693 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5694 (bc_v8i16 (memopv2i64 addr:$src2)))),
5695 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5696 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5697 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5699 // Shuffle with PUNPCKHDQ
5700 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5701 (bc_v4i32 (memopv2i64 addr:$src2)))),
5702 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5703 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5704 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5706 // Shuffle with PUNPCKHQDQ
5707 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5708 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5709 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5710 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5712 // Shuffle with MOVLHPS
5713 def : Pat<(X86Movlhps VR128:$src1,
5714 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5715 (MOVHPSrm VR128:$src1, addr:$src2)>;
5716 def : Pat<(X86Movlhps VR128:$src1,
5717 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5718 (MOVHPSrm VR128:$src1, addr:$src2)>;
5719 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5720 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5721 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5722 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5723 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5724 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5726 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5727 // is during lowering, where it's not possible to recognize the load fold cause
5728 // it has two uses through a bitcast. One use disappears at isel time and the
5729 // fold opportunity reappears.
5730 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5731 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5733 // Shuffle with MOVLHPD
5734 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5735 (scalar_to_vector (loadf64 addr:$src2)))),
5736 (MOVHPDrm VR128:$src1, addr:$src2)>;
5738 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5739 // is during lowering, where it's not possible to recognize the load fold cause
5740 // it has two uses through a bitcast. One use disappears at isel time and the
5741 // fold opportunity reappears.
5742 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5743 (scalar_to_vector (loadf64 addr:$src2)))),
5744 (MOVHPDrm VR128:$src1, addr:$src2)>;
5746 // Shuffle with MOVSS
5747 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5748 (MOVSSrr VR128:$src1, FR32:$src2)>;
5749 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5750 (MOVSSrr (v4i32 VR128:$src1),
5751 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5752 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5753 (MOVSSrr (v4f32 VR128:$src1),
5754 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5755 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5756 // is during lowering, where it's not possible to recognize the load fold cause
5757 // it has two uses through a bitcast. One use disappears at isel time and the
5758 // fold opportunity reappears.
5759 def : Pat<(X86Movss VR128:$src1,
5760 (bc_v4i32 (v2i64 (load addr:$src2)))),
5761 (MOVLPSrm VR128:$src1, addr:$src2)>;
5763 // Shuffle with MOVSD
5764 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5765 (MOVSDrr VR128:$src1, FR64:$src2)>;
5766 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5767 (MOVSDrr (v2i64 VR128:$src1),
5768 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5769 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5770 (MOVSDrr (v2f64 VR128:$src1),
5771 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5772 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5773 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5774 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5775 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5777 // Shuffle with MOVSHDUP
5778 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5779 (MOVSHDUPrr VR128:$src)>;
5780 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5781 (MOVSHDUPrm addr:$src)>;
5783 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5784 (MOVSHDUPrr VR128:$src)>;
5785 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5786 (MOVSHDUPrm addr:$src)>;
5788 // Shuffle with MOVSLDUP
5789 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5790 (MOVSLDUPrr VR128:$src)>;
5791 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5792 (MOVSLDUPrm addr:$src)>;
5794 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5795 (MOVSLDUPrr VR128:$src)>;
5796 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5797 (MOVSLDUPrm addr:$src)>;
5799 // Shuffle with PSHUFHW
5800 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5801 (PSHUFHWri VR128:$src, imm:$imm)>;
5802 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5803 (PSHUFHWmi addr:$src, imm:$imm)>;
5805 // Shuffle with PSHUFLW
5806 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5807 (PSHUFLWri VR128:$src, imm:$imm)>;
5808 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5809 (PSHUFLWmi addr:$src, imm:$imm)>;
5811 // Shuffle with PALIGN
5812 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5813 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5814 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5815 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5816 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5817 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5818 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5819 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5821 // Shuffle with MOVLPS
5822 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5823 (MOVLPSrm VR128:$src1, addr:$src2)>;
5824 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5825 (MOVLPSrm VR128:$src1, addr:$src2)>;
5826 def : Pat<(X86Movlps VR128:$src1,
5827 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5828 (MOVLPSrm VR128:$src1, addr:$src2)>;
5829 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5830 // is during lowering, where it's not possible to recognize the load fold cause
5831 // it has two uses through a bitcast. One use disappears at isel time and the
5832 // fold opportunity reappears.
5833 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5834 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5836 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5837 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5839 // Shuffle with MOVLPD
5840 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5841 (MOVLPDrm VR128:$src1, addr:$src2)>;
5842 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5843 (MOVLPDrm VR128:$src1, addr:$src2)>;
5844 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5845 (scalar_to_vector (loadf64 addr:$src2)))),
5846 (MOVLPDrm VR128:$src1, addr:$src2)>;
5848 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5849 def : Pat<(store (f64 (vector_extract
5850 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5851 (MOVHPSmr addr:$dst, VR128:$src)>;
5852 def : Pat<(store (f64 (vector_extract
5853 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5854 (MOVHPDmr addr:$dst, VR128:$src)>;
5856 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5857 (MOVLPSmr addr:$src1, VR128:$src2)>;
5858 def : Pat<(store (v4i32 (X86Movlps
5859 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5860 (MOVLPSmr addr:$src1, VR128:$src2)>;
5862 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5863 (MOVLPDmr addr:$src1, VR128:$src2)>;
5864 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5865 (MOVLPDmr addr:$src1, VR128:$src2)>;